1
00:00:06,266 --> 00:00:07,901
WELCOME, EVERYONE, TO THE 18th

2
00:00:07,901 --> 00:00:10,069
MEETING OF THE BRAIN NEUROETHICS

3
00:00:10,069 --> 00:00:12,372
WORKING GROUP, NEWG, AS WE

4
00:00:12,372 --> 00:00:16,109
AFFECTIONATELY CALL THIS GROUP.

5
00:00:16,109 --> 00:00:16,976
I'M ANDREA BECKEL-MITCHENER,

6
00:00:16,976 --> 00:00:18,311
DEPUTY DIRECTOR OF THE NIH BRAIN

7
00:00:18,311 --> 00:00:22,916
INITIATIVE, AND I SERVE  AS

8
00:00:22,916 --> 00:00:24,284
DESIGNATED FEDERAL OFFICIAL FOR

9
00:00:24,284 --> 00:00:24,684
TODAY'S MEETING.

10
00:00:24,684 --> 00:00:26,519
I WANT TO EXTEND A WARM WELCOME

11
00:00:26,519 --> 00:00:29,589
TO WORKING GROUP MEMBERS AND

12
00:00:29,589 --> 00:00:31,024
CO-CHAIRS, CHRISTINE GRADY AND

13
00:00:31,024 --> 00:00:32,225
NINA FARAHANY.

14
00:00:32,225 --> 00:00:34,527
I HADN'T TO WELCOME OUR INVITED

15
00:00:34,527 --> 00:00:36,062
SPEAKERS AND PANELISTS AND

16
00:00:36,062 --> 00:00:38,765
GUESTS JOINING US TODAY VIA THE

17
00:00:38,765 --> 00:00:39,666
MAGIC OF VIDEOCAST.

18
00:00:39,666 --> 00:00:42,302
AS MOST OF YOU KNOW, THE NEWG IS

19
00:00:42,302 --> 00:00:44,370
NOT A FORMAL ADVISORY GROUP LIKE

20
00:00:44,370 --> 00:00:45,805
THE MULTI-COUNCIL WORKING GROUP

21
00:00:45,805 --> 00:00:47,173
WHICH WILL MEET TOMORROW.

22
00:00:47,173 --> 00:00:50,176
THE NEWG IS A WORKING GROUP OF

23
00:00:50,176 --> 00:00:51,577
THE TEN ADVISORY COUNCILS OF THE

24
00:00:51,577 --> 00:00:53,780
NIH INSTITUTES AND CENTERS THAT

25
00:00:53,780 --> 00:00:54,881
PARTICIPATE IN THE BRAIN

26
00:00:54,881 --> 00:00:57,283
INITIATIVE, AND IT WAS

27
00:00:57,283 --> 00:00:59,452
ESTABLISHED TO IDENTIFY AND

28
00:00:59,452 --> 00:01:00,553
REFLECT ON ETHICAL

29
00:01:00,553 --> 00:01:01,654
CONSIDERATIONS RELATED TO

30
00:01:01,654 --> 00:01:02,755
RESEARCH SUPPORTED THROUGH THE

31
00:01:02,755 --> 00:01:04,590
BRAIN INITIATIVE, IN THIS ROLE

32
00:01:04,590 --> 00:01:06,592
THE NEWG DISCUSSES AND

33
00:01:06,592 --> 00:01:07,760
DELIBERATES ON ETHICAL, LEGAL,

34
00:01:07,760 --> 00:01:09,329
SOCIETAL QUESTIONS THAT COME UP

35
00:01:09,329 --> 00:01:11,064
IN THE CONTEXT OF BRAIN

36
00:01:11,064 --> 00:01:12,832
INITIATIVE RESEARCH, AND OFFERS

37
00:01:12,832 --> 00:01:14,701
HELPFUL PERSPECTIVES AND POINTS

38
00:01:14,701 --> 00:01:16,202
TO CONSIDER.

39
00:01:16,202 --> 00:01:17,737
AND TODAY'S VIRTUAL MEETING WE

40
00:01:17,737 --> 00:01:20,606
WILL HAVE ITEMS ON THE AGENDA,

41
00:01:20,606 --> 00:01:22,709
YOU ALL SHOULD HAVE RECEIVED THE

42
00:01:22,709 --> 00:01:23,576
AGENDA, IT'S POSTED.

43
00:01:23,576 --> 00:01:25,645
FIRST WE'LL HEAR AN UPDATE FROM

44
00:01:25,645 --> 00:01:27,480
OUR BRAIN DIRECTOR DR. JOHN

45
00:01:27,480 --> 00:01:28,915
NGAI, AND WE'LL DEVOTE THE

46
00:01:28,915 --> 00:01:30,183
MAJORITY OF TODAY'S MEETING AT

47
00:01:30,183 --> 00:01:35,054
THE INTERSECTION OF ARTIFICIAL

48
00:01:35,054 --> 00:01:35,688
INTELLIGENCE, NEUROSCIENCE,

49
00:01:35,688 --> 00:01:37,357
PHENOTYPING, ETHICS, THROUGH THE

50
00:01:37,357 --> 00:01:38,758
PRESENTATION AND EXPLORATION OF

51
00:01:38,758 --> 00:01:40,059
CASE STUDIES TO HELP SET THE

52
00:01:40,059 --> 00:01:44,030
STAGE, WE'LL FIRST HAVE A

53
00:01:44,030 --> 00:01:46,632
TECHNICAL OVERVIEW TO HELP

54
00:01:46,632 --> 00:01:49,602
GROUND POSSESSION ETHICAL

55
00:01:49,602 --> 00:01:54,307
CONSIDERSES IN THE --

56
00:01:54,307 --> 00:01:54,674
CONSIDERATIONS.

57
00:01:54,674 --> 00:01:56,442
WE'LL HEAR UPDATES FROM NEWG

58
00:01:56,442 --> 00:01:57,443
MEMBERS ABOUT EVERYONE HAS BEEN

59
00:01:57,443 --> 00:02:01,414
UP TO IN THE NEUROETHICS SPACE

60
00:02:01,414 --> 00:02:04,017
SINCE WE LAST MET.

61
00:02:04,017 --> 00:02:05,885
A FEW HOUSEKEEPING ITEMS, AS A

62
00:02:05,885 --> 00:02:07,787
REMINDER THIS OPEN SESSION IS

63
00:02:07,787 --> 00:02:10,056
BEING VIDEOCAST LIVE IN THE

64
00:02:10,056 --> 00:02:12,025
PUBLIC DOMAIN, AND THE RECORDING

65
00:02:12,025 --> 00:02:13,126
WILL BE ARCHIVED FOR LATER

66
00:02:13,126 --> 00:02:17,063
VIEWING ON THE NIH AND THE BRAIN

67
00:02:17,063 --> 00:02:17,530
WEBSITE.

68
00:02:17,530 --> 00:02:19,465
PLEASE REMAIN ON MUTE UNLESS

69
00:02:19,465 --> 00:02:20,666
YOU'RE PRESENTING.

70
00:02:20,666 --> 00:02:22,435
DURING DISCUSSIONS WE ASK NEWG

71
00:02:22,435 --> 00:02:24,404
MEMBERS, PLEASE TURN ON YOUR

72
00:02:24,404 --> 00:02:25,738
VIDEO AND USE THE RAISE HAND

73
00:02:25,738 --> 00:02:27,907
OPTION IF YOU HAVE QUESTIONS OR

74
00:02:27,907 --> 00:02:29,108
COMMENTS, MODERATORS WILL CALL

75
00:02:29,108 --> 00:02:30,777
ON YOU TO UNMUTE YOURSELF.

76
00:02:30,777 --> 00:02:32,311
IT'S ESSENTIAL TO USE RAISE HAND

77
00:02:32,311 --> 00:02:33,379
FEATURE BECAUSE RATHER THAN

78
00:02:33,379 --> 00:02:35,615
TYPING IN THE ZOOM CHAT FOR OUR

79
00:02:35,615 --> 00:02:36,883
VIDEOCAST ATTENDEES SO THEY CAN

80
00:02:36,883 --> 00:02:39,318
HEAR AND SEE THE QUESTIONS.

81
00:02:39,318 --> 00:02:41,054
LASTLY, NONE OF THIS CAN HAPPEN

82
00:02:41,054 --> 00:02:43,689
WITHOUT THE INCREDIBLE TEAM WE

83
00:02:43,689 --> 00:02:43,890
HAVE.

84
00:02:43,890 --> 00:02:46,759
I ESPECIALLY WANT TO THANK DR.

85
00:02:46,759 --> 00:02:48,828
NINA HSU, SPECIALIST FOR THIS

86
00:02:48,828 --> 00:02:51,431
WORKING GROUP, THE SEEM MAKING

87
00:02:51,431 --> 00:02:55,401
TODAY'S MEETING POSSIBLE

88
00:02:55,401 --> 00:03:05,878
INCLUDES DOCTOR HENDRICKS, 

89
00:03:05,878 --> 00:03:09,949
DUFRAYA AND THE ROSE LI TEAM AND

90
00:03:09,949 --> 00:03:11,150
VIDEOCAST TEAM.

91
00:03:11,150 --> 00:03:15,555
I PASS TO CHRISTINE AND NINA FOR

92
00:03:15,555 --> 00:03:16,956
OPENING REMARKS.

93
00:03:16,956 --> 00:03:17,824
>> THANK YOU, ANDREA.

94
00:03:17,824 --> 00:03:19,559
I JUST WANT TO SAY HELLO AND

95
00:03:19,559 --> 00:03:19,792
WELCOME.

96
00:03:19,792 --> 00:03:22,295
THANK YOU FOR JOINING US TODAY,

97
00:03:22,295 --> 00:03:24,163
NEWG MEMBERS AND EVERYBODY ELSE.

98
00:03:24,163 --> 00:03:25,898
IT'S A PRIVILEGE FOR ME TO BE

99
00:03:25,898 --> 00:03:30,736
ABLE TO CO-CHAIR THIS GROUP WITH

100
00:03:30,736 --> 00:03:31,471
NINA FARAHANY AND THE GREAT

101
00:03:31,471 --> 00:03:32,672
PEOPLE YOU JUST HEARD ABOUT.

102
00:03:32,672 --> 00:03:37,376
I WANT TO ADD MY THANKS TO NINA

103
00:03:37,376 --> 00:03:41,214
HSU AND  SASKIA HENDRIKS WHO MAE

104
00:03:41,214 --> 00:03:42,715
THIS HAPPEN AND THE OTHER PEOPLE

105
00:03:42,715 --> 00:03:43,349
ANDREA MENTIONED.

106
00:03:43,349 --> 00:03:45,017
WE HAVE AN EXCITING TOPIC AND

107
00:03:45,017 --> 00:03:46,786
GREAT SPEAKERS SO I'M LOOKING

108
00:03:46,786 --> 00:03:48,621
FORWARD VERY MUCH TO THE

109
00:03:48,621 --> 00:03:56,529
DISCUSSION AND SESSIONS THAT WE

110
00:03:56,529 --> 00:03:57,530
HAVE.

111
00:03:57,530 --> 00:03:57,830
NITA?

112
00:03:57,830 --> 00:03:58,598
>> YOU COVERED IT,
INCLUDING

113
00:03:58,598 --> 00:04:01,567
THANKS TO WORK WITH YOU IN THIS

114
00:04:01,567 --> 00:04:01,834
CAPACITY.

115
00:04:01,834 --> 00:04:04,704
I'M EXCITED FOR THE DAY AHEAD.

116
00:04:04,704 --> 00:04:15,148
THERE'S A LOT OF IMPORTANT

117
00:04:16,082 --> 00:04:16,883
MATERIAL TO COVER.

118
00:04:16,883 --> 00:04:18,651
>> ALL RIGHT.

119
00:04:18,651 --> 00:04:19,085
THANKS.

120
00:04:19,085 --> 00:04:22,488
WE'LL JUMP TO JOHN.

121
00:04:22,488 --> 00:04:24,323
>> THANKS, EVERYBODY.

122
00:04:24,323 --> 00:04:26,192
THE WORDS "DITTO" DON'T DO IT SO

123
00:04:26,192 --> 00:04:27,827
I HAVE TO ADD MY THANKS TO

124
00:04:27,827 --> 00:04:30,730
EVERYBODY MENTIONED.

125
00:04:30,730 --> 00:04:38,271
NITA AND CHRISTINE, SASKIA

126
00:04:38,271 --> 00:04:40,673
CHRIS AND DEB AND OTHERS FOR

127
00:04:40,673 --> 00:04:42,341
MAKING SURE EVERYTHING WORKS.

128
00:04:42,341 --> 00:04:44,944
OF COURSE TO THE NEWG MEMBERS

129
00:04:44,944 --> 00:04:46,712
HERE TODAY FOR WHAT I'M SURE

130
00:04:46,712 --> 00:04:51,083
WILL BE A GREAT DISCUSSION WITH

131
00:04:51,083 --> 00:04:52,385
SPECIAL GUESTS.

132
00:04:52,385 --> 00:04:56,322
SINCE I JOINED BRAIN OVER 4 1/2

133
00:04:56,322 --> 00:04:59,058
YEARS AGO WHAT IMPRESSED ME WAS

134
00:04:59,058 --> 00:05:03,229
BEING MINDFUL ABOUT WHAT WE DO

135
00:05:03,229 --> 00:05:04,530
AND IMPACTS ON INDIVIDUALS AND

136
00:05:04,530 --> 00:05:05,064
SOCIETY.

137
00:05:05,064 --> 00:05:08,067
AS WE LOOK AT THIS CURRENT

138
00:05:08,067 --> 00:05:09,602
ITERATION OF THE A.I. REVOLUTION

139
00:05:09,602 --> 00:05:11,571
IT'S BECOME CLEAR TO ME THAT

140
00:05:11,571 --> 00:05:12,872
"BRAIN" CAN AND SHOULD PLAY A

141
00:05:12,872 --> 00:05:16,275
ROLE IN GUIDING OR AT LEAST

142
00:05:16,275 --> 00:05:20,880
PARTICIPATING IN LEADERSHIP WAY

143
00:05:20,880 --> 00:05:22,515
IN THESE IMPORTANT DISCUSSIONS.

144
00:05:22,515 --> 00:05:26,252
LET ME START BY SHARING A FEW

145
00:05:26,252 --> 00:05:27,887
UPDATES, IF I REMEMBER HOW TO DO

146
00:05:27,887 --> 00:05:28,120
THIS.

147
00:05:28,120 --> 00:05:29,655
YOU WOULD THINK I WOULD REMEMBER

148
00:05:29,655 --> 00:05:31,924
HOW TO DO THIS BY NOW.

149
00:05:31,924 --> 00:05:35,861
CAN YOU SEE MY SCREEN, YEAH?

150
00:05:35,861 --> 00:05:36,128
TERRIFIC.

151
00:05:36,128 --> 00:05:37,830
A FEW UPDATES.

152
00:05:37,830 --> 00:05:40,032
WE LIKE TO START WITH RECAPS OF

153
00:05:40,032 --> 00:05:42,101
RECENT EVENTS IN THE NEUROETHICS

154
00:05:42,101 --> 00:05:43,636
DOMAIN AND WHAT OUR FOLKS ARE

155
00:05:43,636 --> 00:05:44,403
DOING.

156
00:05:44,403 --> 00:05:47,807
THERE WAS A MEETING IN APRIL AT

157
00:05:47,807 --> 00:05:49,775
THE INTERNATIONAL NEUROETHICS

158
00:05:49,775 --> 00:05:53,312
SOCIETY, DOCTORS CHURCHILL AND

159
00:05:53,312 --> 00:05:56,616
NINA HSU WHO LEAD THE TEAM

160
00:05:56,616 --> 00:06:00,119
ORGANIZED A WORKSHOP TO TALK

161
00:06:00,119 --> 00:06:02,188
ABOUT DEMYSTIFYING BRAIN

162
00:06:02,188 --> 00:06:02,822
INITIATIVE FUNDING, FUNDING

163
00:06:02,822 --> 00:06:04,257
SCHOLARLY WORK IN THIS DOMAIN,

164
00:06:04,257 --> 00:06:08,194
HAPPY AND PROUD ABOUT THAT.

165
00:06:08,194 --> 00:06:10,997
THE GOAL WAS TO PROVIDE BETTER

166
00:06:10,997 --> 00:06:11,697
UNDERSTANDING OF BRAIN

167
00:06:11,697 --> 00:06:13,132
INITIATIVE IN THIS SPACE AND

168
00:06:13,132 --> 00:06:14,000
GIVE INVESTIGATORS TOOLS TO

169
00:06:14,000 --> 00:06:16,936
REACH OUT TO US TO ENGAGE -- TO

170
00:06:16,936 --> 00:06:19,905
ENGAGE OTHERS IN THIS NETWORK

171
00:06:19,905 --> 00:06:21,007
FOR RESEARCH IN NEUROETHICS, AND

172
00:06:21,007 --> 00:06:22,975
I'LL TOUCH ON THAT BRIEFLY

173
00:06:22,975 --> 00:06:24,744
BECAUSE THIS IS SOMETHING

174
00:06:24,744 --> 00:06:26,479
SPREADING BEYOND JUST "BRAIN"

175
00:06:26,479 --> 00:06:32,752
WHICH MAKES US QUITE HAPPY.

176
00:06:32,752 --> 00:06:35,488
WE HAD A GREAT ANNUAL BRAIN

177
00:06:35,488 --> 00:06:36,355
INITIATIVE CONFERENCE BACK IN

178
00:06:36,355 --> 00:06:36,922
JUNE.

179
00:06:36,922 --> 00:06:38,557
THIS IS OUR 10th ANNUAL

180
00:06:38,557 --> 00:06:39,091
CONFERENCE.

181
00:06:39,091 --> 00:06:44,363
WE HAD A LOT OF GREAT TALKS.

182
00:06:44,363 --> 00:06:46,666
WE HAD A FUN PANEL WITH A BUNCH

183
00:06:46,666 --> 00:06:47,733
OF LUMINARIES TALKING ABOUT NOT

184
00:06:47,733 --> 00:06:50,469
JUST THE PAST BUT THE FUTURE OF

185
00:06:50,469 --> 00:06:50,703
"BRAIN."

186
00:06:50,703 --> 00:06:53,539
AGAIN, AS ALWAYS, WE MADE SURE

187
00:06:53,539 --> 00:06:54,974
THAT THE TOPICS OF NEUROETHICS,

188
00:06:54,974 --> 00:06:57,176
THE IMPACTS OF WHAT WE DO, WAS

189
00:06:57,176 --> 00:06:59,011
FULLY EMBEDDED IN THE PROGRAM IN

190
00:06:59,011 --> 00:07:01,180
MANY DIFFERENT WAYS.

191
00:07:01,180 --> 00:07:03,249
WE HAVE THIS BUILDING OUT A

192
00:07:03,249 --> 00:07:05,318
DECADE OF INNOVATION PANEL

193
00:07:05,318 --> 00:07:08,187
DISCUSSIONS WITH DOCTORS FRANCIS

194
00:07:08,187 --> 00:07:10,056
COLLINS, BARGMANN, NEWSOM, EDDIE

195
00:07:10,056 --> 00:07:10,256
CHANG.

196
00:07:10,256 --> 00:07:12,458
WE HAD SESSIONS LOOKING AT HIGH

197
00:07:12,458 --> 00:07:14,260
DENSITY AND HIGH RESOLUTION

198
00:07:14,260 --> 00:07:14,927
NEUROPHYSIOLOGICAL RECORDINGS OF

199
00:07:14,927 --> 00:07:16,762
THE HUMAN BRAIN, OF COURSE BEING

200
00:07:16,762 --> 00:07:18,164
MINDFUL OF THE ETHICAL

201
00:07:18,164 --> 00:07:19,165
IMPLICATIONS OF THAT.

202
00:07:19,165 --> 00:07:22,435
THERE WAS A GREAT PANEL ON

203
00:07:22,435 --> 00:07:26,806
ADVANCING PARTICIPANT ENGAGEMENT

204
00:07:26,806 --> 00:07:28,007
IN BRAIN-COMPUTER INTERFACE

205
00:07:28,007 --> 00:07:29,442
RESEARCH, BEYOND STANDING ROOM

206
00:07:29,442 --> 00:07:31,644
ONLY SPECIAL SESSION ON EMERGING

207
00:07:31,644 --> 00:07:32,712
PERSPECTIVES ON EMBODIED

208
00:07:32,712 --> 00:07:33,813
NEUROA.I. RESEARCH.

209
00:07:33,813 --> 00:07:35,348
WE'LL BE GETTING TO THAT QUITE A

210
00:07:35,348 --> 00:07:37,850
BIT AS THE DAY GOES ON.

211
00:07:37,850 --> 00:07:42,455
ALSO, LAYING THE FOUNDATION FOR

212
00:07:42,455 --> 00:07:44,523
NEW DATA AND ECOSYSTEMS CENTERED

213
00:07:44,523 --> 00:07:46,826
AROUND THE RECENTLY LAUNCHED

214
00:07:46,826 --> 00:07:50,896
BRAIN BEHAVIOR QUANTIFICATION

215
00:07:50,896 --> 00:07:52,264
AND SYNCHRONIZATION PROGRAM

216
00:07:52,264 --> 00:07:53,432
WE'RE EXCITED ABOUT.

217
00:07:53,432 --> 00:07:55,534
YOU CAN ACCESS MUCH OF THE

218
00:07:55,534 --> 00:07:57,303
CONTENT BY CLICKING ON THE QR

219
00:07:57,303 --> 00:07:59,438
CODE OR GOING TO THE WEBSITE.

220
00:07:59,438 --> 00:08:04,577
EVERYTHING NOW IS FINALLY

221
00:08:04,577 --> 00:08:04,877
ONLINE.

222
00:08:04,877 --> 00:08:07,279
UPCOMING EVENTS, THERE'S THIS

223
00:08:07,279 --> 00:08:09,382
DEEP BRAIN STIMULATION THINK

224
00:08:09,382 --> 00:08:11,917
TANK, UNIVERSITY OF FLORIDA,

225
00:08:11,917 --> 00:08:18,023
ACTUALLY STARTING TODAY.

226
00:08:18,023 --> 00:08:21,093
THERE'S THIS BRAIN-COMPUTER

227
00:08:21,093 --> 00:08:21,761
INTERFACE COMMUNITY THAT JEN

228
00:08:21,761 --> 00:08:23,496
FRENCH PLAYED A ROLE IN, IN

229
00:08:23,496 --> 00:08:23,963
SEPTEMBER.

230
00:08:23,963 --> 00:08:26,365
BROAD ONE ON HUMAN RESEARCH WITH

231
00:08:26,365 --> 00:08:27,566
A.I. SPONSORED BY DEPARTMENT OF

232
00:08:27,566 --> 00:08:29,435
HEALTH AND HUMAN SERVICES.

233
00:08:29,435 --> 00:08:35,441
THEN BACK TO BACK WITH THE IBCI

234
00:08:35,441 --> 00:08:37,643
COMMUNITY WORKSHOP, FDA-NIH

235
00:08:37,643 --> 00:08:42,314
SPONSORED WORKSHOP ON DEVELOPING

236
00:08:42,314 --> 00:08:42,848
IMPLANTED BRAIN-COMPUTER

237
00:08:42,848 --> 00:08:43,883
INTERFACES, WE'LL HAVE A

238
00:08:43,883 --> 00:08:45,851
PRESENCE THERE AS WELL.

239
00:08:45,851 --> 00:08:48,421
THESE ARE ALL VIRTUAL OR HYBRID

240
00:08:48,421 --> 00:08:49,889
SO WE ENCOURAGE YOU TO TUNE IN

241
00:08:49,889 --> 00:08:56,796
AND SEE WHAT'S GOING ON IN THOSE

242
00:08:56,796 --> 00:08:57,062
WORKSHOPS.

243
00:08:57,062 --> 00:09:04,270
NEUROETHICS HAS BEEN A GOOD WORD

244
00:09:04,270 --> 00:09:06,906
BEING SPREAD, THIS BUILDS AND

245
00:09:06,906 --> 00:09:08,541
EXPANDS ON BRAIN'S FOUNDATIONAL

246
00:09:08,541 --> 00:09:09,408
WORK IN NEUROETHICS, THE PURPOSE

247
00:09:09,408 --> 00:09:10,943
OF THIS SPECIAL NOTICE IS TO

248
00:09:10,943 --> 00:09:12,344
DRAW ATTENTION TO FOLK TO

249
00:09:12,344 --> 00:09:14,880
ENCOURAGE THEM TO APPLY FOR

250
00:09:14,880 --> 00:09:18,918
FUNDING IN THIS NEUROETHICS

251
00:09:18,918 --> 00:09:20,786
SPACE, RELEVANT PARENT

252
00:09:20,786 --> 00:09:22,655
ANNOUNCEMENTS LISTED BELOW,

253
00:09:22,655 --> 00:09:25,658
CONTACT IS DR. NINA HSU AT

254
00:09:25,658 --> 00:09:29,195
NINDS, QUITE EXCITED TO SEE THIS

255
00:09:29,195 --> 00:09:31,964
FIELD GROW BEYOND BRAIN INTO

256
00:09:31,964 --> 00:09:33,365
INSTITUTES AND CENTERS

257
00:09:33,365 --> 00:09:34,266
PARTICIPATING IN BRAIN.

258
00:09:34,266 --> 00:09:35,334
THERE'S A REQUEST FOR

259
00:09:35,334 --> 00:09:35,634
INFORMATION.

260
00:09:35,634 --> 00:09:37,536
THIS IS FROM THE COMMON FUND,

261
00:09:37,536 --> 00:09:39,071
NOT FROM "BRAIN," BUT QUITE

262
00:09:39,071 --> 00:09:40,139
RELATED TO THE DISCUSSION WE'RE

263
00:09:40,139 --> 00:09:42,942
GOING TO HAVE TODAY ABOUT THE

264
00:09:42,942 --> 00:09:45,077
CHALLENGES IN ADVANCING A.I.,

265
00:09:45,077 --> 00:09:45,978
INTEGRATING CLINICAL IMAGING

266
00:09:45,978 --> 00:09:46,946
WITH MULTI-LEVEL DATA, YOU CAN

267
00:09:46,946 --> 00:09:54,587
TAKE A LOOK AT THAT, QR CODE ON

268
00:09:54,587 --> 00:09:56,222
BOTTOM RYAN, DOCTORS FERRANTE

269
00:09:56,222 --> 00:10:00,059
AND SHARMA ARE THE CONTACTS.

270
00:10:00,059 --> 00:10:03,562
A FEW PUBLICATIONS THAT WERE

271
00:10:03,562 --> 00:10:05,331
SUPPORTED BY THE BRAIN

272
00:10:05,331 --> 00:10:08,267
NEUROETHICS R01 MECHANISM, A

273
00:10:08,267 --> 00:10:18,711
PAPER SUPPORTED BY AN R01 TO THE

274
00:10:18,711 --> 00:10:22,715
GROUPS LOOKING AT ADOLESCENT OCD

275
00:10:22,715 --> 00:10:23,916
PATIENTS AND CAREGIVER

276
00:10:23,916 --> 00:10:26,452
PERSPECTIVES ON POTENTIAL DEEP

277
00:10:26,452 --> 00:10:27,419
BRAIN STIMULATION TREATMENT, HOW

278
00:10:27,419 --> 00:10:28,854
DO WE ENGAGE WITH PATIENTS AND

279
00:10:28,854 --> 00:10:29,188
FAMILIES.

280
00:10:29,188 --> 00:10:32,258
A LOT OF DISCUSSION AROUND

281
00:10:32,258 --> 00:10:33,559
NEUROMODULATION INFLUENCES ON

282
00:10:33,559 --> 00:10:38,597
INDIVIDUAL PERSONALITIES, 

283
00:10:38,597 --> 00:10:40,566
IDENTITY AND AGENCY, RAISING

284
00:10:40,566 --> 00:10:43,636
ISSUES AROUND PEDIATRIC

285
00:10:43,636 --> 00:10:43,903
PATIENTS.

286
00:10:43,903 --> 00:10:47,540
HERE THEY STUDIED 21 PATIENTS

287
00:10:47,540 --> 00:10:50,976
WITH SEVERE OCD AND CAREGIVERS

288
00:10:50,976 --> 00:10:53,379
THINKING ABOUT FUTURE DBS

289
00:10:53,379 --> 00:10:55,481
TREATMENT, INTERESTING FINDINGS

290
00:10:55,481 --> 00:10:57,383
DOVETAILS NICELY WITH TOPICALLY

291
00:10:57,383 --> 00:11:01,954
COVERED IN FEBRUARY 2024 NEWG

292
00:11:01,954 --> 00:11:03,355
MEETING ON ETHICAL

293
00:11:03,355 --> 00:11:04,657
CONSIDERATIONS AROUND PEDIATRIC

294
00:11:04,657 --> 00:11:05,457
NEUROSTIMULATION, LOOK UP THIS

295
00:11:05,457 --> 00:11:07,760
PAPER AND YOU CAN LOOK AT THE

296
00:11:07,760 --> 00:11:09,528
DISCUSSION THAT WE HAD BACK IN

297
00:11:09,528 --> 00:11:13,132
FEBRUARY ON THIS MORE GENERAL

298
00:11:13,132 --> 00:11:13,332
TOPIC.

299
00:11:13,332 --> 00:11:17,069
THE SECOND PAPER I WANT TO

300
00:11:17,069 --> 00:11:18,504
HIGHLIGHT FROM SARAH GOERING'S

301
00:11:18,504 --> 00:11:20,039
GROUP, UNIVERSITY OF WASHINGTON,

302
00:11:20,039 --> 00:11:25,277
LOOKING AT THIS ISSUE OF MORAL

303
00:11:25,277 --> 00:11:28,547
ENTANGLEMENTS AS HUMAN RESEARCH

304
00:11:28,547 --> 00:11:31,283
PARTNERS PARTICIPANTS ARE

305
00:11:31,283 --> 00:11:32,718
ENGAGED IN HUMAN-BASED RESEARCH.

306
00:11:32,718 --> 00:11:35,921
WE HAVE TO RECOGNIZE THAT

307
00:11:35,921 --> 00:11:39,858
PARTICIPANTS HAVE SPECIFIC

308
00:11:39,858 --> 00:11:40,726
VULNERABLABILITIES MADE MORE

309
00:11:40,726 --> 00:11:41,727
COMPLICATION BECAUSE OF

310
00:11:41,727 --> 00:11:43,662
RELATIONSHIPS AND TRUST WITH

311
00:11:43,662 --> 00:11:45,664
RESEARCHERS, SURGEONS, WHAT HAVE

312
00:11:45,664 --> 00:11:45,931
YOU.

313
00:11:45,931 --> 00:11:49,234
THEY DELVE INTO THE ISSUE OF

314
00:11:49,234 --> 00:11:50,135
MORAL ENTANGLEMENT, INTERESTING

315
00:11:50,135 --> 00:11:53,505
AND TIMELY TOPIC FOR YOUR

316
00:11:53,505 --> 00:11:53,939
CONSIDERATION.

317
00:11:53,939 --> 00:11:55,140
THIS, AGAIN, DOVETAILS WITH A

318
00:11:55,140 --> 00:11:59,078
WORKSHOP WE HELD BACK IN MAY OF

319
00:11:59,078 --> 00:12:02,381
22, GOSH, IT'S ALREADY BEEN

320
00:12:02,381 --> 00:12:04,016
2022, ON CONTINUING TRIAL

321
00:12:04,016 --> 00:12:04,984
RESPONSIBILITIES, AND YOU CAN

322
00:12:04,984 --> 00:12:06,752
LOOK UP THAT WITH THAT QR CODE

323
00:12:06,752 --> 00:12:07,753
RIGHT THERE.

324
00:12:07,753 --> 00:12:08,387
THAT'S IT.

325
00:12:08,387 --> 00:12:10,923
I WANT TO GET US ON TO BUSINESS

326
00:12:10,923 --> 00:12:12,725
TO THE FUN PART OF THE

327
00:12:12,725 --> 00:12:13,025
DISCUSSION.

328
00:12:13,025 --> 00:12:15,561
YOU CAN LEARN MORE ABOUT

329
00:12:15,561 --> 00:12:16,328
EVERYTHING BRAIN BY SUBSCRIBING

330
00:12:16,328 --> 00:12:17,396
TO THE BRAIN BLOCK.

331
00:12:17,396 --> 00:12:22,001
WE PROMISE NOT TO CLOG YOUR

332
00:12:22,001 --> 00:12:23,636
E-MAIL IN-BOX BUT YOU'LL GET

333
00:12:23,636 --> 00:12:25,938
GOOD INFORMATION ABOUT STUFF

334
00:12:25,938 --> 00:12:28,574
COMING UP, WEBSITE, GUIDE

335
00:12:28,574 --> 00:12:28,907
NOTICES.

336
00:12:28,907 --> 00:12:30,309
THE BEST SOURCE IS CONTACT YOUR

337
00:12:30,309 --> 00:12:30,976
PROGRAM OFFICER.

338
00:12:30,976 --> 00:12:32,277
WE'RE MORE THAN HAPPY TO TALK TO

339
00:12:32,277 --> 00:12:33,912
YOU FOLKS AND SEE HOW WE CAN

340
00:12:33,912 --> 00:12:35,881
ENABLE YOUR RESEARCH AND YOU CAN

341
00:12:35,881 --> 00:12:42,154
FIND US AT BRAIN.GOV. 

342
00:12:42,154 --> 00:12:42,921
I WILL STOP THERE.

343
00:12:42,921 --> 00:12:45,858
I DON'T KNOW IF WE HAVE TIME FOR

344
00:12:45,858 --> 00:12:47,292
QUESTIONS OR COMMENTS OR WE

345
00:12:47,292 --> 00:12:50,396
SHOULD JUST DIVE INTO THE GOOD

346
00:12:50,396 --> 00:12:59,138
STUFF. 

347
00:12:59,138 --> 00:13:01,306
>> LOOKS LIKE NOT SEEING
ANY

348
00:13:01,306 --> 00:13:02,107
QUESTIONS, JOHN.

349
00:13:02,107 --> 00:13:03,642
IF ANYONE HAS QUESTIONS OR

350
00:13:03,642 --> 00:13:05,277
COMMENTS, OTHERWISE I'M GOING TO

351
00:13:05,277 --> 00:13:07,379
JUST HAND THE WHOLE THING OVER

352
00:13:07,379 --> 00:13:08,881
TO CHRISTINE AND NITA TO WORK

353
00:13:08,881 --> 00:13:19,358
OUR WAY THROUGH THE AGENDA. 

354
00:13:21,093 --> 00:13:22,161
>> WE'RE LUCKY TO HAVE
WITH US

355
00:13:22,161 --> 00:13:24,697
SEVERAL PEOPLE TO GIVE US

356
00:13:24,697 --> 00:13:27,866
BACKGROUND INFORMATION TO BE

357
00:13:27,866 --> 00:13:29,501
ABLE TO GET STARTED ON THE

358
00:13:29,501 --> 00:13:31,804
DISCUSSION OF THE ETHICS.

359
00:13:31,804 --> 00:13:32,905
SO OUR FIRST SPEAKER WHO IS

360
00:13:32,905 --> 00:13:34,540
GOING TO GIVE AN INTRODUCTION

361
00:13:34,540 --> 00:13:36,709
AND OVERVIEW OF A.I.

362
00:13:36,709 --> 00:13:38,811
NEUROSCIENCE AND ETHICS IS DR.

363
00:13:38,811 --> 00:13:42,848
JOE MONACO FROM THE NINDS.

364
00:13:42,848 --> 00:13:43,148
DR. MONACO?

365
00:13:43,148 --> 00:13:45,584
>> GREAT, THANKS SO MUCH.

366
00:13:45,584 --> 00:13:50,422
I'M GOING TO SHARE MY SCREEN.

367
00:13:50,422 --> 00:13:52,825
I ASSUME YOU CAN SEE THAT.

368
00:13:52,825 --> 00:13:53,926
THANKS FOR HAVING ME.

369
00:13:53,926 --> 00:13:55,561
JOHN SAYS THIS IS THE GOOD

370
00:13:55,561 --> 00:13:56,995
STUFF, AND I HOPE IT IS.

371
00:13:56,995 --> 00:13:59,498
I WANT TO GIVE A VERY BRIEF KIND

372
00:13:59,498 --> 00:14:00,799
OF OVERVIEW OF WHAT WE'RE

373
00:14:00,799 --> 00:14:02,901
TALKING ABOUT WHEN WE'RE TALKING

374
00:14:02,901 --> 00:14:06,605
ABOUT A.I. AND HOW WE CAN APPLY

375
00:14:06,605 --> 00:14:07,806
TO NEUROETHICS.

376
00:14:07,806 --> 00:14:09,108
IT'S FOUND MANY APPLICATIONS AND

377
00:14:09,108 --> 00:14:11,443
WILL FIND MANY MORE IN THE

378
00:14:11,443 --> 00:14:11,744
FUTURE.

379
00:14:11,744 --> 00:14:20,619
I'LL INTRODUCE OUR SUBSEQUENT

380
00:14:20,619 --> 00:14:28,160
SPEAKERS, DORIS T.S.A.O AND

381
00:14:28,160 --> 00:14:31,663
PATRICK MINUTE MINEAUALT. 

382
00:14:31,663 --> 00:14:34,767
THE SECOND HALF WILL INVOLVE

383
00:14:34,767 --> 00:14:36,835
CASE STUDIES, APPLYING A.I. TO

384
00:14:36,835 --> 00:14:39,571
NEUROETHICS AND BRAIN HEALTH,

385
00:14:39,571 --> 00:14:45,244
FOCUSING ON DEEP PHENOTYPING AND

386
00:14:45,244 --> 00:14:46,645
TRAINING LARGE TRANSFORMER-BASED

387
00:14:46,645 --> 00:14:49,281
NEURAL NETWORKS MODELS INCLUDING

388
00:14:49,281 --> 00:14:51,083
MODALITIES LIKE EEG.

389
00:14:51,083 --> 00:14:55,554
SO I WANT TO QUICKLY STEP BACK

390
00:14:55,554 --> 00:14:59,158
AND SAY WHAT IS A.I., WHAT DO WE

391
00:14:59,158 --> 00:14:59,691
MEAN?

392
00:14:59,691 --> 00:15:01,460
PEOPLE HAVE DIFFERENT IDEAS, BUT

393
00:15:01,460 --> 00:15:02,661
AT THE FOUNDATION MODERN

394
00:15:02,661 --> 00:15:05,063
TECHNOLOGY THAT WE CALL A.I. IS

395
00:15:05,063 --> 00:15:06,131
A COMPUTING TECHNOLOGY THAT'S

396
00:15:06,131 --> 00:15:09,535
BASED ON A PARTICULAR MODEL THAT

397
00:15:09,535 --> 00:15:12,004
WE CALLED ARTIFICIAL NEURAL

398
00:15:12,004 --> 00:15:13,639
NETWORK MODEL, ANNs ARE

399
00:15:13,639 --> 00:15:14,072
SIMPLE.

400
00:15:14,072 --> 00:15:16,275
THEY ARE JUST NETWORK MODELS

401
00:15:16,275 --> 00:15:17,276
ESSENTIALLY GRAPHS WHERE YOU

402
00:15:17,276 --> 00:15:19,444
HAVE SIMPLE NODES AND YOU HAVE

403
00:15:19,444 --> 00:15:20,779
CONNECTIONS BETWEEN THOSE NODES

404
00:15:20,779 --> 00:15:22,314
AND THOSE CONNECTIONS ARE

405
00:15:22,314 --> 00:15:24,483
WEIGHTED, AND THE WEIGHTS OF

406
00:15:24,483 --> 00:15:27,252
THOSE CONNECTIONS ARE ADJUSTED

407
00:15:27,252 --> 00:15:30,856
DUE TO LEARNING RULES APPLIED TO

408
00:15:30,856 --> 00:15:31,056
DATA.

409
00:15:31,056 --> 00:15:34,560
THE REASONS EXPLODED OVER 10 OR

410
00:15:34,560 --> 00:15:36,762
12 YEARS, CERTAIN TECHNICAL

411
00:15:36,762 --> 00:15:37,529
BARRIERS WERE SURPASSED.

412
00:15:37,529 --> 00:15:38,964
FOR A LONG TIME THESE TYPES OF

413
00:15:38,964 --> 00:15:43,669
MODELS COULD NOT BE SCALED UP

414
00:15:43,669 --> 00:15:46,371
BECAUSE WE DIDN'T HAVE THE

415
00:15:46,371 --> 00:15:48,807
COMPUTING HARDWARE TO

416
00:15:48,807 --> 00:15:51,877
PARALLELIZE THE DATA, TO FLOW AS

417
00:15:51,877 --> 00:15:54,279
FAST THROUGH THE SYSTEMS AS WE

418
00:15:54,279 --> 00:15:59,952
NEED TO TEST LARGE SYSTEMS, BUT

419
00:15:59,952 --> 00:16:01,653
AROUND 2012 WITH ALEX AND THE

420
00:16:01,653 --> 00:16:04,056
COMPUTER VISION COMMUNITY THESE

421
00:16:04,056 --> 00:16:05,691
BARRIERS WERE BROKEN AND NOW

422
00:16:05,691 --> 00:16:07,559
THAT BASICALLY SET A.I. OFF TO

423
00:16:07,559 --> 00:16:09,828
THE RACES, AS WE ALL KNOW.

424
00:16:09,828 --> 00:16:11,930
WE CAN NOW THROW MORE AND MORE

425
00:16:11,930 --> 00:16:13,565
DATA AND MORE AND MORE COMPUTE

426
00:16:13,565 --> 00:16:17,369
AT THESE MODELS TO BUILD THEM

427
00:16:17,369 --> 00:16:20,539
UP, AND THEY BASICALLY LEARN

428
00:16:20,539 --> 00:16:23,308
FROM HIGH DIMENSIONAL COMPLEX

429
00:16:23,308 --> 00:16:24,409
INPUT-OUTPUT RELATIONSHIPS THAT

430
00:16:24,409 --> 00:16:25,811
ARE OTHERWISE VERY DIFFICULT TO

431
00:16:25,811 --> 00:16:26,578
PROGRAM MANUALLY.

432
00:16:26,578 --> 00:16:30,215
SO IN A SENSE THEY ARE EXPLORING

433
00:16:30,215 --> 00:16:31,216
THE POSSIBLE PROGRAM SPACE TO

434
00:16:31,216 --> 00:16:34,052
SOLVE PROBLEMS OF INTEREST.

435
00:16:34,052 --> 00:16:36,555
AND SO THIS IS TECHNOLOGY THAT

436
00:16:36,555 --> 00:16:38,323
WAS OBVIOUS, WIDE APPLICABILITY.

437
00:16:38,323 --> 00:16:40,459
AND SO TO KIND OF PROVIDE THE

438
00:16:40,459 --> 00:16:42,928
LONG VIEW HERE, THIS IS A

439
00:16:42,928 --> 00:16:45,764
COMPLEX TIMELINE, IT'S NOT

440
00:16:45,764 --> 00:16:47,499
CRITICAL TO UNDERSTANDING ALL OF

441
00:16:47,499 --> 00:16:48,700
THE DETAILS HERE.

442
00:16:48,700 --> 00:16:53,405
THE TOP PURPLE LINE THAT YOU CAN

443
00:16:53,405 --> 00:16:56,575
SEE ARE LANDMARK ADVANCES IN

444
00:16:56,575 --> 00:16:56,909
NEUROSCIENCE.

445
00:16:56,909 --> 00:16:58,877
AND THE BOTTOM THREE TIMELINES,

446
00:16:58,877 --> 00:17:01,947
BLUE, ORANGE AND GREEN, SHOW

447
00:17:01,947 --> 00:17:04,149
DIFFERENT BRANCHES OF THIS

448
00:17:04,149 --> 00:17:06,018
TECHNOLOGY THAT FALLS UNDER THE

449
00:17:06,018 --> 00:17:09,421
UMBRELLA OF A.I., ANNs,

450
00:17:09,421 --> 00:17:10,956
MACHINE LEARNING, STATISTICAL

451
00:17:10,956 --> 00:17:12,925
LEARNING, AND RELATED METHODS.

452
00:17:12,925 --> 00:17:16,194
AND SO THERE'S BEEN A FEW POINTS

453
00:17:16,194 --> 00:17:19,831
IN THE LAST MANY DECADES

454
00:17:19,831 --> 00:17:20,432
ESSENTIALLY WHERE SOME

455
00:17:20,432 --> 00:17:22,434
INSPIRATION FROM THE BRAIN LED

456
00:17:22,434 --> 00:17:24,970
TO ADVANCES IN THE A.I.

457
00:17:24,970 --> 00:17:26,371
TECHNOLOGY, OR IN HOW WE

458
00:17:26,371 --> 00:17:29,207
STRUCTURE THESE TYPES OF MODELS,

459
00:17:29,207 --> 00:17:30,542
PARTICULARLY AROUND HOW WE TRAIN

460
00:17:30,542 --> 00:17:30,876
THEM.

461
00:17:30,876 --> 00:17:35,447
SO IT'S ONLY UNTIL THE '90s

462
00:17:35,447 --> 00:17:38,216
WHEN THE ARABIC PROPAGATION RULE

463
00:17:38,216 --> 00:17:39,985
WAS FORMALIZED, ABLE TO STABLY

464
00:17:39,985 --> 00:17:40,953
TRAIN THESE NETWORKS TO DO

465
00:17:40,953 --> 00:17:41,720
USEFUL THINGS.

466
00:17:41,720 --> 00:17:44,489
WE'RE LOOKING AT 30 YEARS OF

467
00:17:44,489 --> 00:17:45,958
HISTORY THERE.

468
00:17:45,958 --> 00:17:48,827
SO THAT LEADS US TO THE PRESENT

469
00:17:48,827 --> 00:17:51,229
WHERE WE HAVE 10 OR 12 YEARS OF

470
00:17:51,229 --> 00:17:52,331
INFLECTION POINT IN MODERN A.I.,

471
00:17:52,331 --> 00:17:54,066
OF COURSE WE NOW HAVE 10 YEARS

472
00:17:54,066 --> 00:17:56,068
OF THE BRAIN INITIATIVE ITSELF,

473
00:17:56,068 --> 00:18:00,439
SO THIS IS AN AMAZING POINT OF

474
00:18:00,439 --> 00:18:02,407
CONVERGENCE REALLY BECAUSE NOW

475
00:18:02,407 --> 00:18:04,910
WE HAVE THIS EXCELLENT

476
00:18:04,910 --> 00:18:06,445
TECHNOLOGICAL TOOLBOX FOR

477
00:18:06,445 --> 00:18:08,413
ANALYZING AND UNDERSTANDING VAST

478
00:18:08,413 --> 00:18:11,149
QUANTITIES OF COMPLEX DATA, AND

479
00:18:11,149 --> 00:18:12,884
THE BRAIN INITIATIVE HAS BEEN

480
00:18:12,884 --> 00:18:15,620
TEN YEARS GENERATING AND

481
00:18:15,620 --> 00:18:17,622
COLLECTING LARGE SCALE KNOWLEDGE

482
00:18:17,622 --> 00:18:22,427
BASES AND DATASETS ON AN

483
00:18:22,427 --> 00:18:23,295
UNPRECEDENTED SCALES USING

484
00:18:23,295 --> 00:18:24,730
NEUROTECHNOLOGIES THAT IT HAS

485
00:18:24,730 --> 00:18:24,997
INNOVATED.

486
00:18:24,997 --> 00:18:26,765
SO WE CAN USE A.I.-BASED TOOLS

487
00:18:26,765 --> 00:18:30,535
TO DO ALL SORTS OF THINGS TO

488
00:18:30,535 --> 00:18:33,171
UNDERSTAND NEURAL DATA, SO WE

489
00:18:33,171 --> 00:18:34,339
CAN LEARN COMPLEX MODELS OF

490
00:18:34,339 --> 00:18:35,273
NEURAL ACTIVITY.

491
00:18:35,273 --> 00:18:38,944
WE CAN LEARN COMPLEX MODELS OF

492
00:18:38,944 --> 00:18:40,946
COMPLEX BEHAVIORS.

493
00:18:40,946 --> 00:18:43,215
WE CAN TREAT STRUCTURE AND

494
00:18:43,215 --> 00:18:44,216
ACTIVITY AS JOINT OPTIMIZATION

495
00:18:44,216 --> 00:18:46,952
PROGRAM AND SEE WHAT COMES OUT

496
00:18:46,952 --> 00:18:49,688
OF THAT, MAYBE FUNDAMENTAL

497
00:18:49,688 --> 00:18:51,256
COMPUTATIONAL PRINCIPLES, A LOT

498
00:18:51,256 --> 00:18:52,391
OF GREAT OPPORTUNITIES THAT ARE

499
00:18:52,391 --> 00:18:56,261
IN THE CURRENT STATE OF THE ART

500
00:18:56,261 --> 00:18:58,096
IN NEUROSCIENCE AND MANY MORE

501
00:18:58,096 --> 00:19:01,400
GOING FORWARD. 

502
00:19:01,400 --> 00:19:02,601
THERE'S A GREAT REVIEW FROM WANG

503
00:19:02,601 --> 00:19:05,103
A COUPLE YEARS AGO LAYING OUT

504
00:19:05,103 --> 00:19:07,739
THE LANDSCAPE HERE, AND

505
00:19:07,739 --> 00:19:09,174
ESSENTIALLY FOUR MAJOR KINDS OF

506
00:19:09,174 --> 00:19:12,044
MODELS THAT WE CAN THINK ABOUT

507
00:19:12,044 --> 00:19:14,046
THAT ARE IN USE, AND SO

508
00:19:14,046 --> 00:19:15,180
BASICALLY, YOU KNOW, SINCE THESE

509
00:19:15,180 --> 00:19:16,281
THINGS ARE JUST GRAPHS AND

510
00:19:16,281 --> 00:19:17,716
NETWORKS OF NODES YOU CAN

511
00:19:17,716 --> 00:19:18,817
ARRANGE THEM IN DIFFERENT WAYS

512
00:19:18,817 --> 00:19:23,288
TO DO DIFFERENT THINGS. 

513
00:19:23,288 --> 00:19:25,257
SO THAT STRUCTURE, OFTEN YOU

514
00:19:25,257 --> 00:19:27,059
WANT THAT TO MAP ON THE

515
00:19:27,059 --> 00:19:31,396
STRUCTURE YOU'RE TRYING TO

516
00:19:31,396 --> 00:19:35,100
SOLVE. 

517
00:19:35,100 --> 00:19:38,804
STOP LEFT, MULTI-LAYER

518
00:19:38,804 --> 00:19:41,673
PERCEPTRONS, AND TOP RIGHT

519
00:19:41,673 --> 00:19:45,944
RANDOM ASSORTMENT OF NODES THAT

520
00:19:45,944 --> 00:19:49,781
ARE RANDOMLY INTERCONNECTED,

521
00:19:49,781 --> 00:19:50,649
DYNAMIC CONNECTED NEURAL

522
00:19:50,649 --> 00:19:52,617
NETWORK, TRAINED TO LEARN TIME

523
00:19:52,617 --> 00:19:54,586
SERIES AND DYNAMICAL DATA

524
00:19:54,586 --> 00:19:54,920
PATTERNS.

525
00:19:54,920 --> 00:19:56,888
LOWER LEFT YOU CAN UNROLL A

526
00:19:56,888 --> 00:19:58,957
RECURRENT NETWORK IN ORDER TO

527
00:19:58,957 --> 00:20:00,358
APPLY IT TO SEQUENCES AND TIME

528
00:20:00,358 --> 00:20:03,662
SERIES OF DATA SO YOU CAN MAKE

529
00:20:03,662 --> 00:20:04,963
PREDICTIONS ABOUT WHAT TYPE OF

530
00:20:04,963 --> 00:20:06,264
DATA OR WHAT DISTRIBUTION OF

531
00:20:06,264 --> 00:20:09,334
DATA IS THAT YOU CAN EXPECT TO

532
00:20:09,334 --> 00:20:09,901
COME NEXT.

533
00:20:09,901 --> 00:20:11,736
AND THERE'S ALL SORTS OF WAYS

534
00:20:11,736 --> 00:20:16,141
YOU CAN TAKE WHAT ARE OFTEN

535
00:20:16,141 --> 00:20:16,942
CALLED INDUCTIVE BIASES,

536
00:20:16,942 --> 00:20:19,211
STRUCTURES FROM BRAINS, APPLY TO

537
00:20:19,211 --> 00:20:20,979
STRUCTURES OF NEURAL NETWORK

538
00:20:20,979 --> 00:20:24,349
MODELS SUCH AS FOR VISION AND

539
00:20:24,349 --> 00:20:27,085
CONVOLUTIONAL NEURAL NETWORKS.

540
00:20:27,085 --> 00:20:30,055
DORIS WILL TALK MORE ABOUT THAT.

541
00:20:30,055 --> 00:20:32,023
BUT THESE SPAN THE SPACE OF

542
00:20:32,023 --> 00:20:35,060
PROBLEMS THAT WE CAN APPLY A.I.

543
00:20:35,060 --> 00:20:37,362
MODELS TO.

544
00:20:37,362 --> 00:20:39,464
IT'S IMPORTANT TO REMEMBER WHAT

545
00:20:39,464 --> 00:20:40,465
THE SYSTEMS ARE.

546
00:20:40,465 --> 00:20:42,100
AND SO BLAKE RICHARDS AND A

547
00:20:42,100 --> 00:20:43,635
NUMBER OF CO-AUTHORS FROM A FEW

548
00:20:43,635 --> 00:20:47,139
YEARS AGO LAID OUT A GREAT

549
00:20:47,139 --> 00:20:48,673
FRAMEWORK CLARIFYING THAT

550
00:20:48,673 --> 00:20:50,075
FUNDAMENTALLY THERE'S THREE

551
00:20:50,075 --> 00:20:54,012
COMPONENTS TO ALL OF THIS STUFF,

552
00:20:54,012 --> 00:20:55,447
WHICH IS ONE ARCHITECTURE, HOW

553
00:20:55,447 --> 00:20:57,415
ARE NODES CONNECTED, WHAT DOES

554
00:20:57,415 --> 00:21:02,120
THE STRUCTURE LOOK  LIKE.

555
00:21:02,120 --> 00:21:04,189
LEARNING RULE, HOW DO YOU

556
00:21:04,189 --> 00:21:05,957
UPDATE, AND THE OBJECTIVE

557
00:21:05,957 --> 00:21:07,392
FUNCTION, WHAT IS THE

558
00:21:07,392 --> 00:21:08,493
MEASUREMENT OF PERFORMANCE, WHAT

559
00:21:08,493 --> 00:21:11,663
IS THE TASK YOU'RE TRYING TO

560
00:21:11,663 --> 00:21:13,732
LEARN, WHAT IS THE INPUT/OUTPUT

561
00:21:13,732 --> 00:21:15,167
MAPPING YOU WANT TO COME OUT OF

562
00:21:15,167 --> 00:21:15,534
THE MODEL.

563
00:21:15,534 --> 00:21:17,202
THOSE ARE THE THREE ELEMENTS.

564
00:21:17,202 --> 00:21:18,904
AND YOU CAN COME UP WITH

565
00:21:18,904 --> 00:21:20,772
DIFFERENT WAYS OF PUTTING THESE

566
00:21:20,772 --> 00:21:23,074
TOGETHER BUT THERE'S -- YOU CAN

567
00:21:23,074 --> 00:21:25,577
KIND OF MAYBE INTUIT THAT

568
00:21:25,577 --> 00:21:26,478
THERE'S NO FUNDAMENTAL GUARD

569
00:21:26,478 --> 00:21:29,181
RAILS IN THIS KIND OF SYSTEM.

570
00:21:29,181 --> 00:21:30,715
YOU TRAIN THIS UNTIL YOU DON'T

571
00:21:30,715 --> 00:21:35,754
HAVE ANY MORE ERROR AGAINST YOUR

572
00:21:35,754 --> 00:21:38,256
POLLED OUT TEST DATASET, BUT

573
00:21:38,256 --> 00:21:40,225
THERE ARE NO FUNDAMENTAL

574
00:21:40,225 --> 00:21:41,960
GUARANTEES WILL THE BIASES THAT

575
00:21:41,960 --> 00:21:45,497
CAN COME, THAT CAN BE LEARNED BY

576
00:21:45,497 --> 00:21:48,166
THESE MODELS. 

577
00:21:48,166 --> 00:21:49,868
THAT'S A MAJOR KIND OF CONCERN

578
00:21:49,868 --> 00:21:53,071
AS FAR AS POTENTIAL ETHICAL

579
00:21:53,071 --> 00:21:53,838
CONSIDERATIONS OF APPLICATIONS

580
00:21:53,838 --> 00:21:54,606
HERE.

581
00:21:54,606 --> 00:21:57,709
I WANT TO QUICKLY GO OVER HOW

582
00:21:57,709 --> 00:22:01,413
CAN WE USE THESE IN

583
00:22:01,413 --> 00:22:04,049
NEUROSCIENCE, WE KNOW ABOUT

584
00:22:04,049 --> 00:22:06,117
CHATBOTS AND LLMs, THESE

585
00:22:06,117 --> 00:22:07,185
MODELS CAN INGEST LARGE

586
00:22:07,185 --> 00:22:08,420
QUANTITIES OF TEXT DATA, VERY

587
00:22:08,420 --> 00:22:09,955
IMPORTANT, SO THOSE CAN BE USED

588
00:22:09,955 --> 00:22:12,023
TO KIND OF EXPLORE THE

589
00:22:12,023 --> 00:22:18,597
LITERATURE IN A WAY THAT CAN

590
00:22:18,597 --> 00:22:20,966
HELP MODELS DEVELOP THEORIES,

591
00:22:20,966 --> 00:22:21,633
EXPERIMENTALISTS UNDERSTAND WHAT

592
00:22:21,633 --> 00:22:23,235
HAS BEEN DONE OR NOT DONE TO

593
00:22:23,235 --> 00:22:24,469
DRIVE FUTURE SCIENCE.

594
00:22:24,469 --> 00:22:27,872
YOU CAN USE THESE TYPES OF A.I.

595
00:22:27,872 --> 00:22:30,141
MODELS TO SUPPORT OR FACILITATE

596
00:22:30,141 --> 00:22:32,110
DETAILED IN SILICO MODELS THAT

597
00:22:32,110 --> 00:22:33,645
INTERACT WITH THE PHYSICAL WORLD

598
00:22:33,645 --> 00:22:35,180
IN IMPORTANT WAYS.

599
00:22:35,180 --> 00:22:37,816
WE CAN THINK ABOUT THIS AS

600
00:22:37,816 --> 00:22:40,218
ENCODING MODELS TO HELP US

601
00:22:40,218 --> 00:22:41,186
ADVANCE THEORETICAL

602
00:22:41,186 --> 00:22:42,621
UNDERSTANDING, DECODING MODELS

603
00:22:42,621 --> 00:22:45,357
TO HELP PREDICTION AND CONTROL,

604
00:22:45,357 --> 00:22:47,325
AND THINK ABOUT TRANSLATING

605
00:22:47,325 --> 00:22:50,161
THOSE TYPES OF MODELS INTO, YOU

606
00:22:50,161 --> 00:22:54,332
KNOW, REAL WORLD DEVICES

607
00:22:54,332 --> 00:22:57,402
INCLUDING BRAIN-COMPUTER

608
00:22:57,402 --> 00:22:59,271
INTERFACE, BCI, DBS AND NEURO

609
00:22:59,271 --> 00:23:05,477
PROSTHETICS, A LARGE  ROAD MAP F

610
00:23:05,477 --> 00:23:05,810
POSSIBILITIES.

611
00:23:05,810 --> 00:23:06,578
EXAMPLE OF GENERATIVE MODEL THAT

612
00:23:06,578 --> 00:23:08,480
CAN BE TRAINED IN THIS SPACE, A

613
00:23:08,480 --> 00:23:10,548
RESEARCH TEAM BASED OUT OF

614
00:23:10,548 --> 00:23:11,316
UNIVERSITY OF TORONTO THAT

615
00:23:11,316 --> 00:23:16,221
SHOWED YOU CAN TAKE A LARGE DATA

616
00:23:16,221 --> 00:23:18,523
REPOSITORY OF MOLECULAR

617
00:23:18,523 --> 00:23:20,191
SINGLE-CELL CHARACTERIZATIONS IN

618
00:23:20,191 --> 00:23:22,994
THIS CASE 33 MILLION CELLS, AND

619
00:23:22,994 --> 00:23:26,831
TRAIN A GENERATIVE FOUNDATION

620
00:23:26,831 --> 00:23:29,034
MODEL WITH THIS MECHANISM, AND

621
00:23:29,034 --> 00:23:32,437
IT CAN BE USED AS A BASE MODEL

622
00:23:32,437 --> 00:23:33,838
TO PREDICT AND IMPUTE DATA AND

623
00:23:33,838 --> 00:23:35,040
FIGURE OUT STRUCTURE OF WHAT

624
00:23:35,040 --> 00:23:36,808
TYPES OF CELLS ARE RELATED TO

625
00:23:36,808 --> 00:23:40,111
OTHER TYPES OF CELLS, AND

626
00:23:40,111 --> 00:23:40,979
BASICALLY INFER THE STRUCTURE OF

627
00:23:40,979 --> 00:23:42,614
CELLS IN THE HUMAN BODY.

628
00:23:42,614 --> 00:23:45,050
SO THIS PARTICULAR STUDY WAS NOT

629
00:23:45,050 --> 00:23:46,351
RESTRICTED TO JUST BRAIN CELLS

630
00:23:46,351 --> 00:23:47,986
BUT EXACT SAME KIND OF METHOD

631
00:23:47,986 --> 00:23:50,088
CAN BE APPLIED TO KNOWLEDGE

632
00:23:50,088 --> 00:23:53,792
BASES AND CELL ATLASES SUCH AS

633
00:23:53,792 --> 00:23:54,993
BEING CONSTRUCTED BY THE

634
00:23:54,993 --> 00:23:56,528
TRANSFORMATIVE PROJECT OF THE

635
00:23:56,528 --> 00:23:57,729
BRAIN INITIATIVE, AND THEY

636
00:23:57,729 --> 00:23:59,597
PROVIDE A GREAT EXAMPLE WHERE

637
00:23:59,597 --> 00:24:01,766
THEY PRE-TRAINED THIS LARGE BASE

638
00:24:01,766 --> 00:24:03,435
MODEL BUT THEN THEY HAD A

639
00:24:03,435 --> 00:24:04,969
DIFFERENT DOWNSTREAM USE.

640
00:24:04,969 --> 00:24:06,471
THEY WANTED TO THEN APPLY THE

641
00:24:06,471 --> 00:24:09,541
SAME BASE MODEL TO A DIFFERENT

642
00:24:09,541 --> 00:24:09,974
PROBLEM.

643
00:24:09,974 --> 00:24:12,510
SO THEY USED A CERTAIN FORM OF

644
00:24:12,510 --> 00:24:14,012
TRANSFER LEARNING TO COME UP

645
00:24:14,012 --> 00:24:15,347
WITH A SLIGHTLY NEW OBJECTIVE

646
00:24:15,347 --> 00:24:17,649
FUNCTION, AND THEY WERE ABLE TO

647
00:24:17,649 --> 00:24:19,951
FINE TUNE THE REPRESENTATIONS IN

648
00:24:19,951 --> 00:24:21,786
THIS MODEL TO ANSWER DIFFERENT

649
00:24:21,786 --> 00:24:23,121
KINDS OF QUESTIONS.

650
00:24:23,121 --> 00:24:25,523
AND SO THAT'S ONE OF THE MAIN

651
00:24:25,523 --> 00:24:26,858
QUESTIONS THAT, YOU KNOW, WE

652
00:24:26,858 --> 00:24:29,461
WANT TO BE THINKING ABOUT AT THE

653
00:24:29,461 --> 00:24:32,430
MEETING TODAY IS THE DIFFERENT

654
00:24:32,430 --> 00:24:34,065
USES FOR DOMAIN-SPECIFIC MODELS,

655
00:24:34,065 --> 00:24:35,700
FOUNDATIONAL BASE MODELS, OR HOW

656
00:24:35,700 --> 00:24:37,469
TO FINE TUNE MODELS WITH NEW

657
00:24:37,469 --> 00:24:38,670
DATA AND WHAT ARE IMPLICATIONS

658
00:24:38,670 --> 00:24:41,639
FOR THOSE KINDS OF USES.

659
00:24:41,639 --> 00:24:44,609
AND JUST AS ONE OTHER EXAMPLE OF

660
00:24:44,609 --> 00:24:46,945
USING KIND OF THESE LANGUAGE

661
00:24:46,945 --> 00:24:52,250
FOCUSED MODELS FROM A FEW YEARS

662
00:24:52,250 --> 00:24:55,653
AGO, AT STANFORD IN THE

663
00:24:55,653 --> 00:24:57,622
NEUROIMAGING RESEARCH COMMUNITY

664
00:24:57,622 --> 00:24:59,124
USED NATURAL LANGUAGE PROCESSING

665
00:24:59,124 --> 00:25:01,326
TECHNIQUES, MACHINE LEARNING

666
00:25:01,326 --> 00:25:03,628
TECHNIQUES TO AGGREGATE OVER

667
00:25:03,628 --> 00:25:06,264
18,000 STUDIES OF NEUROIMAGING

668
00:25:06,264 --> 00:25:07,999
RESULTS AND ASSOCIATED DATASETS,

669
00:25:07,999 --> 00:25:09,768
WITH fMRI AND PET DATA

670
00:25:09,768 --> 00:25:12,036
MODALITIES, AND THEY WERE

671
00:25:12,036 --> 00:25:16,441
ESSENTIALLY ABLE TO FORM A

672
00:25:16,441 --> 00:25:17,308
BOTTOM-UNLEARNED ONTOLOGY OF

673
00:25:17,308 --> 00:25:20,812
HUMAN  COGNITION AND BRAIN

674
00:25:20,812 --> 00:25:22,013
SYSTEMS THAT COULD BE LEARNED

675
00:25:22,013 --> 00:25:24,516
OUT OF THE VAST AMOUNTS OF

676
00:25:24,516 --> 00:25:25,850
LITERATURE AND ASSOCIATED DATA.

677
00:25:25,850 --> 00:25:28,353
AND SO JUST BY THINKING DEEPLY

678
00:25:28,353 --> 00:25:31,556
ABOUT HOW TO LINK KEY TYPES OF

679
00:25:31,556 --> 00:25:33,391
DATA WE CAN CONSTRUCT, YOU KNOW,

680
00:25:33,391 --> 00:25:37,128
AMAZING NEW KNOWLEDGE AND DATA

681
00:25:37,128 --> 00:25:38,563
RESOURCES THAT CAN BENEFIT, YOU

682
00:25:38,563 --> 00:25:40,732
KNOW, PARTICULAR COMMUNITIES AND

683
00:25:40,732 --> 00:25:41,966
SCIENCE AS A WHOLE.

684
00:25:41,966 --> 00:25:44,803
AND ONE OF THE MOTIVATIONS FOR

685
00:25:44,803 --> 00:25:46,538
THIS KIND OF WORK INDIVIDUAL

686
00:25:46,538 --> 00:25:49,407
SCIENTISTS IN THE LABS HAVE

687
00:25:49,407 --> 00:25:51,042
HUMAN BIASES WHEN INTERPRETING

688
00:25:51,042 --> 00:25:51,376
DATA.

689
00:25:51,376 --> 00:25:54,779
IF WE HAVE THIS, YOU KNOW,

690
00:25:54,779 --> 00:25:56,514
BASICALLY NEUTRAL OR SUPPOSEDLY

691
00:25:56,514 --> 00:25:58,716
IMPARTIAL MODEL THAT'S BEEN

692
00:25:58,716 --> 00:26:01,453
TRAINED ON, YOU KNOW, 18,000

693
00:26:01,453 --> 00:26:02,454
STUDIES, IT PROBABLY AVOIDS

694
00:26:02,454 --> 00:26:04,522
HAVING SOME OF THOSE HUMAN

695
00:26:04,522 --> 00:26:06,357
BIASES THAT INDIVIDUAL

696
00:26:06,357 --> 00:26:07,792
SCIENTISTS MIGHT HAVE, BUT OF

697
00:26:07,792 --> 00:26:08,993
COURSE YOU DON'T ENTIRELY AVOID

698
00:26:08,993 --> 00:26:10,528
THE PROBLEM OF BIAS.

699
00:26:10,528 --> 00:26:13,164
SO WITH ALL THESE THINGS, AND

700
00:26:13,164 --> 00:26:13,832
APPROACHES, THERE ARE

701
00:26:13,832 --> 00:26:15,033
FUNDAMENTAL CHALLENGES.

702
00:26:15,033 --> 00:26:17,535
WE'VE HEARD ABOUT HALLUCINATION

703
00:26:17,535 --> 00:26:20,939
IN THESE MODELS.

704
00:26:20,939 --> 00:26:24,442
DEPENDENCE ON LARGE PRE-TRAINING

705
00:26:24,442 --> 00:26:24,876
DATASETS, INTENSIVE

706
00:26:24,876 --> 00:26:25,443
COMPUTATIONAL REQUIREMENTS,

707
00:26:25,443 --> 00:26:28,146
ENERGY USED TO TRAIN LARGE

708
00:26:28,146 --> 00:26:30,915
MODELS IS NOT FUNDAMENTALLY

709
00:26:30,915 --> 00:26:31,216
SUSTAINABLE.

710
00:26:31,216 --> 00:26:33,318
THERE ARE GAPS IN REASONING

711
00:26:33,318 --> 00:26:35,854
CAPABILITIES, AND SO THIS OTHER

712
00:26:35,854 --> 00:26:39,157
BIASES, WHERE IN TRAINING

713
00:26:39,157 --> 00:26:40,525
DATASET BASICALLY MODELS THAT

714
00:26:40,525 --> 00:26:43,928
WILL INHERIT BIASES IN TRAINING

715
00:26:43,928 --> 00:26:45,797
DATA, AND GENERATIVE MODELS CAN

716
00:26:45,797 --> 00:26:50,602
PRODUCE HARMFUL OFFENSIVE OR

717
00:26:50,602 --> 00:26:52,237
MISLEADING OUTPUT ON THE BASIS,

718
00:26:52,237 --> 00:26:56,641
AND USERS HAVE DIFFERENT

719
00:26:56,641 --> 00:26:58,276
ASSUMPTION ABOUT HOW GOOD THEY

720
00:26:58,276 --> 00:26:58,476
ARE.

721
00:26:58,476 --> 00:27:00,678
CERTAIN USERS MAY THINK THESE

722
00:27:00,678 --> 00:27:03,081
ARE KNOWLEDGEABLE AND WILL TRUST

723
00:27:03,081 --> 00:27:05,917
THEM MORE, OTHERS WILL BE

724
00:27:05,917 --> 00:27:06,184
SKEPTICAL.

725
00:27:06,184 --> 00:27:08,419
THIS USER AND MODEL INTERACTION

726
00:27:08,419 --> 00:27:14,125
AS A LARGE EFFECT ON THE UTILITY

727
00:27:14,125 --> 00:27:16,194
OF THESE MODELS IN NEUROSCIENCE.

728
00:27:16,194 --> 00:27:19,264
THIS IS REALLY ALL DATA DRIVEN,

729
00:27:19,264 --> 00:27:20,231
FROM DATA, IN THE BRAIN

730
00:27:20,231 --> 00:27:21,232
INITIATIVE WE HAVE A LOT OF

731
00:27:21,232 --> 00:27:21,466
DATA.

732
00:27:21,466 --> 00:27:23,768
BRAIN INITIATIVE WAS FOUNDED ON

733
00:27:23,768 --> 00:27:24,536
CORE PRINCIPLES ESTABLISHING

734
00:27:24,536 --> 00:27:27,272
PLATFORMS FOR SHARING DATA AND

735
00:27:27,272 --> 00:27:27,939
CONSIDERATION ETHICAL

736
00:27:27,939 --> 00:27:29,440
IMPLICATIONS OF DATA SHARING,

737
00:27:29,440 --> 00:27:31,543
ESPECIALLY IN HUMANS.

738
00:27:31,543 --> 00:27:33,945
SO THIS IS THE CURRENT 2024

739
00:27:33,945 --> 00:27:37,015
STATE OF THE DATA ARCHIVES.

740
00:27:37,015 --> 00:27:40,285
WE HAVE LOTS OF DATA COMING INTO

741
00:27:40,285 --> 00:27:44,989
OUR DATA TYPE SPECIFIC ARCHIVES

742
00:27:44,989 --> 00:27:46,858
THE BRAIN INITIATIVE SUPPORTS IN

743
00:27:46,858 --> 00:27:51,062
THE ECOSYSTEM.

744
00:27:51,062 --> 00:27:53,131
OPEN NEURO AND DANDI, KEY PARTS

745
00:27:53,131 --> 00:27:56,834
OF THE ECOSYSTEM.

746
00:27:56,834 --> 00:27:59,237
LAST YEAR THERE WAS A MULTI-CASE

747
00:27:59,237 --> 00:28:01,439
STUDY PAPER FROM THE BRAIN SHARE

748
00:28:01,439 --> 00:28:03,174
PROJECT THAT TALKED ABOUT THE

749
00:28:03,174 --> 00:28:05,276
BENEFITS OF SHARING HUMAN BRAIN

750
00:28:05,276 --> 00:28:07,145
DATA, SO SOME GREAT INSIGHT IN

751
00:28:07,145 --> 00:28:10,548
THIS PAPER WHICH I WANT TO

752
00:28:10,548 --> 00:28:12,850
HIGHLIGHT. 

753
00:28:12,850 --> 00:28:17,221
THERE'S ANOTHER PAPER FROM

754
00:28:17,221 --> 00:28:19,891
SASKIA HENDRIKS, WHERE THEY LAY

755
00:28:19,891 --> 00:28:24,996
OUT A SPECTRUM OF RISKS AND

756
00:28:24,996 --> 00:28:25,597
IDENTIFIABILITY PROBLEMS WHERE

757
00:28:25,597 --> 00:28:29,834
IF YOU AGGREGATE DATA FROM

758
00:28:29,834 --> 00:28:32,370
DIFFERENT SOURCES THAT ARE

759
00:28:32,370 --> 00:28:35,807
SHARED THERE'S RE-IDENTIFICATION

760
00:28:35,807 --> 00:28:37,809
RISKS OVER TIME AND SCALE,

761
00:28:37,809 --> 00:28:40,545
APPLICABLE WHEN WE'RE THINKING

762
00:28:40,545 --> 00:28:43,214
ABOUT MAKING TRAINING DATASETS

763
00:28:43,214 --> 00:28:45,249
AVAILABLE FOR FUTURE A.I.

764
00:28:45,249 --> 00:28:46,484
MODELS.

765
00:28:46,484 --> 00:28:48,786
AND SO JUST TO REITERATE NIH

766
00:28:48,786 --> 00:28:51,055
DATA MANAGEMENT SHARING POLICY

767
00:28:51,055 --> 00:28:52,724
PUTS FORWARD GUIDELINES AND

768
00:28:52,724 --> 00:28:54,459
PRINCIPLES FOR HUMAN RESEARCH

769
00:28:54,459 --> 00:28:56,127
PARTICIPANT DATA, YOU CAN SEE

770
00:28:56,127 --> 00:28:58,863
THOSE HERE AND GO TO PUBLISHED

771
00:28:58,863 --> 00:29:02,033
NOTICE FOR MORE DETAIL ABOUT

772
00:29:02,033 --> 00:29:02,233
THAT.

773
00:29:02,233 --> 00:29:03,434
BUT THESE PRINCIPLES APPLY

774
00:29:03,434 --> 00:29:04,769
DEEPLY AS WELL.

775
00:29:04,769 --> 00:29:07,271
AS WE GO FORWARD I WANT TO

776
00:29:07,271 --> 00:29:09,240
REITERATE THERE'S MANY SOURCES

777
00:29:09,240 --> 00:29:10,975
OF BIAS, MANY SOURCES OF

778
00:29:10,975 --> 00:29:12,644
PROBLEMS, WHEN WE THINK ABOUT

779
00:29:12,644 --> 00:29:23,154
HOW TO USE AND INTERPRET A.I.

780
00:29:24,856 --> 00:29:26,557
MODELS, IN NEUROSCIENCE IT'S

781
00:29:26,557 --> 00:29:28,426
DIFFICULT TO ATTRIBUTE CAUSE IF

782
00:29:28,426 --> 00:29:31,162
WE'RE THINKING ABOUT CLINICAL

783
00:29:31,162 --> 00:29:33,665
APPLICATIONS, AND DIAGNOSTIC

784
00:29:33,665 --> 00:29:36,834
TESTING, OR OTHER SIMILAR

785
00:29:36,834 --> 00:29:37,068
AVENUES.

786
00:29:37,068 --> 00:29:40,338
AND A LOT OF THIS COMES DOWN TO

787
00:29:40,338 --> 00:29:40,638
PROVENANCE.

788
00:29:40,638 --> 00:29:42,306
WE NEED TO KNOW WHERE THE DATA

789
00:29:42,306 --> 00:29:43,941
COMES FROM AND THAT'S THE ONLY

790
00:29:43,941 --> 00:29:45,376
WAY REALLY THAT YOU CAN KEEP

791
00:29:45,376 --> 00:29:47,245
PEOPLE IN THE LOOP OF DOWNSTREAM

792
00:29:47,245 --> 00:29:50,281
USES OF DATA AND DOWNSTREAM A.I.

793
00:29:50,281 --> 00:29:51,382
MODELS BASED ON DATA.

794
00:29:51,382 --> 00:29:53,351
AND WHEN YOU KEEP PEOPLE IN THE

795
00:29:53,351 --> 00:29:57,288
LOOP THEN YOU HAVE SOME

796
00:29:57,288 --> 00:29:59,824
ASSURANCE THAT YOUR MODELS WILL

797
00:29:59,824 --> 00:30:05,730
AT LEAST BE GROUNDED IN THAT

798
00:30:05,730 --> 00:30:06,164
INTERACTION.

799
00:30:06,164 --> 00:30:08,332
AND SO I ALSO IN ADDITION TO THE

800
00:30:08,332 --> 00:30:10,101
MEETINGS THAT JOHN HIGHLIGHTED I

801
00:30:10,101 --> 00:30:13,171
WANT TO NOTE THAT EARLIER THIS

802
00:30:13,171 --> 00:30:15,473
YEAR IN MARCH THE NATIONAL

803
00:30:15,473 --> 00:30:17,108
ASSOCIATION OF SCIENCE ENGINEERS

804
00:30:17,108 --> 00:30:19,410
HAD A GREAT WORKSHOP ON

805
00:30:19,410 --> 00:30:21,479
NEUROSCIENCE AND A.I., A SESSION

806
00:30:21,479 --> 00:30:23,047
ON REGULATORY POLICY RELATED TO

807
00:30:23,047 --> 00:30:26,117
NEUROSCIENCE AND A.I., WHERE

808
00:30:26,117 --> 00:30:28,286
JOHN AND NITA FARAHANY SPOKE, SO

809
00:30:28,286 --> 00:30:33,658
THIS IS A VERY HIGH LEVEL BUT

810
00:30:33,658 --> 00:30:34,625
GOOD DISCUSSION ABOUT

811
00:30:34,625 --> 00:30:37,929
ESSENTIALLY WHAT ARE THE LATENT

812
00:30:37,929 --> 00:30:41,299
VULNERABILITIES THAT CAN BE

813
00:30:41,299 --> 00:30:44,936
UNCOVERED RELATED TO DATA

814
00:30:44,936 --> 00:30:49,073
PRIVACY AND SECURITY WHILE

815
00:30:49,073 --> 00:30:50,374
MAINTAINING ADVANCES IN

816
00:30:50,374 --> 00:30:52,343
NEUROSCIENCE AND MEDICINE

817
00:30:52,343 --> 00:30:53,478
EMPHASIZING PUBLIC ENGAGEMENT IS

818
00:30:53,478 --> 00:30:55,646
IMPORTANT, AS WELL AS THE

819
00:30:55,646 --> 00:30:58,282
REGULATORY SIDE OF THINGS.

820
00:30:58,282 --> 00:30:59,517
AND UPCOMING AND NEXT MARCH,

821
00:30:59,517 --> 00:31:04,088
SORRY, NEXT APRIL IN MUNICH,

822
00:31:04,088 --> 00:31:06,824
THERE'S THE INS NEUROETHICS 2025

823
00:31:06,824 --> 00:31:08,359
MEETING, ALSO FOCUSED ON THIS

824
00:31:08,359 --> 00:31:17,235
THEME OF BRAIN AND A.I.

825
00:31:17,235 --> 00:31:18,736
THERE'S THIS OTHER SIDE,

826
00:31:18,736 --> 00:31:19,637
INVESTIGATING SCIENTIFIC CORE,

827
00:31:19,637 --> 00:31:22,106
WHAT ARE THE SHARED FUNDAMENTAL

828
00:31:22,106 --> 00:31:22,740
PRINCIPLES BETWEEN APPARENT

829
00:31:22,740 --> 00:31:24,442
INTELLIGENCE OF THESE A.I.

830
00:31:24,442 --> 00:31:27,712
MODELS AND THE INTELLIGENCE THAT

831
00:31:27,712 --> 00:31:29,914
WE KNOW FROM BIOLOGY, HUMANS AND

832
00:31:29,914 --> 00:31:32,316
OTHER ANIMALS, INTERESTING

833
00:31:32,316 --> 00:31:33,584
QUESTIONS AND SCIENTIFIC

834
00:31:33,584 --> 00:31:34,152
PERSPECTIVES EMERGING ALONG

835
00:31:34,152 --> 00:31:34,952
THOSE QUESTIONS.

836
00:31:34,952 --> 00:31:38,656
IN NOVEMBER WE'LL HAVE A TWO-DAY

837
00:31:38,656 --> 00:31:40,858
HYBRID WORKSHOP TO BRING

838
00:31:40,858 --> 00:31:42,627
TOGETHER MANY FIELDS AND

839
00:31:42,627 --> 00:31:45,029
RESEARCHERS ACROSS CAREER LEVELS

840
00:31:45,029 --> 00:31:46,998
TO REALLY DIG INTO AND EXPLORE

841
00:31:46,998 --> 00:31:48,699
THIS KIND OF SCIENTIFIC

842
00:31:48,699 --> 00:31:49,500
INTERSECTION OF NEUROSCIENCE AND

843
00:31:49,500 --> 00:31:52,336
A.I., AND IT SHOULD BE EXCITING.

844
00:31:52,336 --> 00:31:56,207
WE'LL HAVE THE WEBSITE UP SOON.

845
00:31:56,207 --> 00:31:57,308
SIGN UP FOR THE BLOG.

846
00:31:57,308 --> 00:32:00,411
I WANT TO HAND THE MIC OVER TO

847
00:32:00,411 --> 00:32:07,618
DR. DORIS TSAO GIVING US AN

848
00:32:07,618 --> 00:32:15,726
OVERVIEW OF THE TECHNICAL BASIS

849
00:32:15,726 --> 00:32:16,961
OF THESE MODELS. 

850
00:32:16,961 --> 00:32:19,997
>> THANK YOU, JOE.

851
00:32:19,997 --> 00:32:30,374
IT SAYS I CAN'T SHARE.

852
00:32:32,877 --> 00:32:34,212
I'LL TRY AGAIN.

853
00:32:34,212 --> 00:32:34,545
OKAY.

854
00:32:34,545 --> 00:32:36,080
THANK YOU FOR INVITING ME TO

855
00:32:36,080 --> 00:32:37,381
PARTICIPATE IN THIS WORKSHOP.

856
00:32:37,381 --> 00:32:39,083
I'M EXCITED TO LEARN FROM ALL OF

857
00:32:39,083 --> 00:32:45,489
OF YOU TODAY.

858
00:32:45,489 --> 00:32:47,458
I'M A SYSTEMS NEUROSCIENTIST

859
00:32:47,458 --> 00:32:54,298
STUDYING THE PRIMATE SYSTEM,

860
00:32:54,298 --> 00:32:56,033
STEMMING FROM THE REVOLUTIONED

861
00:32:56,033 --> 00:32:58,336
FIELD OF VISUAL NEUROSCIENCE.

862
00:32:58,336 --> 00:33:00,738
YOU ARE RECOGNIZE THESE TWO

863
00:33:00,738 --> 00:33:01,405
GENTLEMEN, CELEBRATING NOBEL

864
00:33:01,405 --> 00:33:03,140
PRIZE FOR DISCOVERIES ABOUT

865
00:33:03,140 --> 00:33:06,677
PRIMARY VISUAL CORTEX INCLUDING

866
00:33:06,677 --> 00:33:07,979
LANDMARK DISCOVERY OF CELLS.

867
00:33:07,979 --> 00:33:09,680
MOST OF YOU DON'T RECOGNIZE WHO

868
00:33:09,680 --> 00:33:11,816
THIS IS.

869
00:33:11,816 --> 00:33:14,018
THIS IS CHARLIE GROSS, RECORDING

870
00:33:14,018 --> 00:33:18,289
ANTERIORLY IN THE VISUAL SYSTEM,

871
00:33:18,289 --> 00:33:19,724
HE DESCRIBES A MARATHON

872
00:33:19,724 --> 00:33:21,459
EXPERIMENT TRYING TO IDENTIFY

873
00:33:21,459 --> 00:33:22,426
THE OPTIMAL STIMULATORY NEURON

874
00:33:22,426 --> 00:33:24,095
DEEP IN THE VISUAL SYSTEM,

875
00:33:24,095 --> 00:33:26,264
WRITES WE COULD NOT FIND A

876
00:33:26,264 --> 00:33:28,566
SIMPLE PHYSICAL DIMENSION THAT

877
00:33:28,566 --> 00:33:30,668
COULD EXPLAIN THE NEURON'S

878
00:33:30,668 --> 00:33:33,371
RESPONSE, BECAUSE THE USES

879
00:33:33,371 --> 00:33:34,038
HIERARCHICAL DISTRIBUTED

880
00:33:34,038 --> 00:33:34,805
REPRESENTATIONS, CARRIED BY

881
00:33:34,805 --> 00:33:35,806
BILLIONS OF NEURONS.

882
00:33:35,806 --> 00:33:38,542
TURNS OUT THE BEST MODEL FOR

883
00:33:38,542 --> 00:33:40,077
THESE REPRESENTATIONS ARE DEEP

884
00:33:40,077 --> 00:33:45,549
NEURAL NETWORKS WHICH ALSO USE

885
00:33:45,549 --> 00:33:46,083
HIERARCHICAL DISTRIBUTIVE

886
00:33:46,083 --> 00:33:47,184
REPUTATIONS, HE DIDN'T HAVE

887
00:33:47,184 --> 00:33:50,021
ACCESS TO THESE NETWORKS

888
00:33:50,021 --> 00:33:50,788
UNFORTUNATELY IN 1972.

889
00:33:50,788 --> 00:33:53,658
I PLAN TO GIVE A BASIC

890
00:33:53,658 --> 00:33:54,959
INTRODUCTION TO THE TECHNICAL

891
00:33:54,959 --> 00:33:55,893
VOCABULARY OF DEEP NETWORKS TO

892
00:33:55,893 --> 00:33:58,763
MAKE SURE WE'RE ON THE SAME

893
00:33:58,763 --> 00:33:59,897
PAGE.

894
00:33:59,897 --> 00:34:02,633
I'LL EXPLAIN IN SOME DETAIL TWO

895
00:34:02,633 --> 00:34:05,569
VERY INFLUENTIAL NEURAL

896
00:34:05,569 --> 00:34:07,772
NETWORKS, CONVOLUTIONAL NEURAL

897
00:34:07,772 --> 00:34:10,708
NETWORK AND TRANSFORMER, AND RUN

898
00:34:10,708 --> 00:34:11,342
THROUGH MAJOR ACHIEVEMENTS, AND

899
00:34:11,342 --> 00:34:12,677
ALL OF THIS SETS US UP TO

900
00:34:12,677 --> 00:34:16,948
DISCUSS THE ROLE OF ETHICS IN

901
00:34:16,948 --> 00:34:18,482
NEURO-A.I. WHICH JOE HAS GIVEN A

902
00:34:18,482 --> 00:34:20,985
GREAT INTRODUCTION TO.

903
00:34:20,985 --> 00:34:28,426
LET'S GET STARTED WITH A DEEP

904
00:34:28,426 --> 00:34:30,861
DIVE INTO THE CONVOLUTIONAL

905
00:34:30,861 --> 00:34:33,898
NEURAL NETWORK. 

906
00:34:33,898 --> 00:34:35,433
SIMPLE CELLS RESPOND ORIENTATION

907
00:34:35,433 --> 00:34:36,968
AND LOCATION, COMPLEX CELLS

908
00:34:36,968 --> 00:34:38,936
RESPOND TO A BAR AT SPECIFIC

909
00:34:38,936 --> 00:34:41,572
ORIENTATION ANYWHERE WITHIN A

910
00:34:41,572 --> 00:34:43,341
RANGE OF LOCATIONS.

911
00:34:43,341 --> 00:34:45,710
PROPOSED COMPLEX CELLS ARE WIRED

912
00:34:45,710 --> 00:34:50,448
BY POOLING TOGETHER SIMPLE

913
00:34:50,448 --> 00:34:51,315
CELLS.

914
00:34:51,315 --> 00:34:53,384
LATER FUKUSHIMA AND LECUN BUILT

915
00:34:53,384 --> 00:35:02,026
A NETWORK WITH MULTIPLE LAYERS,

916
00:35:02,026 --> 00:35:04,128
THE CONVOLUTIONAL NEURAL NETWORK

917
00:35:04,128 --> 00:35:05,529
COULD RECOGNIZE HANDWRITING.

918
00:35:05,529 --> 00:35:07,631
THE CURRENT REVOLUTION IN A.I.

919
00:35:07,631 --> 00:35:17,241
STARTED IN 2012 WITH THIS PAPER,

920
00:35:17,241 --> 00:35:18,442
BASICALLY TOOK LECUN'S NET AND

921
00:35:18,442 --> 00:35:19,443
TRAINED IT.

922
00:35:19,443 --> 00:35:21,078
THEY TRAINED THE NETWORK TO

923
00:35:21,078 --> 00:35:23,414
CLASSIFY THE IMAGES, LABEL IT AS

924
00:35:23,414 --> 00:35:25,883
A DOG, PARROT, ET CETERA.

925
00:35:25,883 --> 00:35:30,154
THE RESULTING NETWORK CALLED

926
00:35:30,154 --> 00:35:31,355
ALEXNET BLUE EXISTING BENCHMARKS

927
00:35:31,355 --> 00:35:33,557
OUT OF THE WATER.

928
00:35:33,557 --> 00:35:35,960
LET'S EXPLORE COMPONENTS OF A

929
00:35:35,960 --> 00:35:36,594
CONVOLUTIONAL NEURAL NETWORK.

930
00:35:36,594 --> 00:35:40,998
GIVEN A SET OF INPUTS THE FIRST

931
00:35:40,998 --> 00:35:45,803
STEP IS CONVOLVE WITH A FILTER,

932
00:35:45,803 --> 00:35:51,942
CONVOLVE IS A WORD AVERAGING,

933
00:35:51,942 --> 00:35:53,811
YOU GET THE OUTPUT AND THESE

934
00:35:53,811 --> 00:35:56,113
WEIGHTS DEFINE THE FILTER.

935
00:35:56,113 --> 00:35:57,281
PERFORM THIS WEIGHTED AVERAGING

936
00:35:57,281 --> 00:35:59,717
LOCALLY ON A SMALL PATCH OF THE

937
00:35:59,717 --> 00:36:01,352
INPUT, SAY 3 X 3 NEIGHBORHOOD.

938
00:36:01,352 --> 00:36:04,622
AND YOU DO THIS REPEATEDLY

939
00:36:04,622 --> 00:36:08,025
ACROSS THE ENTIRE INPUT ARRAY.

940
00:36:08,025 --> 00:36:12,063
THE VALUE OF FILTER WEIGHTS ARE

941
00:36:12,063 --> 00:36:14,498
PARAMETERS OF THE CONV-NET.

942
00:36:14,498 --> 00:36:15,900
I'LL SHOW YOU HOW THEY CAN BE

943
00:36:15,900 --> 00:36:17,468
LEARNED FROM DATA.

944
00:36:17,468 --> 00:36:19,737
HERE IS A CONCRETE EXAMPLE OF

945
00:36:19,737 --> 00:36:21,272
THIS COMPUTATION WHERE THIS

946
00:36:21,272 --> 00:36:23,908
INPUT IMAGE IS CONVOLVED WITH

947
00:36:23,908 --> 00:36:26,644
THIS FILTER TO GIVE ARRAY OF

948
00:36:26,644 --> 00:36:26,877
OUTPUTS.

949
00:36:26,877 --> 00:36:28,379
YOU CAN SEE ACTUAL FILTERS

950
00:36:28,379 --> 00:36:32,116
LEARNED BY THE FIRST LAYER OF

951
00:36:32,116 --> 00:36:34,085
ALEX-NET, 96 FILTERS, PLOTTED IN

952
00:36:34,085 --> 00:36:37,288
EACH OF THE POSTAGE STAMP ARE

953
00:36:37,288 --> 00:36:38,689
FILTERED WEIGHTS.

954
00:36:38,689 --> 00:36:42,760
TO BUILD A DEEP CONVOLUTIONAL

955
00:36:42,760 --> 00:36:45,262
NEURAL NETWORK YOU TAKE THE

956
00:36:45,262 --> 00:36:46,897
OUTPUT AND PASS IT THROUGH A

957
00:36:46,897 --> 00:36:49,400
GATING FUNCTION, S HERE.

958
00:36:49,400 --> 00:36:50,835
YOU REPEAT THIS PROCESS WITH AS

959
00:36:50,835 --> 00:36:55,339
MANY LAYERS AS YOU LIKE.

960
00:36:55,339 --> 00:36:58,843
SIMPLEST FORM OF GATING FUNCTION

961
00:36:58,843 --> 00:37:01,045
IS A RECTIFIED LINEAR USED,

962
00:37:01,045 --> 00:37:03,247
OUTPUT ZERO IF INPUT IS

963
00:37:03,247 --> 00:37:08,919
NEGATIVE, OTHERWISE PASSES ON

964
00:37:08,919 --> 00:37:09,353
INPUT.

965
00:37:09,353 --> 00:37:10,921
THIS NON-LINEARITY, NEURON

966
00:37:10,921 --> 00:37:12,123
DOESN'T FIRE UNLESS MEMBRANE

967
00:37:12,123 --> 00:37:14,525
POTENTIAL GOES ABOVE A CERTAIN

968
00:37:14,525 --> 00:37:14,959
THRESHOLD.

969
00:37:14,959 --> 00:37:17,128
WHY IS NON-LINEAR GATING

970
00:37:17,128 --> 00:37:17,495
CRITICAL?

971
00:37:17,495 --> 00:37:24,535
IN TERMS OF SIMPLE AND COMPLEX

972
00:37:24,535 --> 00:37:25,936
CELLS, BUILD A COMPLEX CELL BY

973
00:37:25,936 --> 00:37:28,339
ADDING UP A BUNCH OF SIMPLE

974
00:37:28,339 --> 00:37:30,074
CELLS PREFERRING THE SAME

975
00:37:30,074 --> 00:37:31,075
ORIENTATION, VERTICAL.

976
00:37:31,075 --> 00:37:33,377
YOU CAN ASK WHY WOULDN'T THIS

977
00:37:33,377 --> 00:37:34,245
COMPLEX CELL RESPOND TO

978
00:37:34,245 --> 00:37:35,112
HORIZONTAL BAR?

979
00:37:35,112 --> 00:37:37,081
THE REASON IS THAT EACH SIMPLE

980
00:37:37,081 --> 00:37:40,317
CELL BY VIRTUE OF BEING A NEURON

981
00:37:40,317 --> 00:37:42,386
WITH THRESHOLD FOR FIRING HAS

982
00:37:42,386 --> 00:37:44,255
THIS NON-LINEAR GATING BUILT IN,

983
00:37:44,255 --> 00:37:46,724
IF YOU PRESENT A HORIZONTAL BAR

984
00:37:46,724 --> 00:37:48,859
ALL THREE INPUTS OF THE COMPLEX

985
00:37:48,859 --> 00:37:50,594
CELL STAY SILENTS, EACH BELOW

986
00:37:50,594 --> 00:37:50,861
THRESHOLD.

987
00:37:50,861 --> 00:37:52,897
IF YOU PRESENT A VERTICAL BAR,

988
00:37:52,897 --> 00:37:55,299
THEN YOU'LL GET A STRONG

989
00:37:55,299 --> 00:37:56,167
RESPONSE.

990
00:37:56,167 --> 00:37:57,168
SO THE NON-LINEAR GATING IS

991
00:37:57,168 --> 00:37:58,402
CRUCIAL HERE.

992
00:37:58,402 --> 00:38:02,973
IT'S THE MAGIC INGREDIENT IN A

993
00:38:02,973 --> 00:38:06,710
ENABLES HIERARCHICAL COMBINATION

994
00:38:06,710 --> 00:38:09,113
TO CREATE COMPLEX FEATURES.

995
00:38:09,113 --> 00:38:19,657
AND AS A SIDE, MORE ABSTRACTLY.

996
00:38:20,024 --> 00:38:22,560
TO BUILT A DEEP CONVOLUTIONAL

997
00:38:22,560 --> 00:38:25,296
NETWORK YOU STRING BLOCKS

998
00:38:25,296 --> 00:38:27,164
TOGETHER, KEY COMPUTATIONS

999
00:38:27,164 --> 00:38:31,435
WITHIN EACH BLOCK ARE SPATIAL

1000
00:38:31,435 --> 00:38:33,504
FILTERING AND NON-LINEAR GATING.

1001
00:38:33,504 --> 00:38:35,806
FINAL OUTPUT IS END UNITS,

1002
00:38:35,806 --> 00:38:39,510
FIRING TELLS YOU PROBABILITY OF

1003
00:38:39,510 --> 00:38:40,511
INPUT IN IMAGE CLASSES.

1004
00:38:40,511 --> 00:38:41,946
ONE OF THE MOST IMPORTANT

1005
00:38:41,946 --> 00:38:43,514
FEATURES OF DEEP NEURAL NETWORKS

1006
00:38:43,514 --> 00:38:46,183
IS THE WEIGHTS OF THE NETWORK

1007
00:38:46,183 --> 00:38:48,485
ARE LEARNED FROM DATA.

1008
00:38:48,485 --> 00:38:51,021
AND THE IDEA IS TO UPDATE ALL

1009
00:38:51,021 --> 00:38:52,856
THE WEIGHTS TO MINIMIZE THE

1010
00:38:52,856 --> 00:38:55,159
ERROR ON A TRAINING SET.

1011
00:38:55,159 --> 00:38:58,796
IF WE KNOW THIS IS A GRAY PARROT

1012
00:38:58,796 --> 00:39:00,431
THIS OUTPUT WOULD HAVE A HIGH

1013
00:39:00,431 --> 00:39:01,532
ERROR, THIS WOULD HAVE A LOW

1014
00:39:01,532 --> 00:39:01,732
ERROR.

1015
00:39:01,732 --> 00:39:04,168
AND IN ORDER TO FIND THE SET OF

1016
00:39:04,168 --> 00:39:05,703
NETWORK WEIGHTS THAT MINIMIZE

1017
00:39:05,703 --> 00:39:11,275
THE ERROR PERFORM A PROCESS

1018
00:39:11,275 --> 00:39:12,343
CALLED STOCHASTIC GRADIENT

1019
00:39:12,343 --> 00:39:12,576
DESCENT.

1020
00:39:12,576 --> 00:39:13,777
COMPUTE THE ERROR, COMPUTE HOW

1021
00:39:13,777 --> 00:39:15,512
TO CHANGE THE WEIGHTS TO

1022
00:39:15,512 --> 00:39:19,883
DECREASE THE ERROR.

1023
00:39:19,883 --> 00:39:20,884
USING ALGORITHM THAT COMPUTES

1024
00:39:20,884 --> 00:39:22,853
GRADIENT OF ERROR WITH RESPECT

1025
00:39:22,853 --> 00:39:24,488
TO EACH WEIGHT.

1026
00:39:24,488 --> 00:39:26,991
BY FOLLOWING THE GRADIENT YOU

1027
00:39:26,991 --> 00:39:29,627
CAN MINIMIZE THE ERROR.

1028
00:39:29,627 --> 00:39:36,867
NOW WE UNDERSTAND HOW TO BUILD

1029
00:39:36,867 --> 00:39:40,371
AND TRAIN CONV NETS, PIONEERING

1030
00:39:40,371 --> 00:39:42,573
PAPER NEURONS OF DIFFERENT

1031
00:39:42,573 --> 00:39:43,774
LAYERS OF THE MACAQUE HIERARCHY

1032
00:39:43,774 --> 00:39:50,881
COULD BE WELL MODELS BY

1033
00:39:50,881 --> 00:39:58,322
DIFFERENT LAYERS OF CONV NET.

1034
00:39:58,322 --> 00:39:59,323
MACAQUE CORTEX HARBORS SEVERAL

1035
00:39:59,323 --> 00:40:00,858
LAYERS OF THE MAP, EXPLAINING

1036
00:40:00,858 --> 00:40:03,694
THE SPATIAL ORGANIZATION OF

1037
00:40:03,694 --> 00:40:05,429
VISUAL CORTEX USING CONV NET.

1038
00:40:05,429 --> 00:40:08,165
I'D LIKE TO TALK ABOUT

1039
00:40:08,165 --> 00:40:09,400
TRANSFORMERS WHICH REFER TO

1040
00:40:09,400 --> 00:40:11,568
ANOTHER TYPE OF NEURAL NETWORK

1041
00:40:11,568 --> 00:40:12,236
ARCHITECTURE.

1042
00:40:12,236 --> 00:40:15,172
A LOT OF SAME DEVELOPMENTS IN

1043
00:40:15,172 --> 00:40:21,211
A.I. ARE BASED ON TRANSFORMER

1044
00:40:21,211 --> 00:40:22,079
ARCHITECTURE LIKE ChatGPT.

1045
00:40:22,079 --> 00:40:24,615
IT'S WORKS FROM TEXT OR IMAGES

1046
00:40:24,615 --> 00:40:26,583
PATCHES, PASSES THEM THROUGH A

1047
00:40:26,583 --> 00:40:29,186
SERIES OF STAGES TO PRODUCE A

1048
00:40:29,186 --> 00:40:31,488
FINAL OUTPUT, FOR EXAMPLE

1049
00:40:31,488 --> 00:40:32,923
PREDICTIVE NEXT WORK.

1050
00:40:32,923 --> 00:40:37,294
HERE'S THE STRUCTURE OF A BASIC

1051
00:40:37,294 --> 00:40:38,829
TRANSFORMER BLOCK, N OF THESE

1052
00:40:38,829 --> 00:40:41,665
CHAINED TOGETHER, KEY MAGIC IN

1053
00:40:41,665 --> 00:40:42,866
MULTI-HEAD ATTENTION LAYER.

1054
00:40:42,866 --> 00:40:45,602
SO HERE IS HOW ATTENTION WORKS

1055
00:40:45,602 --> 00:40:47,338
IN A TRANSFORMER.

1056
00:40:47,338 --> 00:40:50,307
THINK OF ATTENTION AS ANALOGOUS

1057
00:40:50,307 --> 00:40:54,244
TO CONVOLUTION FILTER IN A CONV

1058
00:40:54,244 --> 00:40:58,215
NET, INSTEAD OF A LOCAL FILTER

1059
00:40:58,215 --> 00:41:01,385
IT COMPUTES BETWEEN ALL OF THE

1060
00:41:01,385 --> 00:41:02,486
INPUT TOKENS.

1061
00:41:02,486 --> 00:41:06,090
THIS CORRELATION IS COMPUTED

1062
00:41:06,090 --> 00:41:12,663
USING THIS EQUATION HERE WHERE

1063
00:41:12,663 --> 00:41:16,166
Q, K, AND V DENOTE PROJECTIONS

1064
00:41:16,166 --> 00:41:17,801
OF THE TOKEN STRING, YOU CAN

1065
00:41:17,801 --> 00:41:21,638
STACK COPIES OF THE BASIC MODULE

1066
00:41:21,638 --> 00:41:23,707
TO IMPLEMENT MULTI-HEAD

1067
00:41:23,707 --> 00:41:24,608
ATTENTION WHERE WEIGHTS AND

1068
00:41:24,608 --> 00:41:31,648
HEADS ARE  INDEPENDENT.

1069
00:41:31,648 --> 00:41:33,717
INTUITIVE WAY OF UNDERSTANDING,

1070
00:41:33,717 --> 00:41:40,290
THINK OF SENTENCE STRING. 

1071
00:41:40,290 --> 00:41:41,258
COMPUTES CORRELATION BETWEEN

1072
00:41:41,258 --> 00:41:43,560
WORDS, FOR EXAMPLE WEARING TO

1073
00:41:43,560 --> 00:41:45,863
JUMPSUIT HIGHLY CORRELATED.

1074
00:41:45,863 --> 00:41:50,567
THAT'S A QUICK TOUR THROUGH THE

1075
00:41:50,567 --> 00:41:50,901
TRANSFORMER.

1076
00:41:50,901 --> 00:41:53,303
NOW PRE-TRAINING AND FINE TUNING

1077
00:41:53,303 --> 00:41:55,038
WHICH JOE MENTIONED HAS A LOT OF

1078
00:41:55,038 --> 00:41:55,806
ETHICAL ISSUES INVOLVED.

1079
00:41:55,806 --> 00:42:00,477
THIS REFERS TO THE IDEA YOU CAN

1080
00:42:00,477 --> 00:42:03,781
TRAIN A NETWORK TO DO ONE TASK,

1081
00:42:03,781 --> 00:42:09,720
RETRAIN TO DO OTHER TASKS.

1082
00:42:09,720 --> 00:42:13,424
ALEX-NET WAS TRAINED ON OBJECT

1083
00:42:13,424 --> 00:42:15,726
CLASSIFICATION, CAN YOU TRAIN IT

1084
00:42:15,726 --> 00:42:17,461
ON DOWNSTREAM TASKS, YOU CAN

1085
00:42:17,461 --> 00:42:18,996
RETRAIN BY SIMPLY HOLDING ALL

1086
00:42:18,996 --> 00:42:20,864
THE WEIGHTS FIXED, EXCEPT FOR

1087
00:42:20,864 --> 00:42:23,434
THE FINAL LAYER, AND ALSO DO

1088
00:42:23,434 --> 00:42:25,502
MORE COMPLEX FINE TUNING BY

1089
00:42:25,502 --> 00:42:27,137
INSERTING NEW ADAPTER LAYERS

1090
00:42:27,137 --> 00:42:28,105
INTO EACH TRANSFORMER BLOCK

1091
00:42:28,105 --> 00:42:31,708
WHILE KEEPING THE WEIGHTS ON THE

1092
00:42:31,708 --> 00:42:33,811
OTHER LAYERS FIXED.

1093
00:42:33,811 --> 00:42:41,385
LET'S TALK BRIEFLY ABOUT MAJOR

1094
00:42:41,385 --> 00:42:42,986
ACHIEVEMENTS WHICH PATRICK WILL

1095
00:42:42,986 --> 00:42:43,687
DISCUSS.

1096
00:42:43,687 --> 00:42:49,460
THE SUCCESS OF DEEP LEARNING HAS

1097
00:42:49,460 --> 00:42:52,329
CAPTURED EVERYONE'S IMAGINATION.

1098
00:42:52,329 --> 00:42:55,599
ALEX'S GROUP AT U.T. AUSTIN

1099
00:42:55,599 --> 00:43:01,738
PUSHED FOR THE STATE OF THE

1100
00:43:01,738 --> 00:43:07,444
LARGE IN LANGUAGE DECODING,

1101
00:43:07,444 --> 00:43:11,048
DECODING FROM BRAIN ACTIVITY.

1102
00:43:11,048 --> 00:43:14,751
THE STORY WAS I GOT UP FROM THE

1103
00:43:14,751 --> 00:43:18,255
AIR MATTRESS AND PRESSED MY FACE

1104
00:43:18,255 --> 00:43:20,224
TO THE GLASS WINDOW, EXPECTING

1105
00:43:20,224 --> 00:43:22,259
EYES BUT SEEING DARKNESS.

1106
00:43:22,259 --> 00:43:24,828
DECODED, I WALKED TO THE WINDOW

1107
00:43:24,828 --> 00:43:26,597
OPENED THE GLASS, PEERED OUT,

1108
00:43:26,597 --> 00:43:29,299
DIDN'T SEE ANYTHING, LOOKED UP.

1109
00:43:29,299 --> 00:43:30,934
AGAIN I SAW NOTHING.

1110
00:43:30,934 --> 00:43:33,470
THIS IS REMARKABLE FROM HUMAN

1111
00:43:33,470 --> 00:43:35,005
fMRI DATA.

1112
00:43:35,005 --> 00:43:37,307
THE WAY THIS WORKS, TAKE YOUR

1113
00:43:37,307 --> 00:43:40,043
fMRI DATA, AND YOU LEARN THE

1114
00:43:40,043 --> 00:43:42,579
FEATURE TUNING OF EACH VOXEL.

1115
00:43:42,579 --> 00:43:45,749
AND THEN YOU USE YOUR LANGUAGE

1116
00:43:45,749 --> 00:43:46,583
MODEL, PRE-TRAINED LANGUAGE

1117
00:43:46,583 --> 00:43:48,485
MODEL, TO PREDICT A SET OF THE

1118
00:43:48,485 --> 00:43:51,221
MOST LIKELY POSSIBILITIES FOR

1119
00:43:51,221 --> 00:43:54,491
THE NEXT WORD. 

1120
00:43:54,491 --> 00:44:00,531
YOU FEED POSITION BUILTS TO

1121
00:44:00,531 --> 00:44:02,599
PREDICT HOW THE BRAIN WOULD

1122
00:44:02,599 --> 00:44:02,833
RESPOND.

1123
00:44:02,833 --> 00:44:08,639
BY COMPARING YOU THEN DECODE THE

1124
00:44:08,639 --> 00:44:10,507
NEXT WORD.

1125
00:44:10,507 --> 00:44:14,878
SIMILARLY A GENERATIVE IMAGE

1126
00:44:14,878 --> 00:44:17,714
MODEL STABLE DIFFUSION DECODED

1127
00:44:17,714 --> 00:44:20,117
VISUAL IMAGES FROM fMRI DATA

1128
00:44:20,117 --> 00:44:21,318
WITH IMPRESSIVE ACCURACY.

1129
00:44:21,318 --> 00:44:22,619
WE'RE APPROACHING THE ABILITY TO

1130
00:44:22,619 --> 00:44:24,054
READ PEOPLE'S MINDS IN REAL

1131
00:44:24,054 --> 00:44:25,489
TIME, WE'RE NOT THERE YET.

1132
00:44:25,489 --> 00:44:29,626
THIS RAISES ALL KINDS OF ETHICAL

1133
00:44:29,626 --> 00:44:30,093
ISSUES.

1134
00:44:30,093 --> 00:44:32,596
NOW I WANT TO WALK THROUGH ONE

1135
00:44:32,596 --> 00:44:37,267
FRAMEWORK FOR THINKING  ABOUT

1136
00:44:37,267 --> 00:44:41,004
ETHICS AND NEUROA.I. PROPOSED IN

1137
00:44:41,004 --> 00:44:43,106
"NATURE" IN 2017.

1138
00:44:43,106 --> 00:44:47,477
FIRST PILLAR IS IMPORTANCE OF

1139
00:44:47,477 --> 00:44:48,245
PROTECTING PRIVACY AND ENSURING

1140
00:44:48,245 --> 00:44:49,780
CONSENT OF THE BRAIN INFORMATION

1141
00:44:49,780 --> 00:44:54,284
IS THE MOST INTIMATE AND PRIVATE

1142
00:44:54,284 --> 00:44:56,253
OF ALL INFORMATION.

1143
00:44:56,253 --> 00:44:59,289
USERS SHOULD BE ABLE TO OPT OUT.

1144
00:44:59,289 --> 00:45:00,290
WHEN DRIVING FACTORS IS

1145
00:45:00,290 --> 00:45:05,362
INCREASING SCALE OF TRAINING

1146
00:45:05,362 --> 00:45:06,930
DATA, DATA GATHERED FROM SOCIAL

1147
00:45:06,930 --> 00:45:08,599
MEDIA, ENORMOUS AMOUNT OF DATA,

1148
00:45:08,599 --> 00:45:10,667
DIFFICULT OR IMPOSSIBLE TO KEEP

1149
00:45:10,667 --> 00:45:12,202
TRACK OF POTENTIAL PRIVACY

1150
00:45:12,202 --> 00:45:13,103
ISSUES.

1151
00:45:13,103 --> 00:45:16,707
ISSUES BECOME MORE TRICKLY WHEN

1152
00:45:16,707 --> 00:45:17,808
CONSIDERING APPLICATIONS TO

1153
00:45:17,808 --> 00:45:19,876
MEDICINE WITH PRIVATE HEALTH

1154
00:45:19,876 --> 00:45:20,644
INFORMATION.

1155
00:45:20,644 --> 00:45:23,580
THERE SHOULD BE A CONSENT

1156
00:45:23,580 --> 00:45:25,015
PROCEDURE TO SPECIFY WHO WILL

1157
00:45:25,015 --> 00:45:27,084
USE THE DATA, WHAT PURPOSES, HOW

1158
00:45:27,084 --> 00:45:28,518
LONG, REGULATIONS TO LIMIT

1159
00:45:28,518 --> 00:45:30,354
POSSIBILITY OF PEOPLE GIVING UP

1160
00:45:30,354 --> 00:45:32,456
DATA OR HAVING DATA WRITTEN TO

1161
00:45:32,456 --> 00:45:34,224
THE BRAIN FOR FINANCIAL

1162
00:45:34,224 --> 00:45:37,294
COMPENSATION, AND FINALLY SHOULD

1163
00:45:37,294 --> 00:45:40,364
ADOPT TECHNOLOGY TO ENSURE

1164
00:45:40,364 --> 00:45:42,399
PRIVACY AND ONE APPROACH IS

1165
00:45:42,399 --> 00:45:43,200
FEDERATED LEARNING.

1166
00:45:43,200 --> 00:45:44,835
INSTEAD OF GATHERING ALL OF THE

1167
00:45:44,835 --> 00:45:48,105
DATA IN ONE PLACE FOR MODEL

1168
00:45:48,105 --> 00:45:49,306
TRAINING, USE DECENTRALIZED

1169
00:45:49,306 --> 00:45:50,307
APPROACH, EACH PARTY TRAINS A

1170
00:45:50,307 --> 00:45:52,442
COMMON MODEL OF THEIR OWN

1171
00:45:52,442 --> 00:45:53,977
PRIVATE DATA, AND THIS IS USED

1172
00:45:53,977 --> 00:45:56,847
TO COMPUTE PART OF THE WEIGHT

1173
00:45:56,847 --> 00:45:57,180
GRADIENT.

1174
00:45:57,180 --> 00:45:58,415
ONLY THIS WEIGHT GRADIENT IS

1175
00:45:58,415 --> 00:45:59,716
SENT TO THE CENTRAL LOCATION

1176
00:45:59,716 --> 00:46:02,019
WHERE THEY ARE ALL AGGREGATED

1177
00:46:02,019 --> 00:46:05,055
AND BROADCAST BACK TO EVERYONE

1178
00:46:05,055 --> 00:46:07,157
SO EACH SITE CAN PERFORM THE

1179
00:46:07,157 --> 00:46:12,929
NEXT ROUND OF TRAINING IN

1180
00:46:12,929 --> 00:46:13,163
PRIVACY.

1181
00:46:13,163 --> 00:46:14,931
SECOND IS PROTECTION OF AGENCY

1182
00:46:14,931 --> 00:46:21,038
AND  IDENTITY.

1183
00:46:21,038 --> 00:46:24,207
ABILITY TO CHOOSE ACTION AND

1184
00:46:24,207 --> 00:46:26,843
BRAIN MACHINE INTERACTION CAN

1185
00:46:26,843 --> 00:46:27,611
BLUR THE LINE.

1186
00:46:27,611 --> 00:46:31,014
A MAN TREATED FOR DEPRESSION

1187
00:46:31,014 --> 00:46:32,649
WITH BRAIN STIMULATORS BEGAN TO

1188
00:46:32,649 --> 00:46:36,987
WONDER ACTIONS WITH DUE TO

1189
00:46:36,987 --> 00:46:38,889
HIMSELF, HE SAID I'M NOT SURE

1190
00:46:38,889 --> 00:46:43,260
WHO I AM, RAISING FUNDAMENTAL

1191
00:46:43,260 --> 00:46:51,468
QUESTIONS. 

1192
00:46:51,468 --> 00:46:53,003
WHAT MAKES US RESPONSIBLE?

1193
00:46:53,003 --> 00:46:54,471
AND IN CONTEXT OF MIND READING

1194
00:46:54,471 --> 00:46:57,574
TECHNOLOGY I TOLD YOU ABOUT

1195
00:46:57,574 --> 00:46:58,909
EARLIER, SUPPOSE ALEX HUTH USED

1196
00:46:58,909 --> 00:47:02,713
THE DEVICE TO READ THOUGHTS OF

1197
00:47:02,713 --> 00:47:04,247
APHASIC PERSON, WOULD THAT

1198
00:47:04,247 --> 00:47:06,650
PERSON BE RESPONSIBLE FOR THOSE

1199
00:47:06,650 --> 00:47:08,418
THOUGHTS AND ACTIONS OR ASCRIBE

1200
00:47:08,418 --> 00:47:10,721
PART OF THE AGENCY TO THE

1201
00:47:10,721 --> 00:47:11,788
LANGUAGE MODEL THAT'S GENERATING

1202
00:47:11,788 --> 00:47:15,959
THE  PREDICTIONS OF THE NEXT

1203
00:47:15,959 --> 00:47:16,326
WORK?

1204
00:47:16,326 --> 00:47:19,162
THIRD ETHICAL CONCERN IS ALL

1205
00:47:19,162 --> 00:47:20,664
THESE NEUROA.I. TECHNOLOGIES

1206
00:47:20,664 --> 00:47:22,232
CREATE POSSIBILITY FOR VASTLY

1207
00:47:22,232 --> 00:47:25,736
ENHANCING HUMAN CAPABILITY, FOR

1208
00:47:25,736 --> 00:47:26,803
EXAMPLE CREATING SUPER

1209
00:47:26,803 --> 00:47:28,038
INTELLIGENCE, I'LL PROPOSE THAT

1210
00:47:28,038 --> 00:47:29,673
GUIDELINES SHOULD BE ESTABLISHED

1211
00:47:29,673 --> 00:47:34,711
TO SET LIMITS ON AUGMENTING

1212
00:47:34,711 --> 00:47:35,679
TECHNOLOGY ANALOGOUS TO THOSE

1213
00:47:35,679 --> 00:47:37,013
FOR CRISPR.

1214
00:47:37,013 --> 00:47:39,483
A.I. SYSTEMS CAN PRODUCE BIASED

1215
00:47:39,483 --> 00:47:42,919
RESULTS THAT REFLECT AND

1216
00:47:42,919 --> 00:47:44,855
PERPETUATE HUMAN BIAS IN

1217
00:47:44,855 --> 00:47:47,724
SOCIETY, FOR EXAMPLE FACE

1218
00:47:47,724 --> 00:47:49,059
RECOGNITION ALGORITHMS TRAINED

1219
00:47:49,059 --> 00:47:50,894
ON WHITE PEOPLE, AND ONE

1220
00:47:50,894 --> 00:47:53,530
APPROACH TO ACTIVELY REMOVE BIAS

1221
00:47:53,530 --> 00:48:03,907
HERE, ONE OF MANY, IS

1222
00:48:04,641 --> 00:48:06,543
ADVERSARIAL DEVISING, TACK ON A

1223
00:48:06,543 --> 00:48:08,612
SECOND NETWORK THAT ATTEMPTS TO

1224
00:48:08,612 --> 00:48:10,280
PREDICT GENDER FROM OUTPUT OF

1225
00:48:10,280 --> 00:48:11,481
THE FIRST NETWORK, AND YOU CAN

1226
00:48:11,481 --> 00:48:13,617
TRAIN YOUR MODEL TO MAKE THE

1227
00:48:13,617 --> 00:48:19,456
CORRECT PREDICTION WHILE  SAME

1228
00:48:19,456 --> 00:48:23,293
TIME PREVENTING TH ADVERSARY

1229
00:48:23,293 --> 00:48:28,331
WITH THE BIASED FACTORS.

1230
00:48:28,331 --> 00:48:29,533
DEEP NEURAL NETWORKS CAPTURE

1231
00:48:29,533 --> 00:48:34,237
COMPLEX FUNCTIONS FROM INPUT TO

1232
00:48:34,237 --> 00:48:35,105
OUTPUT.

1233
00:48:35,105 --> 00:48:36,106
TWO INFLUENTIAL ARCHITECTURES

1234
00:48:36,106 --> 00:48:39,176
ARE CONV NET AND TRANSFORMER,

1235
00:48:39,176 --> 00:48:40,243
PRE-TRAINED FOR ONE APPLICATION,

1236
00:48:40,243 --> 00:48:43,847
FINE TUNED FOR OTHERS, DEEP

1237
00:48:43,847 --> 00:48:45,282
LEARNING ENABLED MAJOR ADVANCES

1238
00:48:45,282 --> 00:48:48,351
IN FUNCTION AND DECODING, AND

1239
00:48:48,351 --> 00:48:49,886
ETHICAL APPROACH TO DEVELOPMENT

1240
00:48:49,886 --> 00:48:52,289
SHOULD CONSIDER PRIVACY AND

1241
00:48:52,289 --> 00:48:59,963
CONSENT, AGENCY AND IDENTITY,

1242
00:48:59,963 --> 00:49:03,066
AUGMENTATION, AND BIAS.

1243
00:49:03,066 --> 00:49:05,035
NOW I'LL INTRODUCE THE NEXT

1244
00:49:05,035 --> 00:49:09,940
SPEAKER, AN A.I. RESEARCHERS,

1245
00:49:09,940 --> 00:49:12,909
SENIOR SCIENTIST AT A

1246
00:49:12,909 --> 00:49:14,444
BRIAN-COMPUTER INTERFACE AT

1247
00:49:14,444 --> 00:49:17,180
META, AND GOOGLE, FOUNDING CTO

1248
00:49:17,180 --> 00:49:20,684
OF NEUROMATCH.

1249
00:49:20,684 --> 00:49:21,184
TAKE IT AWAY, PATRICK.

1250
00:49:21,184 --> 00:49:23,820
>> THANKS SO MUCH FOR THIS

1251
00:49:23,820 --> 00:49:24,487
WONDERFUL INTRODUCTION.

1252
00:49:24,487 --> 00:49:32,495
AND IT'S GREAT TO BE HERE. 

1253
00:49:32,495 --> 00:49:34,898
I'LL SHARE MY SCREEN.

1254
00:49:34,898 --> 00:49:37,067
HOPE YOU CAN SEE THIS.

1255
00:49:37,067 --> 00:49:40,604
I'M GOING TO BE TALKING ABOUT

1256
00:49:40,604 --> 00:49:43,540
THE MORE TECHNICAL ASPECTS OF

1257
00:49:43,540 --> 00:49:44,641
FOUNDATION MODELS FOR

1258
00:49:44,641 --> 00:49:45,408
NEUROSCIENCE.

1259
00:49:45,408 --> 00:49:47,244
SO I'LL START WITH FOUNDATION

1260
00:49:47,244 --> 00:49:48,678
MODELS IN GENERAL.

1261
00:49:48,678 --> 00:49:50,747
I'LL INTRODUCE A RUBRIC FOR

1262
00:49:50,747 --> 00:49:52,282
FOUNDATION MODELS, HOW THEY

1263
00:49:52,282 --> 00:49:54,251
DIFFER FROM CONVENTIONAL MACHINE

1264
00:49:54,251 --> 00:49:57,654
LEARNING MODELS, I'LL GO TO

1265
00:49:57,654 --> 00:50:02,592
TALKING ABOUT FOUNDATION MODELS

1266
00:50:02,592 --> 00:50:04,761
IN NEUROSCIENCE, BRING TAXONOMY

1267
00:50:04,761 --> 00:50:06,630
AND SHOWCASE EXAMPLES AND TALK

1268
00:50:06,630 --> 00:50:08,932
ABOUT THE OPPORTUNITIES AND

1269
00:50:08,932 --> 00:50:11,001
PITFALLS FROM THESE MODELS

1270
00:50:11,001 --> 00:50:12,769
INCLUDING CLASSICAL, ETHICAL

1271
00:50:12,769 --> 00:50:16,339
CONCERNING FOR MACHINE LEARNING,

1272
00:50:16,339 --> 00:50:17,540
AND NEUROSCIENCE, ADDITIONAL

1273
00:50:17,540 --> 00:50:18,808
CONSIDERATIONS REALLY SPECIFIC

1274
00:50:18,808 --> 00:50:21,912
TO FOUNDATION MODELS IN

1275
00:50:21,912 --> 00:50:22,212
PARTICULAR.

1276
00:50:22,212 --> 00:50:23,880
SO I DON'T NEED TO BELABOR THE

1277
00:50:23,880 --> 00:50:26,850
POINT, WE'RE ALL VERY IMPRESSED

1278
00:50:26,850 --> 00:50:31,988
BY THE PROGRESS OF A.I.

1279
00:50:31,988 --> 00:50:33,523
SEVERAL OF US USE FOUNDATION

1280
00:50:33,523 --> 00:50:38,561
MODELS EVERY DAY.

1281
00:50:38,561 --> 00:50:42,499
ON THE LEFT CLAUDE, ChatGPT,

1282
00:50:42,499 --> 00:50:45,335
ALPHA FOLD DEPICTED HERE, WE

1283
00:50:45,335 --> 00:50:49,172
HAVE SINGLE CELL RNAseq

1284
00:50:49,172 --> 00:50:50,373
FOUNDATION MODELS, AND RECENTLY

1285
00:50:50,373 --> 00:50:53,310
FOUNDATION MODELS HAVE BEEN

1286
00:50:53,310 --> 00:50:57,380
EXPANDED TO INCLUDE MODELS

1287
00:50:57,380 --> 00:50:57,681
FOREVIDEOS.

1288
00:50:57,681 --> 00:50:59,582
SORRY TO SAY THE PUPPIES PLAYING

1289
00:50:59,582 --> 00:51:02,953
IN THE SNOW DO NOT EXIST.

1290
00:51:02,953 --> 00:51:05,121
NOW, WHAT CONSTITUTES A

1291
00:51:05,121 --> 00:51:05,889
FOUNDATION MODEL?

1292
00:51:05,889 --> 00:51:09,826
ONE OF THE ELEMENTS IS THAT THEY

1293
00:51:09,826 --> 00:51:11,895
ARE PRE-TRAINED IN A SUPERVISED

1294
00:51:11,895 --> 00:51:15,532
WAY WHICH DORIS ALLUDED TO.

1295
00:51:15,532 --> 00:51:18,601
FINE TUNED ON DOWNSTREAM TASKS.

1296
00:51:18,601 --> 00:51:21,204
TRAINED ON LARGE SCALE DATA,

1297
00:51:21,204 --> 00:51:22,973
LARGE SCALE COMPUTE, CONTAIN A

1298
00:51:22,973 --> 00:51:26,343
LARGE NUMBER OF PARAMETERS.

1299
00:51:26,343 --> 00:51:28,878
THEY USE GENERIC ARCHITECTURES,

1300
00:51:28,878 --> 00:51:31,181
WE TALKED ABOUT CNNs,

1301
00:51:31,181 --> 00:51:32,682
TRANSFORMERS ARE POPULAR FOR

1302
00:51:32,682 --> 00:51:34,217
LARGE LANGUAGE MODELS AND

1303
00:51:34,217 --> 00:51:37,053
FOUNDATION MODELS IN GENERAL.

1304
00:51:37,053 --> 00:51:38,054
THEIR PERFORMANCE IS

1305
00:51:38,054 --> 00:51:38,621
PREDICTABLE.

1306
00:51:38,621 --> 00:51:41,224
WE'LL BE TALKING AND ALLUDED TO

1307
00:51:41,224 --> 00:51:43,760
SCALING LAWS IN THE REST OF THE

1308
00:51:43,760 --> 00:51:44,327
PRESENTATION.

1309
00:51:44,327 --> 00:51:45,562
THEY MAY DISPLAY EMERGENT

1310
00:51:45,562 --> 00:51:47,097
PROPERTIES NOT IMMEDIATELY

1311
00:51:47,097 --> 00:51:50,667
OBVIOUS FROM THE PRE-TRAINING

1312
00:51:50,667 --> 00:51:50,967
TASK.

1313
00:51:50,967 --> 00:51:51,868
FINALLY COMPOSABLE.

1314
00:51:51,868 --> 00:51:53,603
ONCE A FOUNDATION MODEL IS

1315
00:51:53,603 --> 00:51:56,439
TRAINED ITS WAYS CAN BE

1316
00:51:56,439 --> 00:51:59,309
LEVERAGED FOR ANOTHER KIND OF

1317
00:51:59,309 --> 00:51:59,542
TASK.

1318
00:51:59,542 --> 00:52:02,245
SO, WHAT ARE FOUNDATION MODELS

1319
00:52:02,245 --> 00:52:03,880
IN NEUROSCIENCE IN PARTICULAR?

1320
00:52:03,880 --> 00:52:06,082
WELL, THERE'S SEVERAL RECENT

1321
00:52:06,082 --> 00:52:08,718
PAPERS THAT CLAIM TO CREATE

1322
00:52:08,718 --> 00:52:09,486
FOUNDATION MODELS FOR

1323
00:52:09,486 --> 00:52:10,687
NEUROSCIENCE, IT CAN BE HARD TO

1324
00:52:10,687 --> 00:52:12,122
SEE THE FOREST FROM THE TREES.

1325
00:52:12,122 --> 00:52:14,190
HERE IS HOW I LIKE TO THINK

1326
00:52:14,190 --> 00:52:14,858
ABOUT THIS.

1327
00:52:14,858 --> 00:52:20,530
SO HERE IS A DIAGRAM OF AN

1328
00:52:20,530 --> 00:52:22,298
ANIMAL, IT COULD BE A HUMAN

1329
00:52:22,298 --> 00:52:23,133
INTERACTING WITH THE

1330
00:52:23,133 --> 00:52:24,768
ENVIRONMENT.

1331
00:52:24,768 --> 00:52:26,202
THERE'S A WORLD, ANIMAL

1332
00:52:26,202 --> 00:52:28,304
INTERACTS WITH THE WORLD,

1333
00:52:28,304 --> 00:52:31,241
CREATES INPUTS ON RETINA,

1334
00:52:31,241 --> 00:52:34,511
COCHLEA, FOR OLFACTION, THAT

1335
00:52:34,511 --> 00:52:35,445
DRIVES -- CREATES EXOGENOUS

1336
00:52:35,445 --> 00:52:36,946
INPUT WHICH GETS INTEGRATED INTO

1337
00:52:36,946 --> 00:52:39,549
THE STATE OF ITS BRAIN.

1338
00:52:39,549 --> 00:52:42,952
AND DETERMINES THE POSITION THE

1339
00:52:42,952 --> 00:52:45,021
ANIMAL'S BODY WHICH THEN

1340
00:52:45,021 --> 00:52:46,122
EVENTUALLY CHANGES THE CONDITION

1341
00:52:46,122 --> 00:52:46,890
OF THE WORLD.

1342
00:52:46,890 --> 00:52:51,061
AND HOWE OUT -- OUT OF THIS

1343
00:52:51,061 --> 00:52:54,764
SYSTEM WE CAN MEASURE IN LOW

1344
00:52:54,764 --> 00:52:59,903
DIMENSIONAL FASHION WITH

1345
00:52:59,903 --> 00:53:02,972
NON-INVASIVE, fMRI, EEG, OR

1346
00:53:02,972 --> 00:53:03,740
USING NEUROPIXELS.

1347
00:53:03,740 --> 00:53:05,809
SO FOUNDATION MODELS IN

1348
00:53:05,809 --> 00:53:10,713
NEUROSCIENCE TEND TO FOCUS ON

1349
00:53:10,713 --> 00:53:13,349
SOME SUBPARTS OF THIS GRAPH, AND

1350
00:53:13,349 --> 00:53:15,518
DIFFERENT KINDS OF FOUNDATION

1351
00:53:15,518 --> 00:53:19,155
MODELS FOCUS ON -- GENERALLY

1352
00:53:19,155 --> 00:53:21,224
FOCUS ON MODELING PAIRWISE PAIRS

1353
00:53:21,224 --> 00:53:25,195
OF THESE PARTS OF THE GRAPH.

1354
00:53:25,195 --> 00:53:27,397
SO, DORIS TALKED ABOUT THE

1355
00:53:27,397 --> 00:53:34,604
FAMOUS WORK FROM DAN YAMINS AND

1356
00:53:34,604 --> 00:53:37,207
JIM DICARLO, A LINK OF NEURAL

1357
00:53:37,207 --> 00:53:41,811
NETWORKS ON IMAGE CLASSIFICATION

1358
00:53:41,811 --> 00:53:42,812
AND REPRESENTATIONS, AND THE

1359
00:53:42,812 --> 00:53:44,013
VENTRAL STREAM OF THE VISUAL

1360
00:53:44,013 --> 00:53:45,315
CORTEX.

1361
00:53:45,315 --> 00:53:47,550
THERE ARE ALSO MODELS WHICH ARE

1362
00:53:47,550 --> 00:53:50,386
DIRECTLY DATA DRIVEN SO THEY USE

1363
00:53:50,386 --> 00:53:54,591
DATA FROM THE BRAINS OF ANIMALS

1364
00:53:54,591 --> 00:53:58,194
TO TRAIN A NETWORK, SO WE CAN

1365
00:53:58,194 --> 00:54:00,697
CREATE WHAT'S DEEMED A DIGITAL

1366
00:54:00,697 --> 00:54:04,334
TWIN OF A PARTICULAR SET OF

1367
00:54:04,334 --> 00:54:06,202
MOSTLY VISUAL AREAS.

1368
00:54:06,202 --> 00:54:12,976
THIS IS ADVANCED MOSTLY IN

1369
00:54:12,976 --> 00:54:13,309
VISION.

1370
00:54:13,309 --> 00:54:16,012
SOME INVERSE, GOING FROM

1371
00:54:16,012 --> 00:54:21,518
MEASUREMENTS BACK TO WHAT

1372
00:54:21,518 --> 00:54:23,353
STIMULUS WAS, POPULAR IN

1373
00:54:23,353 --> 00:54:24,921
DECODING STUDIES AND fMRI.

1374
00:54:24,921 --> 00:54:26,556
OTHER STUDIES FOCUS ON OUTPUT

1375
00:54:26,556 --> 00:54:28,291
RATHER THAN POSITION OF THE

1376
00:54:28,291 --> 00:54:29,259
BODY.

1377
00:54:29,259 --> 00:54:32,762
SO THESE ARE EXAMPLES FROM

1378
00:54:32,762 --> 00:54:38,902
DEEPLABCUT, A POPULAR TOOL IN

1379
00:54:38,902 --> 00:54:39,903
THE COMPUTATION.

1380
00:54:39,903 --> 00:54:41,704
SOME MODELS FOCUS ON MAPPING OF

1381
00:54:41,704 --> 00:54:43,406
MEASUREMENTS TO WHAT WOULD HAVE

1382
00:54:43,406 --> 00:54:46,442
BEEN THE BEHAVIOR SO WE'VE SEEN

1383
00:54:46,442 --> 00:54:48,645
TREMENDOUS ANSWERS IN

1384
00:54:48,645 --> 00:54:50,380
BRAIN-COMPUTER INTERFACES, IN

1385
00:54:50,380 --> 00:54:50,914
PARTICULAR FOR SPEECH

1386
00:54:50,914 --> 00:54:53,316
PROSTHESIS, SEE THIS IS A PAPER

1387
00:54:53,316 --> 00:54:56,386
JUST CAME OUT IN THE NEW ENGLAND

1388
00:54:56,386 --> 00:54:58,154
JOURNAL OF MEDICINE, THAT

1389
00:54:58,154 --> 00:55:02,091
SHOWCASES ABILITY TO READ THE

1390
00:55:02,091 --> 00:55:03,626
INTENSIVE SPEECH FROM A PATIENT

1391
00:55:03,626 --> 00:55:08,097
WITH A HIGHER ACCURACY THAN

1392
00:55:08,097 --> 00:55:12,468
WOULD BE POSSIBLE USING

1393
00:55:12,468 --> 00:55:14,571
SPEECH-TO-TEXT CAPABILITY SO

1394
00:55:14,571 --> 00:55:15,872
TRULY REMARKABLE CAPABILITIES

1395
00:55:15,872 --> 00:55:17,607
HAVE BEEN COMING IN THIS SPACE.

1396
00:55:17,607 --> 00:55:24,714
BUT I THINK THAT THE THING THAT

1397
00:55:24,714 --> 00:55:26,482
DIFFERENTIATES IS REALLY THIS

1398
00:55:26,482 --> 00:55:29,118
FOCUS ON MODELING THE SIGNALS

1399
00:55:29,118 --> 00:55:30,753
THEMSELVES, IN AN UNSUPERVISED

1400
00:55:30,753 --> 00:55:30,954
WAY.

1401
00:55:30,954 --> 00:55:33,656
NOW, WHY IS THIS IMPORTANT?

1402
00:55:33,656 --> 00:55:35,925
FIRST OF ALL, THIS REPRESENTS A

1403
00:55:35,925 --> 00:55:38,928
LARGE OPPORTUNITY BECAUSE THERE

1404
00:55:38,928 --> 00:55:40,797
ARE VERY LARGE SCALE BRAIN

1405
00:55:40,797 --> 00:55:42,332
ARCHIVES, JOE SPOKE ABOUT SOME

1406
00:55:42,332 --> 00:55:48,271
OF THEM, THERE'S OPEN NEURAL,

1407
00:55:48,271 --> 00:55:51,107
GHANDI, DABI, A LOT FUNDED BY

1408
00:55:51,107 --> 00:55:54,077
NIH, WE HAVE OVER 100,000 HOURS

1409
00:55:54,077 --> 00:55:55,945
OF OPEN NEURAL DATA, CREATING A

1410
00:55:55,945 --> 00:55:59,015
POSSIBILITY OF LEARNING GOOD

1411
00:55:59,015 --> 00:56:03,152
REPRESENTATIONS FROM EXISTING

1412
00:56:03,152 --> 00:56:04,687
NEURAL DATA IN UNSUPERVISED WAY

1413
00:56:04,687 --> 00:56:11,828
TO USE FOR DOWNSTREAM TASKS.

1414
00:56:11,828 --> 00:56:17,834
THIS IS ILLUSTRATED BY A DIAGRAM

1415
00:56:17,834 --> 00:56:20,670
FROM BLAKE RICHARDS WHERE ONE

1416
00:56:20,670 --> 00:56:22,305
COULD RETRAIN A NEURAL NETWORK

1417
00:56:22,305 --> 00:56:23,873
TO FIND REPRESENTATIONS OF

1418
00:56:23,873 --> 00:56:30,647
DISPARATE NEURAL DATA, WHETHER E

1419
00:56:30,647 --> 00:56:32,081
EGEG, CALCIUM IMAGING, FINE TUNE

1420
00:56:32,081 --> 00:56:34,050
FOR DIFFERENT KINDS OF

1421
00:56:34,050 --> 00:56:35,685
DOWNSTREAM TASKS INCLUDING

1422
00:56:35,685 --> 00:56:43,693
PREDICTING THE OUTCOME OF

1423
00:56:43,693 --> 00:56:44,661
CLINICAL TRIALS FOR

1424
00:56:44,661 --> 00:56:46,296
BRAIN-COMPUTER INTERFACE AND

1425
00:56:46,296 --> 00:56:48,064
PSYCHIATRIC DIAGNOSIS.

1426
00:56:48,064 --> 00:56:50,033
WHERE IS A MODEL THAT DOES SUCH

1427
00:56:50,033 --> 00:56:51,134
A THING LOOK LIKE?

1428
00:56:51,134 --> 00:56:53,036
A GOOD EXAMPLE IS THIS RECENT

1429
00:56:53,036 --> 00:56:59,776
PAPER THAT CAME OUT ABOUT A

1430
00:56:59,776 --> 00:57:03,813
MONTH AGO, TOWARDS A UNIVERSAL

1431
00:57:03,813 --> 00:57:06,115
TRANSLATORS NOR NEURAL DYNAMICS,

1432
00:57:06,115 --> 00:57:06,883
FOCUSED ON LEARNING

1433
00:57:06,883 --> 00:57:08,518
REPRESENTATIONS FROM DATA THAT

1434
00:57:08,518 --> 00:57:11,120
CAME FROM THE INTERNATIONAL

1435
00:57:11,120 --> 00:57:11,487
BRAIN LAB.

1436
00:57:11,487 --> 00:57:14,757
SO THIS IS A SET OF NEUROPIXEL

1437
00:57:14,757 --> 00:57:16,859
RECORDINGS IN MICE WHILE

1438
00:57:16,859 --> 00:57:17,827
PERFORMING A VISUAL

1439
00:57:17,827 --> 00:57:19,996
DECISION-MAKING TASK, AND THEY

1440
00:57:19,996 --> 00:57:24,801
HAVE RECORDINGS FROM MANY AREAS

1441
00:57:24,801 --> 00:57:26,135
ALL ACROSS THE MOUSE CORTEX.

1442
00:57:26,135 --> 00:57:29,505
I THINK THEY HAVE LIKE 194 AREAS

1443
00:57:29,505 --> 00:57:31,941
COVERED IN TOTAL.

1444
00:57:31,941 --> 00:57:34,977
SO THEY REALLY COVERED -- THEY

1445
00:57:34,977 --> 00:57:37,413
REALLY FOCUSED ON MODELING THE

1446
00:57:37,413 --> 00:57:40,817
RESPONSES, A HANDFUL OF THESE

1447
00:57:40,817 --> 00:57:43,319
AREAS, AND LEARNED USING A BUNCH

1448
00:57:43,319 --> 00:57:45,288
OF PRE-TRAINING, PRE-TEXT TASKS.

1449
00:57:45,288 --> 00:57:48,358
SO THESE TASKS INCLUDE TASKS

1450
00:57:48,358 --> 00:57:49,992
LIKE CO-SMOOTHING, THAT MEANS I

1451
00:57:49,992 --> 00:57:51,260
TAKE A CERTAIN NUMBER OF

1452
00:57:51,260 --> 00:57:53,029
NEURONS, PREDICT WHAT WOULD HAVE

1453
00:57:53,029 --> 00:57:54,931
BEEN THE RESPONSE OF A NEURON

1454
00:57:54,931 --> 00:57:56,666
WHICH IS LEFT OUT.

1455
00:57:56,666 --> 00:57:59,736
ALSO INCLUDES SOMETHING LIKE

1456
00:57:59,736 --> 00:58:02,038
CAUSAL PREDICTION, SO CAUSAL

1457
00:58:02,038 --> 00:58:05,308
PREDICTION IS TAKING PREVIOUS

1458
00:58:05,308 --> 00:58:07,043
ACTIVITY FROM A PARTICULAR SET

1459
00:58:07,043 --> 00:58:09,712
OF NEURONS AND PREDICTING WHAT'S

1460
00:58:09,712 --> 00:58:11,547
GOING TO HAPPEN NEXT IN THE

1461
00:58:11,547 --> 00:58:11,814
SEQUENCE.

1462
00:58:11,814 --> 00:58:13,549
AND SO ON AND SO FORTH.

1463
00:58:13,549 --> 00:58:15,718
AND THEY LEARN A GOOD

1464
00:58:15,718 --> 00:58:17,153
REPRESENTATION USING A

1465
00:58:17,153 --> 00:58:19,522
COMPLETELY GENERIC MODEL WHICH

1466
00:58:19,522 --> 00:58:21,524
IS TRANSFORMER AS DORIS ALLUDED

1467
00:58:21,524 --> 00:58:22,458
TO.

1468
00:58:22,458 --> 00:58:25,895
THE WEIGH THIS WORKS, IT'S A

1469
00:58:25,895 --> 00:58:27,296
VARIANT OF A PREVIOUSLY

1470
00:58:27,296 --> 00:58:28,398
PUBLISHED NEURAL DATA

1471
00:58:28,398 --> 00:58:32,668
TRANSFORMER WHICH WAS INTRODUCED

1472
00:58:32,668 --> 00:58:37,607
BY JOEL YE AND PANDAREINATH A

1473
00:58:37,607 --> 00:58:38,574
COUPLE YEARS AGO.

1474
00:58:38,574 --> 00:58:41,544
FOR EVERY CHUNK OF TIME THEY

1475
00:58:41,544 --> 00:58:43,279
PERFORM BINNING OF SPIKES, COUNT

1476
00:58:43,279 --> 00:58:45,548
THE TOTAL NUMBER OF SPIKES IN

1477
00:58:45,548 --> 00:58:48,184
EACH BIN, THEN THEY AGGREGATE

1478
00:58:48,184 --> 00:58:50,520
INTO ONE VECTOR, PERFORM A

1479
00:58:50,520 --> 00:58:52,021
LINEAR PROJECTION.

1480
00:58:52,021 --> 00:58:53,456
THAT CREATES A TOKEN.

1481
00:58:53,456 --> 00:58:55,525
SO, THEN ALL OF THESE TOKENS ARE

1482
00:58:55,525 --> 00:58:56,959
PUT THROUGH A DEEP NEURAL

1483
00:58:56,959 --> 00:59:00,229
NETWORK WHICH HAS MANY LAYERS OF

1484
00:59:00,229 --> 00:59:02,331
ATTENTION, IN ORDER TO TRY AND

1485
00:59:02,331 --> 00:59:04,734
RECONSTRUCT THE ACTIVITY THAT

1486
00:59:04,734 --> 00:59:05,635
WAS LEFT OUT.

1487
00:59:05,635 --> 00:59:09,906
THAT WAS MASK OUT.

1488
00:59:09,906 --> 00:59:11,541
THIS PERFORMS REASONABLY WELL.

1489
00:59:11,541 --> 00:59:12,942
NOW, AS YOU MIGHT EXPECT AND

1490
00:59:12,942 --> 00:59:17,647
YOU'LL SEE A LOT OF THESE PLOTS

1491
00:59:17,647 --> 00:59:19,715
IN THE FOLLOWING, I SHOULD SHOW

1492
00:59:19,715 --> 00:59:20,817
MY LASER POINTER.

1493
00:59:20,817 --> 00:59:24,120
THERE YOU GO.

1494
00:59:24,120 --> 00:59:26,088
SO, ONE VERY TYPICAL OUTCOME

1495
00:59:26,088 --> 00:59:29,025
FROM TRAINING THESE KINDS OF

1496
00:59:29,025 --> 00:59:29,926
MODELS IS PERFORMANCE INCREASES

1497
00:59:29,926 --> 00:59:32,528
AS WE ADD MORE AND MORE DATA

1498
00:59:32,528 --> 00:59:35,498
INTO THESE DATASETS SO AS WE

1499
00:59:35,498 --> 00:59:36,365
INCREASE NUMBER OF SESSIONS,

1500
00:59:36,365 --> 00:59:39,001
WHICH WERE USED FOR TRAINING

1501
00:59:39,001 --> 00:59:40,102
THESE MODELS, PERFORMANCE

1502
00:59:40,102 --> 00:59:41,504
INCREASES, IN THIS CASE AVERAGE

1503
00:59:41,504 --> 00:59:44,373
BITS PER SPIKE IS THE METRIC.

1504
00:59:44,373 --> 00:59:47,643
THEY HAVE FOUR PRE-TEXT TASKS,

1505
00:59:47,643 --> 00:59:49,712
LO AND BEHOLD THE MODEL BECOMES

1506
00:59:49,712 --> 00:59:52,014
BETTER AT FOUR TASKS, THAT'S

1507
00:59:52,014 --> 00:59:52,281
EXPECTED.

1508
00:59:52,281 --> 00:59:54,884
WHAT'S MORE INTERESTING IS WHAT

1509
00:59:54,884 --> 00:59:57,954
WE SEE HERE, WHERE NOT ONLY DO

1510
00:59:57,954 --> 00:59:59,722
YOU GET BETTER AT PRETEXT BUT

1511
00:59:59,722 --> 01:00:01,757
ALSO BETTER AT OTHER DOWNSTREAM

1512
01:00:01,757 --> 01:00:02,325
TASKS.

1513
01:00:02,325 --> 01:00:03,960
IN THIS CASE PREDICTING CHOICE

1514
01:00:03,960 --> 01:00:05,928
PROBABILITY OF THE MOUSE FROM

1515
01:00:05,928 --> 01:00:08,564
NEURAL ACTIVITY, AND IN THIS ONE

1516
01:00:08,564 --> 01:00:10,199
IT'S PREDICTING THE AMOUNT OF

1517
01:00:10,199 --> 01:00:14,737
MOTION ENERGY THAT CAME FROM ITS

1518
01:00:14,737 --> 01:00:15,037
WHISKERS.

1519
01:00:15,037 --> 01:00:16,672
SO, THEREIN LIES THIS IDEA MAYBE

1520
01:00:16,672 --> 01:00:18,674
WE COULD BUILD A UNIVERSAL

1521
01:00:18,674 --> 01:00:20,943
TRANSLATOR, SOMETHING THAT CAN

1522
01:00:20,943 --> 01:00:23,145
READ ARBITRARY NEURAL ACTIVITY

1523
01:00:23,145 --> 01:00:24,547
AND FIND MEANINGFUL MANIFOLD

1524
01:00:24,547 --> 01:00:27,750
THAT CAN THEN BE USED FOR

1525
01:00:27,750 --> 01:00:30,052
SEVERAL DOWNSTREAM TASKS FROM

1526
01:00:30,052 --> 01:00:31,487
NEURAL DYNAMICS.

1527
01:00:31,487 --> 01:00:33,856
NOW, LET'S GO THROUGH A QUICK

1528
01:00:33,856 --> 01:00:35,825
TOUR OF THE STATE OF THE ART IN

1529
01:00:35,825 --> 01:00:39,562
SCALING FOUNDATION MODELS FOR

1530
01:00:39,562 --> 01:00:41,097
NEUROSCIENCE, DORIS PRESENTED

1531
01:00:41,097 --> 01:00:42,865
SEVERAL ONES, I'LL PRESENT

1532
01:00:42,865 --> 01:00:46,135
SEVERAL OTHERS ALONG THE SAME

1533
01:00:46,135 --> 01:00:46,469
DIRECTION.

1534
01:00:46,469 --> 01:00:49,238
SO DORIS MENTIONED THE WORD FROM

1535
01:00:49,238 --> 01:00:55,578
ALEX HUTH UNIVERSITY OF TEXAS

1536
01:00:55,578 --> 01:00:56,779
DECODING AUDITORY INFORMATION,

1537
01:00:56,779 --> 01:01:00,049
RATHER TEXT, IN THE CONTEXT OF

1538
01:01:00,049 --> 01:01:02,018
LISTENING TO PODCASTS.

1539
01:01:02,018 --> 01:01:04,353
THIS IS NOW TRANSLATED TO EEG.

1540
01:01:04,353 --> 01:01:10,126
THIS IS A STUDY THAT JUST CAME

1541
01:01:10,126 --> 01:01:12,895
OUT THAT RECORDINGS FROM SINGLE

1542
01:01:12,895 --> 01:01:15,731
SUBJECT I SHOULD SAY SO HIGH

1543
01:01:15,731 --> 01:01:16,966
DENSITY EEG RECORDINGS FROM A

1544
01:01:16,966 --> 01:01:20,570
SING 

1545
01:01:20,570 --> 01:01:23,072
SINGLE SUBJECT, THE SUBJECT WAS

1546
01:01:23,072 --> 01:01:25,174
SPEAKING DIFFERENT SENTENCES IN

1547
01:01:25,174 --> 01:01:29,412
JAPANESE AND WE TRAIN A NEURAL

1548
01:01:29,412 --> 01:01:30,846
NETWORK TO UNDERSTAND OR PREDICT

1549
01:01:30,846 --> 01:01:33,783
WHAT WAS A SENTENCE THEY SPOKE,

1550
01:01:33,783 --> 01:01:38,020
OUT OF 512 POSSIBLE SENTENCES.

1551
01:01:38,020 --> 01:01:38,921
SO CLASSIC MULTI-CLASS

1552
01:01:38,921 --> 01:01:40,256
CLASSIFICATION PROBLEM.

1553
01:01:40,256 --> 01:01:42,658
BUT WHAT THEY DID HERE WHICH WAS

1554
01:01:42,658 --> 01:01:43,859
VERY DIFFERENT FROM THE STATE OF

1555
01:01:43,859 --> 01:01:47,463
THE ART IS THAT THEY RECORDED

1556
01:01:47,463 --> 01:01:49,432
175 HOURS OF THIS SUBJECT,

1557
01:01:49,432 --> 01:01:52,101
PERFORMING THIS TASK.

1558
01:01:52,101 --> 01:01:55,471
THAT MEANS 7 HOURS A DAY FOR 5

1559
01:01:55,471 --> 01:01:59,609
FULL WEEKS, DAY IN, DAY OUT,

1560
01:01:59,609 --> 01:02:02,478
RECORDING FROM EEGs, WHAT'S

1561
01:02:02,478 --> 01:02:04,547
REMARKABLE IS THE SCALING CURVE.

1562
01:02:04,547 --> 01:02:08,150
DECODING ACCESS IF ONE WOULD

1563
01:02:08,150 --> 01:02:10,720
RECORD NORMAL AMOUNTS OF EEG

1564
01:02:10,720 --> 01:02:15,791
DATA, LET'S SAY 1 TO 10 HOURS,

1565
01:02:15,791 --> 01:02:17,760
WAS BELOW 5%, HOWEVER AS THEY

1566
01:02:17,760 --> 01:02:19,729
SCALED THIS DATASET UP TO 175

1567
01:02:19,729 --> 01:02:25,534
HOURS THEY WERE ABLE TO GET TO

1568
01:02:25,534 --> 01:02:27,970
50% ACCURACY ON THIS 512-WEIGHT

1569
01:02:27,970 --> 01:02:30,906
CLASSIFICATION, WHICH IS TRULY

1570
01:02:30,906 --> 01:02:31,207
ASTOUNDING.

1571
01:02:31,207 --> 01:02:32,208
NOW, THAT MEANS THAT MAYBE

1572
01:02:32,208 --> 01:02:35,077
THERE'S A LITTLE BIT MORE

1573
01:02:35,077 --> 01:02:39,749
INFORMATION IN EEG THAT'S BEEN

1574
01:02:39,749 --> 01:02:40,449
CONVENTIONALLY APPRECIATED.

1575
01:02:40,449 --> 01:02:42,051
NOW, OF COURSE IT'S NOT

1576
01:02:42,051 --> 01:02:43,986
PRACTICAL TO HAVE SOMEONE HAVE

1577
01:02:43,986 --> 01:02:46,322
TO TRAIN FOR 175 HOURS TO TRAIN

1578
01:02:46,322 --> 01:02:51,127
A DECODER, SO OTHER GROUPS ARE

1579
01:02:51,127 --> 01:02:53,329
FOCUSING INSTEAD OF CAPTURING A

1580
01:02:53,329 --> 01:02:54,764
REASONABLE AMOUNT OF DATA FROM

1581
01:02:54,764 --> 01:02:56,265
LOTS AND LOTS OF DIFFERENT

1582
01:02:56,265 --> 01:03:01,203
SUBJECTS, SO THIS IS A STUDY

1583
01:03:01,203 --> 01:03:03,706
FROM A PART OF MET A REALITY

1584
01:03:03,706 --> 01:03:08,210
LABS, ONE OF MY FORMER

1585
01:03:08,210 --> 01:03:09,178
EMPLOYERS, ELECTROMYOGRAPHY

1586
01:03:09,178 --> 01:03:09,445
DECODING.

1587
01:03:09,445 --> 01:03:12,381
AND THEY ALSO FOUND THAT THEY

1588
01:03:12,381 --> 01:03:14,550
COULD TRAIN A NETWORK, GENERIC

1589
01:03:14,550 --> 01:03:16,118
NETWORK, THAT WOULD TRANSFER

1590
01:03:16,118 --> 01:03:17,286
ACROSS DIFFERENT SUBJECTS AND

1591
01:03:17,286 --> 01:03:20,656
THAT WILL ALLOW THEM TO DECODE

1592
01:03:20,656 --> 01:03:22,825
GESTURES OR HANDWRITING FROM

1593
01:03:22,825 --> 01:03:26,595
RECORDINGS OF EMG ON THE WRIST.

1594
01:03:26,595 --> 01:03:28,798
NOW, NOTICE HOW THE

1595
01:03:28,798 --> 01:03:30,099
CLASSIFICATION ERROR DECREASES

1596
01:03:30,099 --> 01:03:31,400
IN A VERY PREDICTABLE WAY AS A

1597
01:03:31,400 --> 01:03:32,735
FUNCTION OF THE NUMBER OF

1598
01:03:32,735 --> 01:03:33,502
PARTICIPANTS.

1599
01:03:33,502 --> 01:03:36,539
IN THIS CASE THEY WENT UP TO

1600
01:03:36,539 --> 01:03:37,373
4800 TRAINING PARTICIPANTS, IN

1601
01:03:37,373 --> 01:03:40,176
THAT CASE THEY WERE ABLE TO GET

1602
01:03:40,176 --> 01:03:41,944
TO CLASSIFICATION ERROR OF LESS

1603
01:03:41,944 --> 01:03:46,315
THAN 10%. 

1604
01:03:46,315 --> 01:03:48,184
AND SIMILARLY FOR HANDWRITING.

1605
01:03:48,184 --> 01:03:50,686
AS DORIS ALLUDED TO, THESE

1606
01:03:50,686 --> 01:03:52,855
TECHNIQUES HAVE ALSO BEEN NOW

1607
01:03:52,855 --> 01:03:58,394
ADAPTED FOR DECODING IMAGES

1608
01:03:58,394 --> 01:04:03,966
FROM fMRI,  ONE IMPORTANT ASPECT

1609
01:04:03,966 --> 01:04:07,136
OF THIS PARTICULAR MODEL WHICH

1610
01:04:07,136 --> 01:04:09,772
SEEKS TO DECODE IMAGES FROM

1611
01:04:09,772 --> 01:04:13,609
VISUAL CORTEX IS THIS MODEL,

1612
01:04:13,609 --> 01:04:14,844
SDXL, STABLE DIFFUSION.

1613
01:04:14,844 --> 01:04:16,445
IT'S A DIFFUSION MODEL THAT

1614
01:04:16,445 --> 01:04:18,547
SERVES TO -- CAN BE USED TO

1615
01:04:18,547 --> 01:04:25,121
WRITE TEXT AND GENERATE IMAGES,

1616
01:04:25,121 --> 01:04:26,655
MIGHT HAVE USED DALI, FOR

1617
01:04:26,655 --> 01:04:28,624
INSTANCE, FOR THIS PURPOSE.

1618
01:04:28,624 --> 01:04:32,228
THIS MODEL WAS  PRE-TRAINED,

1619
01:04:32,228 --> 01:04:34,330
ADAPTED IN THE CONTEXT OF

1620
01:04:34,330 --> 01:04:35,097
fMRI.

1621
01:04:35,097 --> 01:04:36,499
AND SO WE HAVE THIS

1622
01:04:36,499 --> 01:04:38,567
COMPOSABILITY WHERE WE CAN TAKE

1623
01:04:38,567 --> 01:04:40,536
POWERFUL MODELS AND INTEGRATE

1624
01:04:40,536 --> 01:04:43,506
THEM INTO LARGER MODELS FOR

1625
01:04:43,506 --> 01:04:44,273
BRAIN DECODING.

1626
01:04:44,273 --> 01:04:47,109
INDEED IN THIS CASE THE FIELD

1627
01:04:47,109 --> 01:04:49,111
HAS TREMENDOUSLY IMPROVED OVER

1628
01:04:49,111 --> 01:04:49,545
TIME.

1629
01:04:49,545 --> 01:04:50,346
THESE WERE STATE-OF-THE-ART

1630
01:04:50,346 --> 01:04:51,714
IMAGES TWO YEARS AGO.

1631
01:04:51,714 --> 01:04:53,382
THESE CAME OUT LAST YEAR.

1632
01:04:53,382 --> 01:04:56,018
AND THIS WAS WHAT WAS PRESENTED

1633
01:04:56,018 --> 01:05:00,823
IN THIS CURRENT PAPER, SO AS YOU

1634
01:05:00,823 --> 01:05:06,262
CAN SEE SEMANTIC CONTENT IS

1635
01:05:06,262 --> 01:05:08,264
PRESERVED, WE'RE STARTING TO SEE

1636
01:05:08,264 --> 01:05:08,831
MORE DETAILS.

1637
01:05:08,831 --> 01:05:11,700
YOU MIGHT COME UP WITH THE

1638
01:05:11,700 --> 01:05:13,202
CONCLUSION WE NEED TO THROW THE

1639
01:05:13,202 --> 01:05:17,139
DATA INTO A BIG PILE OF DATA AND

1640
01:05:17,139 --> 01:05:18,374
LINEAR ALGEBRA AND STIRRING THE

1641
01:05:18,374 --> 01:05:20,109
PILE UNTIL WE GET THE RIGHT

1642
01:05:20,109 --> 01:05:21,844
RESULT BUT OF COURSE THESE

1643
01:05:21,844 --> 01:05:24,046
MODELS ARE NOT MAGIC, AND THEY

1644
01:05:24,046 --> 01:05:27,783
RELY ON SCALING LAWS THAT NEED

1645
01:05:27,783 --> 01:05:30,219
EXPONENTIAL AMOUNTS OF DATA FOR

1646
01:05:30,219 --> 01:05:31,754
LINEAR INCREASES AND

1647
01:05:31,754 --> 01:05:32,955
PERFORMANCE, SLOPE READING

1648
01:05:32,955 --> 01:05:34,023
MATTERS, ILLUSTRATED BY THIS

1649
01:05:34,023 --> 01:05:37,326
RECENT PAPER THAT SHOWED THE

1650
01:05:37,326 --> 01:05:39,728
PERFORMANCE OF DIFFERENT

1651
01:05:39,728 --> 01:05:41,497
BRAIN-BASED IMAGING PHENOTYPE

1652
01:05:41,497 --> 01:05:43,766
PREDICTIONS FOR COMPUTATIONAL

1653
01:05:43,766 --> 01:05:45,401
PSYCHIATRY.

1654
01:05:45,401 --> 01:05:46,735
THEY SHOW THEY REQUIRE

1655
01:05:46,735 --> 01:05:47,703
PROHIBITIVELY LARGE SAMPLES TO

1656
01:05:47,703 --> 01:05:48,804
MAKE THIS WORK.

1657
01:05:48,804 --> 01:05:52,842
THEY USE U.K. BIOBANK DATA, AND

1658
01:05:52,842 --> 01:05:59,615
WHAT THEY DID IS TO TAKE DTI

1659
01:05:59,615 --> 01:06:01,817
DATA, fMRI, STRUCTURAL, TO

1660
01:06:01,817 --> 01:06:02,818
PREDICT COVARIATES ALSO

1661
01:06:02,818 --> 01:06:03,786
RECORDING IN DATASETS AND

1662
01:06:03,786 --> 01:06:05,120
EXTRAPOLATED TO FIGURE OUT WHAT

1663
01:06:05,120 --> 01:06:06,956
WOULD BE THE PERFORMANCE IF WE

1664
01:06:06,956 --> 01:06:08,824
HAD A MUCH LARGER DATASET THAN

1665
01:06:08,824 --> 01:06:09,625
EVEN WHAT WAS AVAILABLE, WHICH

1666
01:06:09,625 --> 01:06:12,428
IS, YOU KNOW, MORE OR LESS

1667
01:06:12,428 --> 01:06:13,329
50,000 PEOPLE.

1668
01:06:13,329 --> 01:06:17,800
AND IT'S NOT GOOD.

1669
01:06:17,800 --> 01:06:19,668
IT'S NOT PRETTY.

1670
01:06:19,668 --> 01:06:21,303
APART FROM AGE AND SEX,

1671
01:06:21,303 --> 01:06:23,038
OBVIOUSLY MUCH EASIER WAYS TO

1672
01:06:23,038 --> 01:06:24,106
GET TO THIS INFORMATION.

1673
01:06:24,106 --> 01:06:29,612
BUT WHAT I WANT TO SHOW HERE IS

1674
01:06:29,612 --> 01:06:30,813
PREDICTION ACCURACY FOR

1675
01:06:30,813 --> 01:06:32,781
DEPRESSION, FOR BALANCE ACCURACY

1676
01:06:32,781 --> 01:06:33,048
ACTUALLY.

1677
01:06:33,048 --> 01:06:36,151
WITH A SMALL NUMBER OF SAMPLES

1678
01:06:36,151 --> 01:06:40,556
ABOUT TO GET TO . 52 PREDICTION

1679
01:06:40,556 --> 01:06:43,492
ACCURACY, WITH A MILLION THEY

1680
01:06:43,492 --> 01:06:45,261
COULD INCREASE ACCURACY TO .6,

1681
01:06:45,261 --> 01:06:46,562
QUITE BAD COMPARED TO

1682
01:06:46,562 --> 01:06:49,965
ALTERNATIVE, WHICH IS A PIECE OF

1683
01:06:49,965 --> 01:06:50,165
PAPER.

1684
01:06:50,165 --> 01:06:52,368
SO, IN SUMMARY, THERE ARE MANY

1685
01:06:52,368 --> 01:06:55,771
USE CASES FOR FOUNDATION MOD

1686
01:06:55,771 --> 01:06:58,374
UNLESS NEUROSCIENCE, SINCE

1687
01:06:58,374 --> 01:07:00,009
UNSUPERVISED LEARNING IS MORE

1688
01:07:00,009 --> 01:07:00,876
SCALABLE THAN TRADITIONAL

1689
01:07:00,876 --> 01:07:03,178
SUPERVISED LEARNING, AND WE CAN

1690
01:07:03,178 --> 01:07:05,381
THEN FINE TUNE ON SMALL SCALE

1691
01:07:05,381 --> 01:07:08,751
DOWNSTREAM DATA FOR USE CASES WE

1692
01:07:08,751 --> 01:07:10,719
CARE ABOUT.

1693
01:07:10,719 --> 01:07:12,588
THERE'S IMPRESSIVE EXAMPLES OF

1694
01:07:12,588 --> 01:07:14,323
NEURAL DECODING, CREATING A

1695
01:07:14,323 --> 01:07:16,425
COMPOSABLE LEGO BLOCK MODEL OF

1696
01:07:16,425 --> 01:07:19,028
DEVELOPMENT WHICH CAN ACCELERATE

1697
01:07:19,028 --> 01:07:22,998
SCIENCE POTENTIALLY BUT IT'S NOT

1698
01:07:22,998 --> 01:07:23,465
A PANACEA.

1699
01:07:23,465 --> 01:07:25,834
SCALING LAWS NEED TO BE ON ONE

1700
01:07:25,834 --> 01:07:27,269
SIDE IN THIS COMPUTATIONAL

1701
01:07:27,269 --> 01:07:29,104
PHENOTYPING STUDY, AS WE SAW.

1702
01:07:29,104 --> 01:07:31,740
BRIEFLY I'LL SPEAK ABOUT SOME

1703
01:07:31,740 --> 01:07:33,275
ETHICAL CONSIDERATIONS FOR THIS

1704
01:07:33,275 --> 01:07:34,276
KIND OF MODEL.

1705
01:07:34,276 --> 01:07:37,346
AND I DO WANT TO SEPARATE THIS

1706
01:07:37,346 --> 01:07:39,415
DISCUSSION IN TERMS OF NUMBER

1707
01:07:39,415 --> 01:07:40,516
ONE BASIC SCIENCE, ANIMAL

1708
01:07:40,516 --> 01:07:43,919
RESEARCH, WHERE I THINK THE

1709
01:07:43,919 --> 01:07:46,088
BALANCE OF THE OUTCOMES IS

1710
01:07:46,088 --> 01:07:47,289
MOSTLY POSITIVE.

1711
01:07:47,289 --> 01:07:48,490
LESS ANIMALS INVOLVED IN

1712
01:07:48,490 --> 01:07:50,926
RESEARCH BECAUSE WE HAVE GOOD

1713
01:07:50,926 --> 01:07:52,127
DECODING MODELS, POTENTIALLY

1714
01:07:52,127 --> 01:07:54,196
FASTER ITERATION TIMES.

1715
01:07:54,196 --> 01:07:55,497
BUT OF COURSE WE'RE ALSO

1716
01:07:55,497 --> 01:07:59,001
INTERESTED IN CLINICAL

1717
01:07:59,001 --> 01:08:01,303
APPLICATIONS WHERE IT CREATES

1718
01:08:01,303 --> 01:08:02,538
ALL THE COMPLEXITIES OF HUMAN

1719
01:08:02,538 --> 01:08:07,876
SUBJECTS RESEARCH OF COURSE.

1720
01:08:07,876 --> 01:08:10,279
NOW, CLASSICALLY, PEOPLE HAVE

1721
01:08:10,279 --> 01:08:10,946
CONSIDERED ETHICAL

1722
01:08:10,946 --> 01:08:12,581
CONSIDERATIONS OF MACHINE

1723
01:08:12,581 --> 01:08:14,216
LEARNING AND NEUROSCIENCE, DORIS

1724
01:08:14,216 --> 01:08:16,952
MENTIONED SOME OF THEM,

1725
01:08:16,952 --> 01:08:21,156
FAIRNESS, BIAS, DATA OWNERSHIP,

1726
01:08:21,156 --> 01:08:23,859
CONSENT, PRIVACY, SECURITY,

1727
01:08:23,859 --> 01:08:25,094
AUTONOMY, AGENCY, THESE STILL

1728
01:08:25,094 --> 01:08:25,861
REMAIN WITH FOUNDATION MODELS.

1729
01:08:25,861 --> 01:08:27,496
YOU ABOUT A TOP OF THAT WE DO

1730
01:08:27,496 --> 01:08:29,898
ADD A NEW SET OF CONCERNS WHICH

1731
01:08:29,898 --> 01:08:31,433
ARE RATHER UNIQUE TO THIS

1732
01:08:31,433 --> 01:08:32,668
SITUATION OF FOUNDATION MODELS.

1733
01:08:32,668 --> 01:08:43,112
I THINK THE LARGEST ONE IS

1734
01:08:45,214 --> 01:08:48,083
INSCRUTABLEITY, HARD TO

1735
01:08:48,083 --> 01:08:49,952
UNDERSTAND WHY THE MODELS DO THE

1736
01:08:49,952 --> 01:08:52,788
THINGS THEY DO, IT BECOMES

1737
01:08:52,788 --> 01:08:54,023
DIFFICULT TO CREATE

1738
01:08:54,023 --> 01:08:54,990
COMPREHENSIVE EVALUATIONS OF

1739
01:08:54,990 --> 01:08:56,291
THESE MODELS.

1740
01:08:56,291 --> 01:08:58,494
A SECONDARY ISSUE HERE WHICH IS

1741
01:08:58,494 --> 01:09:01,096
EQUITY OF ACCESS TO DATA TOOLING

1742
01:09:01,096 --> 01:09:03,432
AND INFRASTRUCTURE.

1743
01:09:03,432 --> 01:09:04,400
WE RISK MARGINALIZING ACADEMIC

1744
01:09:04,400 --> 01:09:07,703
RESEARCH IF, YOU KNOW, WE NEED A

1745
01:09:07,703 --> 01:09:10,406
BIG GPU CLUSTER TO DO RESEARCH

1746
01:09:10,406 --> 01:09:15,110
ON THIS KIND OF MODELING, WHICH

1747
01:09:15,110 --> 01:09:17,312
RAISES SOME ISSUES ABOUT

1748
01:09:17,312 --> 01:09:18,514
AUDITING EXISTING MODELS OUT IN

1749
01:09:18,514 --> 01:09:19,481
THE WILD.

1750
01:09:19,481 --> 01:09:22,017
MANY OF THE MODELS THAT WE NOW

1751
01:09:22,017 --> 01:09:24,119
HAVE ACCESS TO ARE OPEN WEIGHTS

1752
01:09:24,119 --> 01:09:27,189
BUT NOT OPEN SOURCE SO WE DON'T

1753
01:09:27,189 --> 01:09:28,524
KNOW EXACTLY THE TRAINING DATA,

1754
01:09:28,524 --> 01:09:32,227
WHICH IS INSIDE OF THEM, NOR THE

1755
01:09:32,227 --> 01:09:35,531
WAYS IN THIS THEY WERE TRAINED.

1756
01:09:35,531 --> 01:09:38,600
MANY ACADEMIC LABS WITH NOT WELL

1757
01:09:38,600 --> 01:09:41,336
EQUIPPED TO FULLY UNDERSTAND AND

1758
01:09:41,336 --> 01:09:43,539
TRAIN THESE MODELS FROM SCRATCH,

1759
01:09:43,539 --> 01:09:45,641
WHICH LEADS THE MODEL TRAINING

1760
01:09:45,641 --> 01:09:49,211
TO THE HANDS OF MORE WELL

1761
01:09:49,211 --> 01:09:50,279
RESOURCED ENTITIES AND

1762
01:09:50,279 --> 01:09:50,546
COMPANIES.

1763
01:09:50,546 --> 01:09:52,147
AND WE MIGHT NEED TO CONSIDER

1764
01:09:52,147 --> 01:09:56,318
NEW FUNDING MODELS IN ORDER TO

1765
01:09:56,318 --> 01:09:57,619
SUPPORT THE CREATION AND

1766
01:09:57,619 --> 01:10:00,489
EVALUATION OF THESE KINDS OF

1767
01:10:00,489 --> 01:10:01,790
FOUNDATION MODELS.

1768
01:10:01,790 --> 01:10:02,991
SO I HAVE ANOTHER SLIDE.

1769
01:10:02,991 --> 01:10:05,894
I WILL SKIP IT BECAUSE I'M OUT

1770
01:10:05,894 --> 01:10:06,395
OF TIME.

1771
01:10:06,395 --> 01:10:09,765
I HAVE A LOT OF CONTENT THAT I

1772
01:10:09,765 --> 01:10:11,533
WASN'T ABLE TO GO THROUGH.

1773
01:10:11,533 --> 01:10:17,339
I WROTE THIS UP AS A BLOG POST

1774
01:10:17,339 --> 01:10:19,007
ON NEUROAI.SCIENCE WHERE YOU CAN

1775
01:10:19,007 --> 01:10:20,409
HAVE THE LINKS THAT I MENTIONED

1776
01:10:20,409 --> 01:10:20,609
HERE.

1777
01:10:20,609 --> 01:10:22,845
THANK YOU FOR HAVING ME.

1778
01:10:22,845 --> 01:10:24,480
I LOOK FORWARD TO YOUR

1779
01:10:24,480 --> 01:10:24,913
QUESTIONS.

1780
01:10:24,913 --> 01:10:25,280
>> WONDERFUL.

1781
01:10:25,280 --> 01:10:26,548
THANK YOU SO MUCH.

1782
01:10:26,548 --> 01:10:29,918
THANK YOU TO ALL THREE OF YOU,

1783
01:10:29,918 --> 01:10:31,353
JOE, DORIS, PATRICK.

1784
01:10:31,353 --> 01:10:33,122
I HOPE YOU DON'T MIND I USED

1785
01:10:33,122 --> 01:10:34,323
YOUR FIRST NAMES.

1786
01:10:34,323 --> 01:10:35,724
I'LL SEE IF ANYBODY HAS

1787
01:10:35,724 --> 01:10:36,725
QUESTIONS FOR THE GROUP.

1788
01:10:36,725 --> 01:10:39,461
AND I REALLY APPRECIATED THAT

1789
01:10:39,461 --> 01:10:41,964
LAST PART ABOUT DISTINGUISHING

1790
01:10:41,964 --> 01:10:42,965
BASIC SCIENCE FROM CLINICAL

1791
01:10:42,965 --> 01:10:44,700
APPLICATION BEFORE WE GET INTO

1792
01:10:44,700 --> 01:10:45,701
CLINICAL APPLICATION, I'LL SEE

1793
01:10:45,701 --> 01:10:47,069
IF PEOPLE HAVE QUESTIONS.

1794
01:10:47,069 --> 01:10:50,205
ANY QUESTIONS FROM THE GROUP?

1795
01:10:50,205 --> 01:10:52,040
>> I WANT TO ADD MY
THANKS.

1796
01:10:52,040 --> 01:10:53,475
THAT WAS REALLY VERY

1797
01:10:53,475 --> 01:10:54,343
INFORMATIVE, GREAT WAY TO SET

1798
01:10:54,343 --> 01:10:57,179
THE STAGE FOR THE LATER

1799
01:10:57,179 --> 01:10:57,479
DISCUSSIONS.

1800
01:10:57,479 --> 01:10:59,248
AND FOR THOSE, PATRICK MENTIONED

1801
01:10:59,248 --> 01:11:00,582
A BLOG POST.

1802
01:11:00,582 --> 01:11:02,317
I ALWAYS TRY TO ENCOURAGED MY

1803
01:11:02,317 --> 01:11:05,387
STUDENTS IN MY TEACHING DAYS,

1804
01:11:05,387 --> 01:11:06,889
ALWAYS GOOD TO READ ASSIGNMENT

1805
01:11:06,889 --> 01:11:07,956
AHEAD OF THE LECTURE.

1806
01:11:07,956 --> 01:11:09,558
I DID THAT LAST NIGHT.

1807
01:11:09,558 --> 01:11:10,759
IT'S VERY INFORMATIVE.

1808
01:11:10,759 --> 01:11:21,136
AGAIN, THANK YOU FOR BEING HERE.

1809
01:11:21,136 --> 01:11:24,873
ANY QUESTIONS? 

1810
01:11:24,873 --> 01:11:27,943
DR. KOROSHETZ?

1811
01:11:27,943 --> 01:11:28,243
>> HI.

1812
01:11:28,243 --> 01:11:30,779
MAYBE TANGENTIAL BUT GETS TO THE

1813
01:11:30,779 --> 01:11:37,786
POINT OF MAYBE ONE OF THE ISSUES

1814
01:11:37,786 --> 01:11:39,521
BEING WHAT PEOPLE WOULD ALLOW

1815
01:11:39,521 --> 01:11:44,459
THE DATA TO BE USED FOR, AND

1816
01:11:44,459 --> 01:11:48,730
BECAUSE THE DATA IS GOING INTO

1817
01:11:48,730 --> 01:11:50,799
THIS, YOU KNOW, GIGANTIC BANK,

1818
01:11:50,799 --> 01:11:56,505
PEOPLE ARE HAVING, YOU KNOW,

1819
01:11:56,505 --> 01:11:57,573
MULTIPLE DIFFERENT AGENDAS ON

1820
01:11:57,573 --> 01:12:00,442
FOLKS WHO ARE EXPLORING THIS

1821
01:12:00,442 --> 01:12:03,178
DATA, YOU KNOW, IF YOU HAD YOUR

1822
01:12:03,178 --> 01:12:04,846
DRUTHERS, WE TALKED ABOUT THIS

1823
01:12:04,846 --> 01:12:07,015
BEFORE, YOU KNOW, YOU WOULD WANT

1824
01:12:07,015 --> 01:12:09,651
TO HAVE SOMEBODY BE ABLE TO OPT

1825
01:12:09,651 --> 01:12:11,053
OUT OF THE RESEARCH BUT WHAT

1826
01:12:11,053 --> 01:12:12,154
WOULD THAT MEAN?

1827
01:12:12,154 --> 01:12:15,224
YOU SEND THEM AN E-MAIL AND THEY

1828
01:12:15,224 --> 01:12:16,992
SAY, I DON'T WANT MY -- I DON'T

1829
01:12:16,992 --> 01:12:18,393
REALLY WANT MY DATA USED FOR

1830
01:12:18,393 --> 01:12:21,396
THIS MODEL?

1831
01:12:21,396 --> 01:12:23,599
BUT IT'S OKAY FOR THAT MODEL.

1832
01:12:23,599 --> 01:12:25,334
SO THAT SEEMS THE RIGHT THING TO

1833
01:12:25,334 --> 01:12:27,836
DO BUT NOT FEASIBLE.

1834
01:12:27,836 --> 01:12:28,503
THE QUESTION COMES UP, WHETHER

1835
01:12:28,503 --> 01:12:31,006
OR NOT YOU COULD COLLECT DATA

1836
01:12:31,006 --> 01:12:36,278
FROM PEOPLE THAT WOULD THEN BE

1837
01:12:36,278 --> 01:12:40,816
ABLE TO BE PHUT  PUT IN A.I. MOL

1838
01:12:40,816 --> 01:12:47,089
THAT WE TICKETS -- PREDICTS WHAT

1839
01:12:47,089 --> 01:12:49,524
KIND THEY WANT THEIR DATA TO BE

1840
01:12:49,524 --> 01:12:51,460
IN, WONDERING IF THERE'S AN A.I.

1841
01:12:51,460 --> 01:12:55,430
SOLUTION TO OUR CONSENT PROBLEM.

1842
01:12:55,430 --> 01:12:58,700
>> DO ANY OF THE SPEAKERS
-- I

1843
01:12:58,700 --> 01:13:00,335
THINK OUR FIRST CASE MIGHT

1844
01:13:00,335 --> 01:13:02,437
INTRODUCE SOME RESPONSES TO THAT

1845
01:13:02,437 --> 01:13:03,305
QUESTION.

1846
01:13:03,305 --> 01:13:09,411
LET'S SEE IF ANY OF THE SPEAKERS

1847
01:13:09,411 --> 01:13:10,712
WHO WENT SO FAR WANT TO RESPOND

1848
01:13:10,712 --> 01:13:15,017
TO THAT.

1849
01:13:15,017 --> 01:13:16,451
>> I CAN SAY A FEW WORDS.

1850
01:13:16,451 --> 01:13:19,921
I DO THINK THAT WE HAVE TO BE

1851
01:13:19,921 --> 01:13:21,790
CAREFUL ABOUT HOW WE DESIGN

1852
01:13:21,790 --> 01:13:23,325
CONSENT FOR THIS KIND OF MODEL.

1853
01:13:23,325 --> 01:13:26,061
I DON'T KNOW THAT IT'S A GREAT

1854
01:13:26,061 --> 01:13:28,363
IDEA TO DIRECTLY TRY TO READ

1855
01:13:28,363 --> 01:13:31,133
THIS INFORMATION FROM THE BRAIN.

1856
01:13:31,133 --> 01:13:34,202
THAT SEEMS ALSO RIFE WITH

1857
01:13:34,202 --> 01:13:36,171
ETHICAL CONCERNS.

1858
01:13:36,171 --> 01:13:40,509
WHEN WE'RE DOING INFORM CONSENT,

1859
01:13:40,509 --> 01:13:43,478
IT IS VERY DIFFICULTY THINK FOR

1860
01:13:43,478 --> 01:13:44,913
PATIENTS TO FULLY GRASP WHAT

1861
01:13:44,913 --> 01:13:47,883
THEY ARE ACTUALLY SIGNING UP

1862
01:13:47,883 --> 01:13:48,216
FOR.

1863
01:13:48,216 --> 01:13:51,386
AND AS WE'VE SEEN IN THE

1864
01:13:51,386 --> 01:13:52,154
DEPLOYMENT OF DIFFERENT PRIVACY

1865
01:13:52,154 --> 01:13:55,090
LAWS FOR INSTANCE IN EUROPE,

1866
01:13:55,090 --> 01:13:56,525
WE'VE SEEN THAT GIVING CONSENT

1867
01:13:56,525 --> 01:13:58,493
TO SOMETHING AS SIMPLE AS

1868
01:13:58,493 --> 01:14:02,331
WHETHER ONE USES COOKIES OR NOT

1869
01:14:02,331 --> 01:14:05,734
ON A WEBSITE IS A COMPLICATED

1870
01:14:05,734 --> 01:14:05,967
ORDEAL.

1871
01:14:05,967 --> 01:14:10,205
WHEN YOU ADD THE EXTRA FACTOR OF

1872
01:14:10,205 --> 01:14:12,607
THE PRIVACY OF BRAIN DATA THAT

1873
01:14:12,607 --> 01:14:14,176
BECOMES VERY DIFFICULTY THINK

1874
01:14:14,176 --> 01:14:20,282
FOR A LAYPERSON TO MAKE INFORMED

1875
01:14:20,282 --> 01:14:21,817
CONSENT IN AN AUTOMATIC WAY.

1876
01:14:21,817 --> 01:14:22,150
>> THANKS.

1877
01:14:22,150 --> 01:14:25,220
WE'LL TALK ABOUT THAT MORE.

1878
01:14:25,220 --> 01:14:25,654
TIM?

1879
01:14:25,654 --> 01:14:26,154
>> HI.

1880
01:14:26,154 --> 01:14:28,390
I WANTED TO THANK THE THREE

1881
01:14:28,390 --> 01:14:29,858
SPEAKERS FOR THE PRESENTATIONS,

1882
01:14:29,858 --> 01:14:30,292
REALLY INFORMATIVE.

1883
01:14:30,292 --> 01:14:32,561
I HAVE A QUESTION, MORE FOR

1884
01:14:32,561 --> 01:14:33,862
PATRICK, CERTAINLY ANYONE IS

1885
01:14:33,862 --> 01:14:35,297
WELCOME TO COMMENT.

1886
01:14:35,297 --> 01:14:43,739
THAT'S ABOUT YOUR COMMENT ABOUT

1887
01:14:43,739 --> 01:14:46,308
INSCRUTABILITY OF MODELS AND

1888
01:14:46,308 --> 01:14:46,875
DATA GENERATED.

1889
01:14:46,875 --> 01:14:50,512
WHAT ARE ARE SOME WAYS OF

1890
01:14:50,512 --> 01:14:50,912
GETTING PAST THAT

1891
01:14:50,912 --> 01:14:52,514
INSCRUTABILITY, IS IT POSSIBLE

1892
01:14:52,514 --> 01:14:55,350
TO PUT IN STANDARDS HOW THE DATA

1893
01:14:55,350 --> 01:14:57,119
IS DEVELOPED AND INTERPRETED?

1894
01:14:57,119 --> 01:14:59,087
IS IT IMPORTANT TO DO THE

1895
01:14:59,087 --> 01:15:07,195
ANALYSIS FROM TWO DIFFERENT

1896
01:15:07,195 --> 01:15:10,365
APPROACHES, TWO A.I. APPROACHES,

1897
01:15:10,365 --> 01:15:11,366
IT'S GOING TO BE ALMOST

1898
01:15:11,366 --> 01:15:13,869
IMPOSSIBLE TO GO BACK AND DOUBLE

1899
01:15:13,869 --> 01:15:16,805
CHECK THESE TYPES OF DATA

1900
01:15:16,805 --> 01:15:17,506
OUTPUTS.

1901
01:15:17,506 --> 01:15:26,348
>> YEAH, ABSOLUTELY.

1902
01:15:26,348 --> 01:15:27,783
IN TERMS OF INSCRUTABILITY,

1903
01:15:27,783 --> 01:15:31,052
OFTENTIMES THE DATA IS CLOSED.

1904
01:15:31,052 --> 01:15:34,456
WHEN IT'S HUMAN SUBJECTS

1905
01:15:34,456 --> 01:15:37,426
RESEARCH AND BRAIN DATA, CONSENT

1906
01:15:37,426 --> 01:15:38,193
CONTAINS SENSITIVE HEALTH DATA,

1907
01:15:38,193 --> 01:15:40,061
GOOD REASONS TO PUT THAT DATA

1908
01:15:40,061 --> 01:15:40,595
BEHIND CLOSED DOORS.

1909
01:15:40,595 --> 01:15:43,665
BUT OF COURSE THAT MAKES IT

1910
01:15:43,665 --> 01:15:46,401
HARDER FOR RESEARCHERS TO THEN

1911
01:15:46,401 --> 01:15:47,803
AUDIT THESE MODELS, SO IF THE

1912
01:15:47,803 --> 01:15:49,104
DATA IS NOT AVAILABLE THAT'S

1913
01:15:49,104 --> 01:15:50,338
VERY HARD TO DO.

1914
01:15:50,338 --> 01:15:53,275
BUT EVEN IN THE CASES WHERE DATA

1915
01:15:53,275 --> 01:15:55,811
IS OPEN AND AVAILABLE IT IS VERY

1916
01:15:55,811 --> 01:15:58,413
HARD TO SEE WHAT'S INSIDE OF

1917
01:15:58,413 --> 01:16:00,715
THESE LARGE SCALE MODELS.

1918
01:16:00,715 --> 01:16:03,385
SO LLAMA WAS TRAINED, A LARGE

1919
01:16:03,385 --> 01:16:05,720
LANGUAGE MODEL FROM META, WAS

1920
01:16:05,720 --> 01:16:07,422
TRAINED ON 15 TRILLION TOKENS OF

1921
01:16:07,422 --> 01:16:08,723
TEXT, RIGHT?

1922
01:16:08,723 --> 01:16:12,461
SO JUST TO GIVE YOU AN IDEA, I

1923
01:16:12,461 --> 01:16:21,203
CALCULATED THAT THE LARGEST

1924
01:16:21,203 --> 01:16:24,039
MODEL COMMONLY READ, IT'S ABOUT

1925
01:16:24,039 --> 01:16:26,241
TWO MILLION TOKENS.

1926
01:16:26,241 --> 01:16:32,714
HIS LIFE'S WORK, EIGHT MILLION

1927
01:16:32,714 --> 01:16:34,783
PEOPLE'S WORK IN THIS DATASET,

1928
01:16:34,783 --> 01:16:36,318
NO HUMAN IS ABLE TO LOOK AT THIS

1929
01:16:36,318 --> 01:16:37,686
DATA ONE AFTER ANOTHER.

1930
01:16:37,686 --> 01:16:40,121
PEOPLE ARE MOVING IN THE FIELD

1931
01:16:40,121 --> 01:16:49,431
OF LARGE LANGUAGE MODELS TOWARDS

1932
01:16:49,431 --> 01:16:50,866
MECHANISTIC

1933
01:16:50,866 --> 01:16:51,533
INTERPRETABLEABILITY.

1934
01:16:51,533 --> 01:16:53,702
IT'S NEUROSCIENCE ON ARTIFICIAL

1935
01:16:53,702 --> 01:16:54,903
NEURAL NETWORKS, IT'S REALLY

1936
01:16:54,903 --> 01:17:04,112
EMERGING AS AN INTERESTING AREA 

1937
01:17:04,112 --> 01:17:05,347
WHAT IS THE LOADING, HOW MUCH

1938
01:17:05,347 --> 01:17:07,315
DOES THAT INFLUENCE THE WEIGHTS

1939
01:17:07,315 --> 01:17:09,718
OF THE MODEL?

1940
01:17:09,718 --> 01:17:12,454
OR CAN WE CONTROL A MODEL IN A

1941
01:17:12,454 --> 01:17:13,121
SPECIFIC WAY?

1942
01:17:13,121 --> 01:17:18,026
AND I THINK KEEPING AN EYE ON

1943
01:17:18,026 --> 01:17:20,228
THIS RESEARCH IS VERY VALUABLE.

1944
01:17:20,228 --> 01:17:23,031
WE'VE SEEN A COUPLE INSTANCES

1945
01:17:23,031 --> 01:17:26,635
WHERE THERE HAS BEEN A DIALOGUE

1946
01:17:26,635 --> 01:17:27,903
BETWEEN NEUROSCIENTISTS AND

1947
01:17:27,903 --> 01:17:30,839
ARTIFICIAL INTELLIGENCE

1948
01:17:30,839 --> 01:17:31,406
RESEARCHERS.

1949
01:17:31,406 --> 01:17:36,411
I'M THINKING OF A PAPER BY GRACE

1950
01:17:36,411 --> 01:17:38,280
LINDSEY AND DAVID BOW THAT WE

1951
01:17:38,280 --> 01:17:39,714
SHOULD TEST THE TOOLS OF

1952
01:17:39,714 --> 01:17:43,518
NEUROSCIENCE ON A.I. AND MAYBE

1953
01:17:43,518 --> 01:17:44,853
BRING THAT BACK TO NEUROSCIENCE.

1954
01:17:44,853 --> 01:17:47,489
SO IT IS THE CASE THAT THERE

1955
01:17:47,489 --> 01:17:49,324
MIGHT BE SOME POSITIVE OUTCOMES

1956
01:17:49,324 --> 01:17:53,161
FROM LIKE REALLY ENGAGING IN

1957
01:17:53,161 --> 01:17:55,330
THIS RESEARCH ON, YEAH, BREAKING

1958
01:17:55,330 --> 01:17:56,665
APART MODELS AND FIGURING HOW

1959
01:17:56,665 --> 01:17:59,834
THEY WORK THE SAME WAY WE DO

1960
01:17:59,834 --> 01:18:04,205
NEUROSCIENCE ON REAL BRAINS.

1961
01:18:04,205 --> 01:18:04,873
>> INTERESTING.

1962
01:18:04,873 --> 01:18:06,074
>> OTHER QUESTIONS?

1963
01:18:06,074 --> 01:18:07,609
GO AHEAD, JOE, PLEASE.

1964
01:18:07,609 --> 01:18:14,616
>> I WANT TO ADD THERE ARE
SOME

1965
01:18:14,616 --> 01:18:16,785
RECENT APPROACHES INCLUDE

1966
01:18:16,785 --> 01:18:18,753
ENSEMBLE MODELS, YOU HAVE A

1967
01:18:18,753 --> 01:18:20,622
SENSE OF HOW THEY ARE IN

1968
01:18:20,622 --> 01:18:22,123
CONFLICT OR CAN BE RECONCILED

1969
01:18:22,123 --> 01:18:27,495
AND SO IN THAT WAY YOU CAN

1970
01:18:27,495 --> 01:18:31,433
MITIGATE SPURIOUS VISIBILITY,

1971
01:18:31,433 --> 01:18:34,603
SOMETIMES INSCRUTABILITY IS A

1972
01:18:34,603 --> 01:18:38,473
PROBLEM BUT THERE'S SPURIOUS

1973
01:18:38,473 --> 01:18:40,508
SCRUTABILITY, MAY NOT BE

1974
01:18:40,508 --> 01:18:42,677
GROUNDED TO ACTUAL DATA, WE NEED

1975
01:18:42,677 --> 01:18:45,447
TO KEEP PEOPLE IN THE LOOP WITH

1976
01:18:45,447 --> 01:18:47,749
PROVENANCE AND DATA CONTROLS

1977
01:18:47,749 --> 01:18:51,252
LIKE WE DO FOR SHARING

1978
01:18:51,252 --> 01:18:51,886
ECOSYSTEMS.

1979
01:18:51,886 --> 01:18:53,755
>> THANK YOU.

1980
01:18:53,755 --> 01:18:53,922
TOR?

1981
01:18:53,922 --> 01:18:55,090
>> TERRIFIC TALKS, THANK
YOU SO

1982
01:18:55,090 --> 01:18:55,390
MUCH.

1983
01:18:55,390 --> 01:18:56,858
THIS IS A BROAD QUESTION BUT I

1984
01:18:56,858 --> 01:18:58,493
WANT TO PUT IT OUT THERE.

1985
01:18:58,493 --> 01:19:00,762
THE MESSAGE THAT I'M GETTING IS

1986
01:19:00,762 --> 01:19:03,832
THAT IT'S A COMBINATION OF DATA,

1987
01:19:03,832 --> 01:19:05,033
SKILLS, COMPUTE RESOURCES, YOU

1988
01:19:05,033 --> 01:19:07,335
NEED ALL THOSE THINGS.

1989
01:19:07,335 --> 01:19:08,536
THAT'S THE KEY.

1990
01:19:08,536 --> 01:19:11,172
WE WANT THE RIGHT PEOPLE TO HAVE

1991
01:19:11,172 --> 01:19:12,374
ACCESS TO ALL THOSE RESOURCES TO

1992
01:19:12,374 --> 01:19:15,543
DO THINGS THAT ARE IN THE PUBLIC

1993
01:19:15,543 --> 01:19:16,077
INTEREST.

1994
01:19:16,077 --> 01:19:18,613
BUT OF COURSE IF PEOPLE -- IF

1995
01:19:18,613 --> 01:19:20,215
EVERYBODY HAS ACCESS TO ALL

1996
01:19:20,215 --> 01:19:22,150
THESE THINGS THERE WOULD BE

1997
01:19:22,150 --> 01:19:23,218
CRAZY THINGS THAT WOULD HAPPEN,

1998
01:19:23,218 --> 01:19:25,086
A LOT OF THE WRONG STUFF, USED

1999
01:19:25,086 --> 01:19:26,121
FOR THE WRONG PURPOSES.

2000
01:19:26,121 --> 01:19:28,356
SO I FEEL LIKE ONE OF THE BIG

2001
01:19:28,356 --> 01:19:30,558
PICTURE QUESTIONS FOR ME IS HOW

2002
01:19:30,558 --> 01:19:36,598
DOES ONE OR HOW DOES, YOU KNOW,

2003
01:19:36,598 --> 01:19:38,767
NIH CREATE SOMETHING WHERE ALL

2004
01:19:38,767 --> 01:19:40,235
THOSE RESOURCES ARE COMING

2005
01:19:40,235 --> 01:19:42,604
TOGETHER FOR WORK IN THE PUBLIC

2006
01:19:42,604 --> 01:19:44,239
INTEREST BUT, YOU KNOW, FOR

2007
01:19:44,239 --> 01:19:46,007
ACADEMIA BUT THAT ARE NOT MAYBE

2008
01:19:46,007 --> 01:19:49,611
SHARED IN THE WRONG WAYS.

2009
01:19:49,611 --> 01:19:51,246
>> WHEN YOU SAY CREATE
SOMETHING

2010
01:19:51,246 --> 01:19:54,749
YOU MEAN LIKE GUIDANCE OF SOME

2011
01:19:54,749 --> 01:19:55,216
SORT?

2012
01:19:55,216 --> 01:19:57,318
>> WELL, YEAH.

2013
01:19:57,318 --> 01:19:58,420
>> A MECHANISM?

2014
01:19:58,420 --> 01:20:01,056
>> THINGS LIKE STRUCTURES
AND

2015
01:20:01,056 --> 01:20:03,892
TRAINING PROGRAMS TO CREATE

2016
01:20:03,892 --> 01:20:04,526
ACCESS, LARGE FEDERATED

2017
01:20:04,526 --> 01:20:05,193
DATASETS.

2018
01:20:05,193 --> 01:20:05,493
>> RIGHT.

2019
01:20:05,493 --> 01:20:07,262
>> WHERE PEOPLE HAVE
ACCESS TO

2020
01:20:07,262 --> 01:20:09,698
THOSE, THEY HAVE TRAINING TO,

2021
01:20:09,698 --> 01:20:14,703
YOU KNOW, GAIN SKILLS, SORT OF

2022
01:20:14,703 --> 01:20:17,238
RAPIDLY EXPANDING SPACE, ACCESS

2023
01:20:17,238 --> 01:20:20,709
TO COMPETE RESOURCES WHICH MOST

2024
01:20:20,709 --> 01:20:21,710
ACADEMIC RESOURCES DON'T HAVE ON

2025
01:20:21,710 --> 01:20:27,415
THE SCALE OF LARGE COMPANIES.

2026
01:20:27,415 --> 01:20:27,849
>> THANKS.

2027
01:20:27,849 --> 01:20:28,983
KAREEM?

2028
01:20:28,983 --> 01:20:29,250
>> HI.

2029
01:20:29,250 --> 01:20:31,019
THANK YOU FOR THIS FANTASTIC

2030
01:20:31,019 --> 01:20:31,886
PRESENTATION.

2031
01:20:31,886 --> 01:20:33,321
NICE TO SEE YOU, PATRICK.

2032
01:20:33,321 --> 01:20:34,856
I HAD A QUESTION RELATED TO

2033
01:20:34,856 --> 01:20:37,358
SCALING BECAUSE I THINK THIS IS

2034
01:20:37,358 --> 01:20:38,460
SOMETHING THAT'S VERY IMPORTANT

2035
01:20:38,460 --> 01:20:40,662
BECAUSE WE'RE SEEING THIS WITH

2036
01:20:40,662 --> 01:20:43,798
NLP BEFORE, THERE WAS A LEAP IN

2037
01:20:43,798 --> 01:20:45,667
THE PERFORMANCE WITH SCALING.

2038
01:20:45,667 --> 01:20:47,869
THE DYNAMICS OF SCALING ARE

2039
01:20:47,869 --> 01:20:49,070
SOMETIMES PREDICTABLE, CAN BE

2040
01:20:49,070 --> 01:20:50,171
ESTIMATED AS YOU SHOWED TOWARDS

2041
01:20:50,171 --> 01:20:53,141
THE END BUT SOMETIMES THEY ARE

2042
01:20:53,141 --> 01:20:54,442
NOT.

2043
01:20:54,442 --> 01:20:55,443
FOR EXAMPLE FACE TRANSITIONS

2044
01:20:55,443 --> 01:20:56,945
WHEN YOU GO UP WITH SCALE, ALL

2045
01:20:56,945 --> 01:21:00,548
OF A SUDDEN YOU HAVE A HUGE LEAP

2046
01:21:00,548 --> 01:21:02,650
IN PERFORMANCE, THINGS THAT ARE

2047
01:21:02,650 --> 01:21:04,486
SUPER DIFFICULT TO PREDICT.

2048
01:21:04,486 --> 01:21:08,456
MY POINT TO THE NEURO-A.I. SIDE

2049
01:21:08,456 --> 01:21:10,091
OF THINGS DEVELOPING MODELS FOR

2050
01:21:10,091 --> 01:21:11,526
NEUROSCIENCE, PUT YOUR BRAINS ON

2051
01:21:11,526 --> 01:21:13,495
THIS, IT ALSO HAS ETHICAL

2052
01:21:13,495 --> 01:21:14,262
IMPLICATIONS.

2053
01:21:14,262 --> 01:21:15,997
IN OTHER WORDS IF WE BUILD

2054
01:21:15,997 --> 01:21:18,433
MODELS THAT WORK AT A GIVEN

2055
01:21:18,433 --> 01:21:19,634
SIZE, MANAGEABLE SAY IN

2056
01:21:19,634 --> 01:21:21,936
ACADEMIA, BUT AS YOU POINT OUT

2057
01:21:21,936 --> 01:21:24,005
SOME UPSCALING MIGHT NOT BE

2058
01:21:24,005 --> 01:21:26,741
DOABLE IN THE FRAMEWORK OF A LAB

2059
01:21:26,741 --> 01:21:28,610
OR INSTITUTION, MIGHT REQUIRE

2060
01:21:28,610 --> 01:21:31,446
MORE RESOURCES, THAT THINGS THAT

2061
01:21:31,446 --> 01:21:33,081
MAYBE HIGH TECH COMPANIES, BIG

2062
01:21:33,081 --> 01:21:34,382
TECH COMPANIES HAVE ACCESS TO AT

2063
01:21:34,382 --> 01:21:35,183
THIS POINT.

2064
01:21:35,183 --> 01:21:37,519
IF WE DO BUILD THESE MODELS WE

2065
01:21:37,519 --> 01:21:39,354
WOULD LIKE TO PREDICT HOW WELL

2066
01:21:39,354 --> 01:21:43,858
THEY WILL PERFORM AT SCALE.

2067
01:21:43,858 --> 01:21:46,027
BECAUSE THAT'S VERY DIFFICULT TO

2068
01:21:46,027 --> 01:21:47,529
DO, MIGHT BE DECISIONS MADE IN

2069
01:21:47,529 --> 01:21:50,198
TERMS OF FUNDING, FOR EXAMPLE,

2070
01:21:50,198 --> 01:21:52,567
OF WHAT TYPE OF RESEARCH SHOULD

2071
01:21:52,567 --> 01:21:55,436
BE FUNDED BECAUSE IT IS LIKELY

2072
01:21:55,436 --> 01:22:00,041
TO LEAD TO PROGRAM SHIFT OR REAL

2073
01:22:00,041 --> 01:22:00,809
IMPROVEMENT IN PERFORMANCE, SO

2074
01:22:00,809 --> 01:22:03,545
HOW DO YOU SEE THINGS EVOLVING

2075
01:22:03,545 --> 01:22:05,079
SPECIFICALLY IN THE NEUROSCIENCE

2076
01:22:05,079 --> 01:22:06,614
CONTEXT NOW WITH THESE MODELS

2077
01:22:06,614 --> 01:22:08,883
AND HOW THEY SCALE?

2078
01:22:08,883 --> 01:22:12,086
>> YEAH, THAT'S A GREAT
QUESTIO.

2079
01:22:12,086 --> 01:22:15,723
SO I THINK GENERALLY MODELS, WE

2080
01:22:15,723 --> 01:22:17,792
CAN PREDICT VERY WELL, BASED OFF

2081
01:22:17,792 --> 01:22:22,697
OF THE AMOUNT OF DATA, HOW IT IS

2082
01:22:22,697 --> 01:22:23,798
THAT LIKELIHOOD OR GOODNESS OF

2083
01:22:23,798 --> 01:22:25,867
FIT OF THE MODEL WILL EVOLVE AS

2084
01:22:25,867 --> 01:22:28,069
WE INCREASE THE NUMBER OF

2085
01:22:28,069 --> 01:22:29,470
PARAMETERS, AS WE INCREASE SIZE

2086
01:22:29,470 --> 01:22:31,573
OF THE DATA, AS WE INCREASE THE

2087
01:22:31,573 --> 01:22:34,008
SIZE OF THE COMPUTE.

2088
01:22:34,008 --> 01:22:35,877
FOR THE PRE-TRAINING TASK, WHAT

2089
01:22:35,877 --> 01:22:37,612
IS MUCH HARDER IS TO PREDICT

2090
01:22:37,612 --> 01:22:39,581
WHAT'S GOING TO HAPPEN ON THE

2091
01:22:39,581 --> 01:22:40,782
DOWNSTREAM TASKS, RIGHT?

2092
01:22:40,782 --> 01:22:45,153
AND THAT REQUIRES A KIND OF LEAP

2093
01:22:45,153 --> 01:22:46,688
OF FAITH, IF YOU WILL.

2094
01:22:46,688 --> 01:22:49,858
SO, THIS IS A PROBLEM THAT HAS

2095
01:22:49,858 --> 01:22:51,392
BEEN, YOU KNOW, DOCUMENTED IN

2096
01:22:51,392 --> 01:22:56,731
THE A.I. FIELD SO JUST TO GIVE

2097
01:22:56,731 --> 01:22:57,866
ONE EXAMPLE, IF YOU'RE ASKING

2098
01:22:57,866 --> 01:23:00,568
HOW GOOD IT WILL BE AT

2099
01:23:00,568 --> 01:23:03,538
PREDICTING THE NEXT TOKEN, MARY

2100
01:23:03,538 --> 01:23:06,274
HAD A LITTLE SOMETHING, YOU CAN

2101
01:23:06,274 --> 01:23:08,810
PREDICT THAT VERY WELL BY

2102
01:23:08,810 --> 01:23:11,012
TRAINING EVER LARGER MODELS AND

2103
01:23:11,012 --> 01:23:12,747
SEEING HOW THE LIKELIHOOD

2104
01:23:12,747 --> 01:23:14,382
CHANGES AS A FUNCTION OF THIS

2105
01:23:14,382 --> 01:23:16,017
MODEL SIZE.

2106
01:23:16,017 --> 01:23:19,187
BUT IF YOU'RE ASKING YOURSELF

2107
01:23:19,187 --> 01:23:22,156
HOW WELL WILL AN LLM BE ABLE TO

2108
01:23:22,156 --> 01:23:23,892
MULTIPLY TWO DIFFERENT NUMBERS

2109
01:23:23,892 --> 01:23:25,026
TOGETHER, TWO DIFFERENT LARGE

2110
01:23:25,026 --> 01:23:28,162
NUMBERS, THAT'S A LOT HARDER TO

2111
01:23:28,162 --> 01:23:29,364
PREDICT.

2112
01:23:29,364 --> 01:23:34,502
AND UNFORTUNATELY WE DON'T HAVE

2113
01:23:34,502 --> 01:23:36,037
GREAT WAYS OF PREDICTING SCALE

2114
01:23:36,037 --> 01:23:36,271
MODELS.

2115
01:23:36,271 --> 01:23:38,006
PEOPLE ARE TRYING TO DO THAT

2116
01:23:38,006 --> 01:23:41,509
BECAUSE THEY HAVE TREMENDOUS

2117
01:23:41,509 --> 01:23:42,710
INCENTIVES FROM MONETARY

2118
01:23:42,710 --> 01:23:45,246
PERSPECTIVE TO PREDICT HOW WELL

2119
01:23:45,246 --> 01:23:49,083
THIS MODEL WORKS ON SOME

2120
01:23:49,083 --> 01:23:51,286
ESOTERIC DOWNSTREAM TASKS.

2121
01:23:51,286 --> 01:23:54,122
IT FELT AN OPEN FRONTIER BUT WE

2122
01:23:54,122 --> 01:23:56,291
CAN DO SMALLER SCALE

2123
01:23:56,291 --> 01:23:57,292
EXPERIMENTS, PREDICT HOW WELL

2124
01:23:57,292 --> 01:23:59,260
WE'LL DO ON INITIAL PASS, IT'S

2125
01:23:59,260 --> 01:24:02,430
POSSIBLE TO CREATE SCALING LAWS

2126
01:24:02,430 --> 01:24:04,399
FOR DOWNSTREAM TASKS BUT

2127
01:24:04,399 --> 01:24:06,401
UNFORTUNATELY IF YOUR DOWNSTREAM

2128
01:24:06,401 --> 01:24:09,003
TASK IS ACCURACY OF .1% WITH

2129
01:24:09,003 --> 01:24:09,637
LARGEST MODEL, LARGEST DATASETS,

2130
01:24:09,637 --> 01:24:11,639
IS IT GOING TO GO THIS WAY OR

2131
01:24:11,639 --> 01:24:14,709
THIS WAY OR THIS WAY, IS IT

2132
01:24:14,709 --> 01:24:14,909
FLAT?

2133
01:24:14,909 --> 01:24:18,846
IT'S VERY HARD TO SEE.

2134
01:24:18,846 --> 01:24:19,147
>> THANKS.

2135
01:24:19,147 --> 01:24:22,784
CAROLYN, I'M GOING TO ASK YOU TO

2136
01:24:22,784 --> 01:24:24,218
HOLD UNTIL LATER BECAUSE WE'RE

2137
01:24:24,218 --> 01:24:24,619
BEHIND SCHEDULE.

2138
01:24:24,619 --> 01:24:26,421
I WANT TO GET US INTO THE FIRST

2139
01:24:26,421 --> 01:24:28,623
CASE, I HOPE YOU CAN RAISE YOUR

2140
01:24:28,623 --> 01:24:29,390
QUESTION LATER.

2141
01:24:29,390 --> 01:24:31,893
SO, OUR FIRST CASE IS GOING TO

2142
01:24:31,893 --> 01:24:34,095
BE ABOUT DEEP PHENOTYPING IN

2143
01:24:34,095 --> 01:24:34,963
MENTAL HEALTH FOR PSYCHIATRY

2144
01:24:34,963 --> 01:24:36,064
WHICH BRINGS SOME OF THIS

2145
01:24:36,064 --> 01:24:38,666
TOGETHER IN TERMS OF THINKING

2146
01:24:38,666 --> 01:24:39,767
ABOUT BOTH APPLICATIONS OF SOME

2147
01:24:39,767 --> 01:24:43,805
OF THESE MODELS THAT WE'VE HEARD

2148
01:24:43,805 --> 01:24:51,145
ABOUT YOU ALSO THE REAL HUMANS

2149
01:24:51,145 --> 01:24:53,147
INTERACTING IN NATURAL WORLD AND

2150
01:24:53,147 --> 01:24:55,316
REAL WORLD DATA.

2151
01:24:55,316 --> 01:24:56,851
I'LL ASK SUSAN WRIGHT TO

2152
01:24:56,851 --> 01:25:01,322
INTRODUCE SPEAKERS AND LET'S

2153
01:25:01,322 --> 01:25:03,091
JUMP IN.

2154
01:25:03,091 --> 01:25:08,663
>> FIRST SPEAKER IS DR.
JUSTIN

2155
01:25:08,663 --> 01:25:09,430
BAKER, SCIENTIFIC DIRECTOR,

2156
01:25:09,430 --> 01:25:12,066
DIRECTOR OF THE LABORATORY FOR

2157
01:25:12,066 --> 01:25:16,104
FUNCTIONAL IMAGING AND

2158
01:25:16,104 --> 01:25:17,305
BIOINFORMATICS McQUEEN

2159
01:25:17,305 --> 01:25:18,072
HOSPITAL, ASSOCIATE PROFESSOR AT

2160
01:25:18,072 --> 01:25:19,707
HARVARD MEDICAL SCHOOL, USING

2161
01:25:19,707 --> 01:25:22,110
LARGE SCALE STUDIES AND DEEP

2162
01:25:22,110 --> 01:25:22,677
MULTI-LEVEL PHENOTYPING

2163
01:25:22,677 --> 01:25:26,914
APPROACHES TO UNDERSTAND NATURE

2164
01:25:26,914 --> 01:25:37,625
AND UNDERLYING BIOLOGY OF MENTAL

2165
01:25:37,625 --> 01:25:37,992
ILLNESS.

2166
01:25:37,992 --> 01:25:41,729
SHOULD I INTRODUCE THE SECOND

2167
01:25:41,729 --> 01:25:43,064
SPEAKER?

2168
01:25:43,064 --> 01:25:44,766
I'LL TAKE A PAUSE.

2169
01:25:44,766 --> 01:25:45,099
>> SURE.

2170
01:25:45,099 --> 01:25:46,534
>> GO AHEAD, DR. BAKER.

2171
01:25:46,534 --> 01:25:54,976
>> LET ME SHARE MY SCREEN
HERE.

2172
01:25:54,976 --> 01:25:56,044
WELL, I'M JUSTIN BAKER.

2173
01:25:56,044 --> 01:25:58,780
THANK YOU FOR THE OPPORTUNITY TO

2174
01:25:58,780 --> 01:26:02,517
JOIN THIS INTERESTING SESSION

2175
01:26:02,517 --> 01:26:02,817
TODAY.

2176
01:26:02,817 --> 01:26:06,387
SO I'LL TALK ABOUT THIS IDEA OF

2177
01:26:06,387 --> 01:26:08,322
DEEP PHENOTYPING, GOING MUCH

2178
01:26:08,322 --> 01:26:12,260
BEYOND THE MORE SUPERFICIAL

2179
01:26:12,260 --> 01:26:13,127
CHARACTERIZATION OF PATIENTS

2180
01:26:13,127 --> 01:26:15,196
WITH MENTAL HEALTH CONDITIONS

2181
01:26:15,196 --> 01:26:17,198
AND MY COLLEAGUE DR. SILVERMAN

2182
01:26:17,198 --> 01:26:20,468
WILL FOLLOW FLESHING OUT ETHICAL

2183
01:26:20,468 --> 01:26:21,569
CONSIDERATIONS, SOME OF WHICH

2184
01:26:21,569 --> 01:26:24,372
HAVE ALREADY BEEN TOUCHED ON

2185
01:26:24,372 --> 01:26:25,373
THIS AFTERNOON.

2186
01:26:25,373 --> 01:26:27,375
SO WHERE WE START IS THINKING

2187
01:26:27,375 --> 01:26:29,877
ABOUT HOW DO WE MEASURE

2188
01:26:29,877 --> 01:26:31,179
SOMEONE'S MENTAL HEALTH.

2189
01:26:31,179 --> 01:26:33,047
HOW DO WE DETERMINE SOMEONE

2190
01:26:33,047 --> 01:26:34,215
MIGHT BE EXPERIENCING A MENTAL

2191
01:26:34,215 --> 01:26:35,183
HEALTH CONDITION?

2192
01:26:35,183 --> 01:26:37,085
AS WE'VE HEARD ALREADY WE CAN DO

2193
01:26:37,085 --> 01:26:38,686
THIS WITH MULTIPLE BIOLOGICAL

2194
01:26:38,686 --> 01:26:39,587
VARIABLES, THERE'S THE NOTION OF

2195
01:26:39,587 --> 01:26:40,922
PEOPLE WHO HAVE A CONDITION,

2196
01:26:40,922 --> 01:26:43,624
MEME WHO DON'T HAVE A CONDITION,

2197
01:26:43,624 --> 01:26:44,525
SO-CALLED CASES OF CONTROLS BUT

2198
01:26:44,525 --> 01:26:49,664
IN THE REAL WORLD THIS IS A MUCH

2199
01:26:49,664 --> 01:26:52,300
MORE COMPLICATED TRANSFORMATION,

2200
01:26:52,300 --> 01:26:54,502
MANY TYPES OF PSYCHIATRIC CASES,

2201
01:26:54,502 --> 01:26:56,771
THIS NOTION OF CONTINUUM FROM

2202
01:26:56,771 --> 01:26:59,941
HEALTH INTO DISEASE, AND FINALLY

2203
01:26:59,941 --> 01:27:02,810
THE NOTION OF COMORBIDITY, SAME

2204
01:27:02,810 --> 01:27:03,444
INDIVIDUAL MILD EXPERIENCE MANY

2205
01:27:03,444 --> 01:27:06,948
MENTAL HEALTH CONDITIONS AT THE

2206
01:27:06,948 --> 01:27:09,817
SAME TIME, AND SIMPLY PLOTTING

2207
01:27:09,817 --> 01:27:13,721
TWO OR EVEN TEN BIOLOGICAL

2208
01:27:13,721 --> 01:27:14,822
VARIABLES AND SEE WHERE

2209
01:27:14,822 --> 01:27:17,125
INDIVIDUALS MAYBE SIT WITHIN

2210
01:27:17,125 --> 01:27:21,329
THAT PARAMETER SPACE IS AN

2211
01:27:21,329 --> 01:27:22,530
APPROACH WHICH HASN'T PANNED OUT

2212
01:27:22,530 --> 01:27:26,267
AS WE TRY TO PREDICT OR ESTIMATE

2213
01:27:26,267 --> 01:27:27,668
SOMEONE'S DIAGNOSIS OR MENTAL

2214
01:27:27,668 --> 01:27:30,972
STATE BASED ON SUCH MEASURES.

2215
01:27:30,972 --> 01:27:35,343
SO THE APPROACH THAT WE'VE

2216
01:27:35,343 --> 01:27:36,544
TAKEN, AND ALIGNS WITH THE

2217
01:27:36,544 --> 01:27:38,279
TOPICS THAT WE'VE BEEN

2218
01:27:38,279 --> 01:27:41,015
DISCUSSING SO FAR, IS REALLY TO

2219
01:27:41,015 --> 01:27:42,783
USE MULTI-MODAL MEASUREMENTS AND

2220
01:27:42,783 --> 01:27:43,417
LATENT CONSTRUCT MODELS TO

2221
01:27:43,417 --> 01:27:45,153
ASSESS MENTAL HEALTH CONDITIONS

2222
01:27:45,153 --> 01:27:45,820
AND BEHAVIORS.

2223
01:27:45,820 --> 01:27:50,091
AND SO THE TYPES OF MULTI-MODAL

2224
01:27:50,091 --> 01:27:52,293
MEASUREMENTS WE CAN TAKE ARE

2225
01:27:52,293 --> 01:27:54,795
TRADITIONAL SELF-REPORT SUCH AS

2226
01:27:54,795 --> 01:27:55,029
SURVEYS.

2227
01:27:55,029 --> 01:27:58,299
WE CAN PULL IN DATA FROM BRAIN

2228
01:27:58,299 --> 01:27:58,833
RECORDINGS, MRIs, BLOOD

2229
01:27:58,833 --> 01:28:01,469
SAMPLES, PULL IN DATA FROM THE

2230
01:28:01,469 --> 01:28:02,770
ELECTRONIC HEALTH RECORD, BUT AS

2231
01:28:02,770 --> 01:28:04,205
I'LL TALK ABOUT TODAY THERE'S

2232
01:28:04,205 --> 01:28:06,274
ALSO MANY OTHER TIMES OF DATA,

2233
01:28:06,274 --> 01:28:08,342
RAW DATA SOURCES ONE CAN BRING

2234
01:28:08,342 --> 01:28:10,645
IN TO INFORM SOMEONE'S MENTAL

2235
01:28:10,645 --> 01:28:12,947
STATE, WHETHER IT'S VIDEO,

2236
01:28:12,947 --> 01:28:14,148
RECORDINGS FROM MICROPHONES,

2237
01:28:14,148 --> 01:28:15,917
RECORDINGS FROM WEARABLES, PHONE

2238
01:28:15,917 --> 01:28:18,452
SENSORS, THESE ARE ALL DATA

2239
01:28:18,452 --> 01:28:24,025
SOURCES WHICH IN TODAY'S WORLD

2240
01:28:24,025 --> 01:28:26,093
ARE SOMEWHAT UBIQUITOUS AND CAN

2241
01:28:26,093 --> 01:28:27,395
INFORM UNDERLYING MENTAL TATE.

2242
01:28:27,395 --> 01:28:28,629
LIKE WITH BRAIN RECORDINGS ONE

2243
01:28:28,629 --> 01:28:31,132
HAS TO MOVE THROUGH RAW DATA

2244
01:28:31,132 --> 01:28:32,733
SOURCES, BUT IN INDIVIDUAL CASES

2245
01:28:32,733 --> 01:28:34,802
WE WANT TO DO INTELLIGENT

2246
01:28:34,802 --> 01:28:36,804
FEATURE EXTRACTION OF THOSE RAW

2247
01:28:36,804 --> 01:28:39,240
DATA SOURCES, SO WHETHER THAT'S

2248
01:28:39,240 --> 01:28:42,843
EXTRACTING BRAIN SIGNALS SUCH AS

2249
01:28:42,843 --> 01:28:44,478
FUNCTIONAL OR STRUCTURAL MRI,

2250
01:28:44,478 --> 01:28:46,581
GENOMES USING NATURAL LANGUAGE

2251
01:28:46,581 --> 01:28:47,882
PROCESSING TO PARSE ELECTRONIC

2252
01:28:47,882 --> 01:28:50,518
HEALTH RECORD OR APPROACHES WE

2253
01:28:50,518 --> 01:28:52,053
CAN DERIVE FROM COMPUTER SCIENCE

2254
01:28:52,053 --> 01:28:53,888
SUCH AS COMPUTER VISION TO

2255
01:28:53,888 --> 01:28:55,122
EXTRACT INFORMATION FROM FACES,

2256
01:28:55,122 --> 01:28:58,326
AND OTHER TECHNIQUES TO TRY TO

2257
01:28:58,326 --> 01:29:01,028
PULL AND MAKE SENSE OF RAW DATA

2258
01:29:01,028 --> 01:29:01,262
SOURCES.

2259
01:29:01,262 --> 01:29:05,399
AGAIN, THE IDEA IS TO BE ABLE TO

2260
01:29:05,399 --> 01:29:09,370
MAP THOSE FEATURES INTO LATENT

2261
01:29:09,370 --> 01:29:11,105
CONSTRUCTS OF INTEREST, HOW DO

2262
01:29:11,105 --> 01:29:13,241
WE INFER SLEEP OR CIRCADIAN

2263
01:29:13,241 --> 01:29:15,843
PATTERNS FROM THOSE PATTERNS?

2264
01:29:15,843 --> 01:29:16,844
INFER EXECUTIVE FUNCTION,

2265
01:29:16,844 --> 01:29:18,913
LANGUAGE FUNCTION, LEVEL OF

2266
01:29:18,913 --> 01:29:21,882
AFFECT REGULATION, STRESS

2267
01:29:21,882 --> 01:29:24,385
SENSITIVITY, SO ON.

2268
01:29:24,385 --> 01:29:26,387
WE BELIEVE WHERE FUTURE MENTAL

2269
01:29:26,387 --> 01:29:29,457
HEALTH ASSESSMENT IS GOING IS

2270
01:29:29,457 --> 01:29:30,758
THAT THIS COMPREHENSIVE

2271
01:29:30,758 --> 01:29:32,193
PHENOTYPING APPROACH WHERE WE

2272
01:29:32,193 --> 01:29:34,595
POOL DIFFERENT RAW DATA SOURCES

2273
01:29:34,595 --> 01:29:36,397
AND MAP THEM INTO LATENT

2274
01:29:36,397 --> 01:29:37,798
PSYCHIATRIC VARIABLES WILL ALLOW

2275
01:29:37,798 --> 01:29:41,936
US TO THEN INTERFACE WITH

2276
01:29:41,936 --> 01:29:44,772
SOFTWARE THAT SYNTHESIZES ACROSS

2277
01:29:44,772 --> 01:29:45,973
DIFFERENT CONSTRUCTS, DETECT

2278
01:29:45,973 --> 01:29:49,610
CHANGES WITH LEVEL OF

2279
01:29:49,610 --> 01:29:51,479
INDIVIDUALS, FEEDS THOSE CHANGES

2280
01:29:51,479 --> 01:29:52,446
AND SYNTHESES BACK TO

2281
01:29:52,446 --> 01:29:53,581
INDIVIDUALS WHETHER IT'S THE

2282
01:29:53,581 --> 01:29:55,182
INDIVIDUAL WHO IS AT THE CENTER

2283
01:29:55,182 --> 01:29:56,484
OF CARE, THE PERSON WITH THE

2284
01:29:56,484 --> 01:29:58,019
CONDITION, OR SOMEONE WITHIN

2285
01:29:58,019 --> 01:30:00,221
THEIR CARE NETWORK, WHETHER A

2286
01:30:00,221 --> 01:30:02,056
FAMILY NETWORK, PEER, DOCTOR, SO

2287
01:30:02,056 --> 01:30:02,390
ON.

2288
01:30:02,390 --> 01:30:06,294
AND WE SEE THIS AS AN EVOLVING

2289
01:30:06,294 --> 01:30:07,328
FIELD WITH EVOLUTION IN TERMS OF

2290
01:30:07,328 --> 01:30:08,829
HOW WE MEASURE THESE THINGS, BUT

2291
01:30:08,829 --> 01:30:10,665
ALSO JUST AS IMPORTANT IS HOW WE

2292
01:30:10,665 --> 01:30:12,733
THEN DELIVER THEM BACK TO

2293
01:30:12,733 --> 01:30:14,535
INDIVIDUALS IN WAYS THAT ARE

2294
01:30:14,535 --> 01:30:18,372
ETHICAL AND LEAD TO CLINICAL

2295
01:30:18,372 --> 01:30:20,775
IMPROVEMENT OVER TIME.

2296
01:30:20,775 --> 01:30:23,944
AND SO I'LL START BY BRINGING UP

2297
01:30:23,944 --> 01:30:29,984
ONE KEY AREA, CHARACTERIZING

2298
01:30:29,984 --> 01:30:31,519
DYADIC INTERACTIONS, THE BREAD

2299
01:30:31,519 --> 01:30:33,254
AND BUTTER OF MEDICINE, HOW A

2300
01:30:33,254 --> 01:30:35,022
PATIENT AND THEIR DOCTOR ARE

2301
01:30:35,022 --> 01:30:36,991
GOING TO BE SHARING INFORMATION.

2302
01:30:36,991 --> 01:30:38,192
ONE CHALLENGE OF THIS IS THAT

2303
01:30:38,192 --> 01:30:42,897
THE WAY THIS IS DONE IN TODAY'S

2304
01:30:42,897 --> 01:30:44,098
MEDICINE, ESPECIALLY PSYCHIATRY,

2305
01:30:44,098 --> 01:30:46,300
IT'S VERY MUCH SUBJECTIVE.

2306
01:30:46,300 --> 01:30:48,035
TRAINING TO CONDUCT DYADIC

2307
01:30:48,035 --> 01:30:49,603
INTERACTION FOR MEDICAL DATA

2308
01:30:49,603 --> 01:30:51,839
EXTRACTION IS IT TAKES MONTHS TO

2309
01:30:51,839 --> 01:30:55,776
YEARS, ON TOP OF THAT ACTUALLY

2310
01:30:55,776 --> 01:30:56,844
ADMINISTERING THESE INTERVIEWS

2311
01:30:56,844 --> 01:30:58,813
AND SCORING THEM OR BEING ABLE

2312
01:30:58,813 --> 01:31:00,348
TO INTERPRET THEM CAN TAKE LONG

2313
01:31:00,348 --> 01:31:01,315
PERIODS OF TIME.

2314
01:31:01,315 --> 01:31:06,020
AND THEY ARE NOT WELL SUITED FOR

2315
01:31:06,020 --> 01:31:07,888
CAPTURING ROBUST POINT ESTIMATES

2316
01:31:07,888 --> 01:31:09,857
OF SYMPTOM SEVERITY ESPECIALLY

2317
01:31:09,857 --> 01:31:11,592
WHEN YOU CONSIDER PSYCHIATRIC

2318
01:31:11,592 --> 01:31:12,460
CONDITIONS HAVE MULTIPLE

2319
01:31:12,460 --> 01:31:14,729
DIMENSIONS TO THEM IN TERMS OF

2320
01:31:14,729 --> 01:31:15,730
AFFECT, PSYCHOSIS, ET CETERA.

2321
01:31:15,730 --> 01:31:21,068
AND SO TO DO THIS AT SCALE IS

2322
01:31:21,068 --> 01:31:22,737
SOMETHING WHICH TRADITIONAL

2323
01:31:22,737 --> 01:31:24,572
HUMAN INTERACTIONS IS PROBABLY

2324
01:31:24,572 --> 01:31:26,741
NOT WELL SUITED, ENDS UP BEING

2325
01:31:26,741 --> 01:31:28,209
EXPENSIVE AND UNRELIABLE AS A

2326
01:31:28,209 --> 01:31:28,509
RESULT.

2327
01:31:28,509 --> 01:31:30,711
SO, ONE OF THE AREAS WE'VE

2328
01:31:30,711 --> 01:31:33,447
ATTEMPTED TO INNOVATE IS BRING

2329
01:31:33,447 --> 01:31:34,415
TOGETHER COMPUTER SCIENCE WITH

2330
01:31:34,415 --> 01:31:36,183
CLINICIANS TO TRY TO COME UP

2331
01:31:36,183 --> 01:31:38,652
WITH MORE INTELLIGENT WAYS OF

2332
01:31:38,652 --> 01:31:40,321
CAPTURING THIS.

2333
01:31:40,321 --> 01:31:43,057
AND SO I'LL JUST HIGHLIGHT ONE

2334
01:31:43,057 --> 01:31:44,392
EXAMPLE OF THIS IS COLLABORATION

2335
01:31:44,392 --> 01:31:47,995
THAT WE'VE HAD WITH A GROUP OUT

2336
01:31:47,995 --> 01:31:49,830
OF CARNEGIE MELLON UNIVERSITY,

2337
01:31:49,830 --> 01:31:52,800
ORIGINALLY OUT OF UNIVERSITY OF

2338
01:31:52,800 --> 01:31:54,435
SOUTHERN CALIFORNIA,

2339
01:31:54,435 --> 01:31:54,735
MULTI-SENSE.

2340
01:31:54,735 --> 01:31:58,372
IT TAKES OFF THE SHELF CAMERAS

2341
01:31:58,372 --> 01:32:00,141
AND MICROPHONES AND USES TOOLS

2342
01:32:00,141 --> 01:32:01,742
READILY AVAILABLE FROM COMPUTER

2343
01:32:01,742 --> 01:32:02,710
SCIENCE, THIS TECHNOLOGY HAS

2344
01:32:02,710 --> 01:32:04,412
BEEN AROUND FOR MORE THAN A

2345
01:32:04,412 --> 01:32:05,980
DECADE NOW.

2346
01:32:05,980 --> 01:32:09,250
CAN IN REAL TIME EXTRACT FACIAL

2347
01:32:09,250 --> 01:32:10,684
EXPRESSION DATA, POSTURE DATA,

2348
01:32:10,684 --> 01:32:13,187
GAZE, THE WAY SOMEONE IS

2349
01:32:13,187 --> 01:32:16,123
SPEAKING, AND BE ABLE TO INFER

2350
01:32:16,123 --> 01:32:18,592
AND PROVIDE INPUTS BACK TO

2351
01:32:18,592 --> 01:32:20,327
POTENTIALLY CLINICIAN WHO COULD

2352
01:32:20,327 --> 01:32:21,662
MAKE CLINICAL DECISIONS IN REAL

2353
01:32:21,662 --> 01:32:25,266
TIME BASED ON THESE SIGNALS.

2354
01:32:25,266 --> 01:32:27,234
SO WE USING THIS TECHNOLOGY

2355
01:32:27,234 --> 01:32:28,969
WORKED DIRECTLY WITH SOME

2356
01:32:28,969 --> 01:32:30,337
CLINICIANS AT MCLEAN HOSPITAL

2357
01:32:30,337 --> 01:32:33,474
WHERE I'M BASED TO TRY TO TAKE

2358
01:32:33,474 --> 01:32:36,877
INFORMATION FROM A VERY BRIEF

2359
01:32:36,877 --> 01:32:37,978
CLINICAL INTERACTION THAT WOULD

2360
01:32:37,978 --> 01:32:38,979
CONTAIN THE SAME AMOUNT OF

2361
01:32:38,979 --> 01:32:40,381
INFORMATION ONE MIGHT BE ABLE TO

2362
01:32:40,381 --> 01:32:45,653
GET FROM HOURS OF INTERVIEWS IN

2363
01:32:45,653 --> 01:32:48,489
A RESEARCH CONTEXT.

2364
01:32:48,489 --> 01:32:49,924
WE CONSTRUCTED SEMI STRUCTURE

2365
01:32:49,924 --> 01:32:51,692
INTERVIEW WHICH TOOK QUESTIONS

2366
01:32:51,692 --> 01:32:52,660
OTHER PROMPTS THAT CLINICIAN

2367
01:32:52,660 --> 01:32:55,729
MIGHT BE ABLE TO USE IN THE

2368
01:32:55,729 --> 01:32:57,731
COURSE OF THEIR STANDARD

2369
01:32:57,731 --> 01:32:59,767
CLINICAL EVALUATIONS, AND BE

2370
01:32:59,767 --> 01:33:02,603
ABLE TO THEN COMBINE THOSE

2371
01:33:02,603 --> 01:33:03,704
PROMPTS WITH FACIAL EXPRESSION

2372
01:33:03,704 --> 01:33:05,673
DATA FROM BOTH THE PATIENT AS

2373
01:33:05,673 --> 01:33:07,208
WELL AS THE CLINICIAN.

2374
01:33:07,208 --> 01:33:09,743
SO YOU CAN SEE ON THE BOTTOM

2375
01:33:09,743 --> 01:33:12,480
HERE A HEAT MAP SHOWING HOW

2376
01:33:12,480 --> 01:33:13,814
THESE INDIVIDUAL PATIENTS IN

2377
01:33:13,814 --> 01:33:15,416
THIS CASE AND CLINICIAN

2378
01:33:15,416 --> 01:33:18,252
RESPONDED OVER THE COURSE OF A

2379
01:33:18,252 --> 01:33:18,853
VERY BRIEF 10-MINUTE

2380
01:33:18,853 --> 01:33:19,920
INTERACTION, WHERE AT THE

2381
01:33:19,920 --> 01:33:21,222
BEGINNING OF THE INTERVIEW THE

2382
01:33:21,222 --> 01:33:22,656
PATIENT'S ALONE IN THE ROOM, AND

2383
01:33:22,656 --> 01:33:24,024
YOU CAN SEE THE TOP OF THE HEAT

2384
01:33:24,024 --> 01:33:26,694
MAP SHOWS THEY ARE DOING A LOT

2385
01:33:26,694 --> 01:33:27,695
OF BROW RAISING, POTENTIALLY

2386
01:33:27,695 --> 01:33:28,562
LOOKING AROUND THE ROOM.

2387
01:33:28,562 --> 01:33:30,297
AS SOON AS THE DOCTOR ENTERS THE

2388
01:33:30,297 --> 01:33:32,933
ROOM THE PATIENT BEGINS TO

2389
01:33:32,933 --> 01:33:36,303
SMILE, SHOWN WITH CERTAIN ACTION

2390
01:33:36,303 --> 01:33:37,938
UNITS, SHOWING HIGHER LEVELS OF

2391
01:33:37,938 --> 01:33:39,507
ACTIVITY.

2392
01:33:39,507 --> 01:33:44,645
JUST LIKE YOU MIGHT WITH A

2393
01:33:44,645 --> 01:33:46,947
NEUROSCIENCE EXPERIMENT, HAVE A

2394
01:33:46,947 --> 01:33:50,651
VENN-RELATED POTENTIAL, HERE THE

2395
01:33:50,651 --> 01:33:51,418
POTENTIAL WE'RE EVALUATING IS IN

2396
01:33:51,418 --> 01:33:53,187
THE FACE.

2397
01:33:53,187 --> 01:33:55,155
WHEN THE DOCTOR ENTERS, THE

2398
01:33:55,155 --> 01:33:56,156
PATIENT SMILES.

2399
01:33:56,156 --> 01:33:57,291
IN RESPONSE TO EACH QUESTION WE

2400
01:33:57,291 --> 01:33:58,459
GET NOT JUST WHAT THE PATIENT

2401
01:33:58,459 --> 01:34:00,861
SAYS IN RESPONSE TO PROMPTS BUT

2402
01:34:00,861 --> 01:34:02,263
HOW THEIR FACE RESPONDS.

2403
01:34:02,263 --> 01:34:04,798
INTERESTINGLY WE ALSO CAN SEE

2404
01:34:04,798 --> 01:34:09,503
EXACTLY HOW THE EVALUATOR 

2405
01:34:09,503 --> 01:34:09,737
BEHAVES.

2406
01:34:09,737 --> 01:34:12,072
THIS BECOMES LIKE A BEHAVIORAL

2407
01:34:12,072 --> 01:34:13,807
EEG, USE WHATEVER KIND OF

2408
01:34:13,807 --> 01:34:16,277
STIMULUS PROBES WE WANT TO, TO

2409
01:34:16,277 --> 01:34:18,345
EVALUATE A PATIENT OR

2410
01:34:18,345 --> 01:34:19,346
PARTICIPANT, RESEARCH

2411
01:34:19,346 --> 01:34:21,215
PARTICIPANT.

2412
01:34:21,215 --> 01:34:24,485
WE CAN READ OFF OR EVENTUALLY BE

2413
01:34:24,485 --> 01:34:26,654
ABLE TO INTERPRET AND IN A SENSE

2414
01:34:26,654 --> 01:34:28,856
DO SOME OF THE KIND OF MIND

2415
01:34:28,856 --> 01:34:32,126
READING YOU WERE HEARING FROM

2416
01:34:32,126 --> 01:34:33,227
THE EARLIER PRESENTATIONS SIMPLY

2417
01:34:33,227 --> 01:34:34,428
BASED ON HOW SOMEONE'S FACE

2418
01:34:34,428 --> 01:34:36,297
RESPONDS OR OTHER PARTS OF THEIR

2419
01:34:36,297 --> 01:34:38,799
BODY MIGHT RESPOND TO AN

2420
01:34:38,799 --> 01:34:41,435
EVALUATION.

2421
01:34:41,435 --> 01:34:43,170
THESE KINDS OF TECHNOLOGIES ARE

2422
01:34:43,170 --> 01:34:44,838
CREATED BECAUSE WE ASSUME ANY

2423
01:34:44,838 --> 01:34:46,340
ONE CLINICIAN ISN'T GOING TO

2424
01:34:46,340 --> 01:34:48,108
NECESSARILY BE ATTUNED TO ALL

2425
01:34:48,108 --> 01:34:50,077
THE SUBTLETIES OF MOVEMENT AND

2426
01:34:50,077 --> 01:34:52,046
BEHAVIOR A PATIENT MY EXPERIENCE

2427
01:34:52,046 --> 01:34:54,982
OR MIGHT BE MANIFESTING, BUT WE

2428
01:34:54,982 --> 01:34:55,849
COULD DEVELOP SYSTEMS THAT

2429
01:34:55,849 --> 01:34:58,385
PERFORM AS WELL AS THE BEST

2430
01:34:58,385 --> 01:35:02,222
CLINICIAN ON THEIR BEST DAY.

2431
01:35:02,222 --> 01:35:04,825
AND SO THEN TO BRING IN ANOTHER

2432
01:35:04,825 --> 01:35:08,562
SET OF INFORMATION WE WANTED TO

2433
01:35:08,562 --> 01:35:09,697
APPLY THIS TO FOLLOWING SEVEN

2434
01:35:09,697 --> 01:35:12,499
INDIVIDUALS WITH A NUMBER OF

2435
01:35:12,499 --> 01:35:13,467
POTENTIAL PSYCHIATRIC CONDITIONS

2436
01:35:13,467 --> 01:35:15,803
OVER A LONG PERIOD OF TIME,

2437
01:35:15,803 --> 01:35:17,438
NATURALISTIC PERIOD OF TIME OVER

2438
01:35:17,438 --> 01:35:18,439
WHICH THEY MIGHT EXPERIENCE

2439
01:35:18,439 --> 01:35:20,074
SIGNIFICANT CHANGES IN THEIR

2440
01:35:20,074 --> 01:35:20,941
MENTAL HEALTH.

2441
01:35:20,941 --> 01:35:23,978
SO WE SOMETIMES REFER TO THIS AS

2442
01:35:23,978 --> 01:35:24,845
THE ONE HUNDRED PERSON-YEAR

2443
01:35:24,845 --> 01:35:27,815
STUDY BECAUSE THE IDEA WAS TO

2444
01:35:27,815 --> 01:35:28,682
STUDY ONE HUNDRED INDIVIDUALS

2445
01:35:28,682 --> 01:35:32,086
WHO EACH HAD A CONDITION SUCH AS

2446
01:35:32,086 --> 01:35:34,722
DEPRESSION, BIPOLAR DISORDER OR

2447
01:35:34,722 --> 01:35:35,589
PSYCHOTIC CONDITION AND TRACK

2448
01:35:35,589 --> 01:35:37,758
THEM OVER A PERIODS OF TIME

2449
01:35:37,758 --> 01:35:39,159
LIKELY TO EXPERIENCE SIGNIFICANT

2450
01:35:39,159 --> 01:35:41,061
SHIFTS IN MENTAL STATE AND ASK

2451
01:35:41,061 --> 01:35:44,531
THEM TO REPORT ON THEIR

2452
01:35:44,531 --> 01:35:45,532
SUBJECTIVE EXPERIENCES WITH

2453
01:35:45,532 --> 01:35:48,702
SELF-REPORTS, AND THEN ALSO MEET

2454
01:35:48,702 --> 01:35:51,338
PERIODICALLY WITH CLINICIAN AND

2455
01:35:51,338 --> 01:35:53,207
COMBINE THOSE SELF-REPORTS AND

2456
01:35:53,207 --> 01:35:54,174
CLINICIAN EVALUATIONS WITH

2457
01:35:54,174 --> 01:35:58,679
OBJECTIVE MEASURES WE CAN DERIVE

2458
01:35:58,679 --> 01:36:01,548
FROM WEARABLE DATA, PHONE SENSOR

2459
01:36:01,548 --> 01:36:03,050
DATA, BEHAVIORAL TASKS ON A

2460
01:36:03,050 --> 01:36:03,283
DEVICE.

2461
01:36:03,283 --> 01:36:06,453
AGAIN, THE IDEA HERE IS TO

2462
01:36:06,453 --> 01:36:08,756
UNDERSTAND WHICH OBJECTIVE AND

2463
01:36:08,756 --> 01:36:09,723
SUBJECTIVE MEASURES ARE

2464
01:36:09,723 --> 01:36:11,792
CORRELATING WITH ONE ANOTHER,

2465
01:36:11,792 --> 01:36:16,964
AND TO WHAT EXTENT THOSE

2466
01:36:16,964 --> 01:36:18,632
COHERENT SYSTEMS ARE DEPENDENT

2467
01:36:18,632 --> 01:36:20,034
PERSONALIZED VERY SPECIFIC TO

2468
01:36:20,034 --> 01:36:23,170
THE INDIVIDUAL, OR MORE

2469
01:36:23,170 --> 01:36:24,371
GENERALIZED ACROSS INDIVIDUALS.

2470
01:36:24,371 --> 01:36:26,340
I THINK AS YOU'LL SEE AS PART OF

2471
01:36:26,340 --> 01:36:31,278
THIS WORK THIS BRINGS UP ETHICAL

2472
01:36:31,278 --> 01:36:32,246
CONSIDERATIONS AS WE MOVE

2473
01:36:32,246 --> 01:36:36,417
TECHNOLOGIES INTO A CLINICAL

2474
01:36:36,417 --> 01:36:36,617
REALM.

2475
01:36:36,617 --> 01:36:40,788
SO, IN THIS STUDY WE HAD BROAD

2476
01:36:40,788 --> 01:36:43,323
ELIGIBILITY, PATIENTS COULD

2477
01:36:43,323 --> 01:36:46,894
BE -- ANY ADULT, THEY HAD TO OWN

2478
01:36:46,894 --> 01:36:47,895
A SMARTPHONE, HISTORY OF

2479
01:36:47,895 --> 01:36:51,165
FLUCTUATION IN MOOD OR

2480
01:36:51,165 --> 01:36:52,166
COGNITION.

2481
01:36:52,166 --> 01:36:54,668
SO I'LL TAKE YOU THROUGH THE

2482
01:36:54,668 --> 01:36:56,003
INDIVIDUAL PARTICIPANT TIMELINE.

2483
01:36:56,003 --> 01:36:58,405
PEOPLE WERE FOLD OVER A YEAR,

2484
01:36:58,405 --> 01:37:02,476
USING THIS PARTICULAR WEARABLE

2485
01:37:02,476 --> 01:37:05,512
DEVICE, PHONE APPLICATION ON

2486
01:37:05,512 --> 01:37:06,714
THEIR PHONE, PULLING CONTINUOUS

2487
01:37:06,714 --> 01:37:11,218
DATA FROM THE WEARABLE AND PHONE

2488
01:37:11,218 --> 01:37:11,452
SENSORS.

2489
01:37:11,452 --> 01:37:12,853
PHONE SENSORS INCLUDED LOCATION

2490
01:37:12,853 --> 01:37:16,890
DATA, HOW THEY USED THEIR PHONE,

2491
01:37:16,890 --> 01:37:19,727
AS WELL AS CALL AND TEXT

2492
01:37:19,727 --> 01:37:23,664
METADATA, NOT THE CONTEXT OF THE

2493
01:37:23,664 --> 01:37:25,332
CALLS, BUT SIGNIFICANT OR

2494
01:37:25,332 --> 01:37:28,736
NUMBERS THAT THEY ARE CALLING OR

2495
01:37:28,736 --> 01:37:30,137
TEXTING WITH FREQUENTLY.

2496
01:37:30,137 --> 01:37:32,339
THEY ALSO WERE PERFORMING DAILY

2497
01:37:32,339 --> 01:37:34,108
ASSESSMENTS ON THEIR PHONE IN

2498
01:37:34,108 --> 01:37:36,376
TERMS OF ANSWERS TO SURVEYS,

2499
01:37:36,376 --> 01:37:38,579
PROVIDED DAILY VOICE DIARIES TO

2500
01:37:38,579 --> 01:37:40,214
PROVIDE MORE COLOR ON WHAT

2501
01:37:40,214 --> 01:37:41,882
EXACTLY WAS GOING ON WITH THEM.

2502
01:37:41,882 --> 01:37:43,951
FINALLY THEY WOULD HAVE PERIODIC

2503
01:37:43,951 --> 01:37:45,919
CLINICAL ASSESSMENTS THE WAY YOU

2504
01:37:45,919 --> 01:37:47,020
MIGHT EXPERIENCE IN YOU'RE A

2505
01:37:47,020 --> 01:37:48,322
PATIENT FOLLOWING UP WITH A

2506
01:37:48,322 --> 01:37:50,190
CLINICIAN EVERY MONTH OR SO AS

2507
01:37:50,190 --> 01:37:52,593
PART OF A STANDARD CLINICAL

2508
01:37:52,593 --> 01:37:54,661
VISIT, AND AS MENTIONED WE WOULD

2509
01:37:54,661 --> 01:37:58,198
COLLECT BEHAVIORAL DATA TO

2510
01:37:58,198 --> 01:38:02,703
CAPTURE INFORMATION ABOUT THEIR

2511
01:38:02,703 --> 01:38:03,570
COGNITIVE PERFORMANCE.

2512
01:38:03,570 --> 01:38:05,973
WITH THESE RAW DATA TYPES WE CAN

2513
01:38:05,973 --> 01:38:08,075
EXTRACT MANY DIFFERENT FEATURE,

2514
01:38:08,075 --> 01:38:09,276
SOME I'VE ALREADY MENTIONED.

2515
01:38:09,276 --> 01:38:12,012
AGAIN THIS IS A COMBINATION OF

2516
01:38:12,012 --> 01:38:14,081
THINGS WE CAN COLLECT BOTH

2517
01:38:14,081 --> 01:38:15,182
ACTIVELY MEANING THINGS THAT

2518
01:38:15,182 --> 01:38:18,252
PATIENTS HAVE TO RESPOND TO IN A

2519
01:38:18,252 --> 01:38:20,654
WAY THEY ARE AWARE OF BUT ALSO

2520
01:38:20,654 --> 01:38:21,889
PASSIVELY, SO THINGS ALTHOUGH

2521
01:38:21,889 --> 01:38:24,691
PATIENTS CONSENT, THEY ARE NOT

2522
01:38:24,691 --> 01:38:26,760
ACTUALLY AWARE OF HOW WE -- WHEN

2523
01:38:26,760 --> 01:38:29,596
AND HOW WE'RE PULLING IN RAW

2524
01:38:29,596 --> 01:38:31,899
DATA OR PASSIVE SENSING DATA.

2525
01:38:31,899 --> 01:38:35,335
I WON'T GO INTO IT TOO MUCH

2526
01:38:35,335 --> 01:38:36,970
TODAY, WE THEN MAPPED THESE

2527
01:38:36,970 --> 01:38:40,908
FEATURES INTO A NUMBER OF LATENT

2528
01:38:40,908 --> 01:38:45,712
CONSTRUCTS SUCH AS ENERGY LEVEL,

2529
01:38:45,712 --> 01:38:47,581
SLEEP, PROCESSING SPEED WITH

2530
01:38:47,581 --> 01:38:48,582
POTENTIALLY MECHANISTIC LINKS TO

2531
01:38:48,582 --> 01:38:49,516
OCCURRENCE OF THESE CONDITIONS

2532
01:38:49,516 --> 01:38:52,186
OVER TIME.

2533
01:38:52,186 --> 01:38:56,557
AND SO I'LL TAKE YOU THROUGH A

2534
01:38:56,557 --> 01:39:00,360
COUPLE OF CASE -- SORT OF

2535
01:39:00,360 --> 01:39:01,695
EXAMPLES OF INDIVIDUALS,

2536
01:39:01,695 --> 01:39:03,530
PARTICIPANTS IN THE STUDY TO

2537
01:39:03,530 --> 01:39:04,631
FUEL THE DISCUSSION.

2538
01:39:04,631 --> 01:39:08,135
SO HERE I'M SHOWING EXAMPLE OF A

2539
01:39:08,135 --> 01:39:09,770
SINGLE PARTICIPANT FOLLOWED OVER

2540
01:39:09,770 --> 01:39:14,274
MORE THAN TWO YEARS. .

2541
01:39:14,274 --> 01:39:16,910
SOMEONE WITH REPEATED PERIODS OF

2542
01:39:16,910 --> 01:39:18,245
DENSE DEPRESSION.

2543
01:39:18,245 --> 01:39:19,746
SO, WHAT I'LL HIGHLIGHT HERE,

2544
01:39:19,746 --> 01:39:23,350
THIS IS ONE OF THE FIRST

2545
01:39:23,350 --> 01:39:26,653
EXAMPLES OF WHAT IT'S LIKE IN

2546
01:39:26,653 --> 01:39:29,056
TERMS OF LINKING -- WE ALL HAVE

2547
01:39:29,056 --> 01:39:30,791
IDEAS POTENTIALLY OF WHAT

2548
01:39:30,791 --> 01:39:31,458
SOMEONE WITH DEPRESSION

2549
01:39:31,458 --> 01:39:33,760
EXPERIENCES BUT HERE IS AN

2550
01:39:33,760 --> 01:39:35,295
EXAMPLE OF DAY-TO-DAY CHANGES

2551
01:39:35,295 --> 01:39:36,930
THAT SOMEONE MIGHT SHOW.

2552
01:39:36,930 --> 01:39:38,999
SO YOU CAN SEE AT THE BEGINNING

2553
01:39:38,999 --> 01:39:41,401
OF THE STUDY THIS PARTICULAR

2554
01:39:41,401 --> 01:39:49,209
INDIVIDUAL WAS ENDORSING A LOT

2555
01:39:49,209 --> 01:39:51,144
OF POSITIVE VALENCE REPORTS,

2556
01:39:51,144 --> 01:39:52,479
ENERGETIC, SO ON.

2557
01:39:52,479 --> 01:39:55,215
A HUNDRED DAYS IN STARTED

2558
01:39:55,215 --> 01:39:57,851
REPORTING FEELING VERY NEGATIVE,

2559
01:39:57,851 --> 01:39:59,920
IRRITABLE, ASHAMED.

2560
01:39:59,920 --> 01:40:03,023
YOU CAN SEE THE NATURAL COURSE

2561
01:40:03,023 --> 01:40:04,324
OF DEPRESSION IN THIS

2562
01:40:04,324 --> 01:40:05,192
INDIVIDUAL.

2563
01:40:05,192 --> 01:40:08,595
YOU CAN SEE AFTER ABOUT DAY 200

2564
01:40:08,595 --> 01:40:09,930
SHE SHIFTS BACK TO POSITIVE

2565
01:40:09,930 --> 01:40:10,998
STATE, AGAIN TO NEGATIVE STATE,

2566
01:40:10,998 --> 01:40:12,432
SO ON.

2567
01:40:12,432 --> 01:40:13,634
YOU CAN SEE EPISODES BECOME

2568
01:40:13,634 --> 01:40:16,703
SHORTER OVER THE COURSE OF THE

2569
01:40:16,703 --> 01:40:17,237
STUDY.

2570
01:40:17,237 --> 01:40:19,306
I WILL MENTION OUR RESEARCH TEAM

2571
01:40:19,306 --> 01:40:21,608
WAS NOT TREATING THIS INDIVIDUAL

2572
01:40:21,608 --> 01:40:23,810
SO THIS WAS AN OBSERVATIONAL

2573
01:40:23,810 --> 01:40:24,344
STUDY.

2574
01:40:24,344 --> 01:40:25,646
SO WE'RE NOT COMMENTING ON THE

2575
01:40:25,646 --> 01:40:27,447
CAUSAL NATURE OF WHY SHE CHANGED

2576
01:40:27,447 --> 01:40:30,817
YOU ABOUT WE CAN NEVERTHELESS

2577
01:40:30,817 --> 01:40:35,088
STUDY HOW THIS INDIVIDUAL

2578
01:40:35,088 --> 01:40:36,390
SUBJECTIVELY AND OBJECTIVELY

2579
01:40:36,390 --> 01:40:37,591
VARIABLES WERE CHANGING.

2580
01:40:37,591 --> 01:40:43,630
WE CAN LINK UP THIS INDIVIDUAL'S

2581
01:40:43,630 --> 01:40:45,632
SELF-REPORT WITH

2582
01:40:45,632 --> 01:40:47,567
CLINICIAN-ADMINISTERED REPORT SO

2583
01:40:47,567 --> 01:40:49,102
MADRAS IS A COMMON DEPRESSION

2584
01:40:49,102 --> 01:40:51,738
RATING SCALE.

2585
01:40:51,738 --> 01:40:55,342
WHEN SHE BECOMES VERY SUBJECTIVE

2586
01:40:55,342 --> 01:40:57,311
DEPRESSED HER MADRS SCORE GOES

2587
01:40:57,311 --> 01:40:59,179
UP, BUT NOT SURPRISINGLY AS

2588
01:40:59,179 --> 01:41:00,948
THESE EPISODES BECOME MORE

2589
01:41:00,948 --> 01:41:02,482
FREQUENT TOWARDS THE LATTER

2590
01:41:02,482 --> 01:41:05,619
PART, WE LOSE THE GRANULARITY IN

2591
01:41:05,619 --> 01:41:15,629
HER MOOD SYMPTOMS BASED ON

2592
01:41:15,629 --> 01:41:18,031
CLINICIAN-RATED SYMPTOMS.

2593
01:41:18,031 --> 01:41:19,666
AS WE LINK REPORTS, I'M SHOWING

2594
01:41:19,666 --> 01:41:22,202
DATA EXTRACTED FROM THE WEARABLE

2595
01:41:22,202 --> 01:41:24,404
SHE WAS USING THROUGHOUT THE

2596
01:41:24,404 --> 01:41:24,738
STUDY.

2597
01:41:24,738 --> 01:41:26,273
AND WHAT WE'RE SEEING IN THESE

2598
01:41:26,273 --> 01:41:29,643
HEAT MAPS IS A REPRESENTATION OF

2599
01:41:29,643 --> 01:41:31,945
ACTIVITY, SO WHEN SHE'S AWAKE

2600
01:41:31,945 --> 01:41:34,014
YOU SEE THESE HOT COLORS, REDS

2601
01:41:34,014 --> 01:41:34,815
AND ORANGES.

2602
01:41:34,815 --> 01:41:37,617
WHEN SHE'S ASLEEP YOU CAN SEE

2603
01:41:37,617 --> 01:41:40,587
THE COOLER COLORS, THE BLUES.

2604
01:41:40,587 --> 01:41:42,122
WHAT'S APPARENT IN GOING THROUGH

2605
01:41:42,122 --> 01:41:44,858
THESE IS THAT WHEN SHE BECOMES

2606
01:41:44,858 --> 01:41:46,393
DEPRESSED, A NUMBER OF CHANGES

2607
01:41:46,393 --> 01:41:49,930
OCCUR IN HER SLEEP AND ACTIVITY.

2608
01:41:49,930 --> 01:41:55,302
ONE, YOU CAN SEE HERE IN THIS

2609
01:41:55,302 --> 01:41:57,471
SECOND PANEL OF THE HEAT MAPS,

2610
01:41:57,471 --> 01:41:59,539
HER SLEEP BECOMES LONGER, SO

2611
01:41:59,539 --> 01:42:01,641
SHE'S WAKING UP CONSIDERABLY

2612
01:42:01,641 --> 01:42:04,244
LATER IN THE MORNING, AS

2613
01:42:04,244 --> 01:42:06,646
COMPARED WITH FEELING WELL.

2614
01:42:06,646 --> 01:42:08,315
ALSO YOU'LL NOTICE IF YOU CAN

2615
01:42:08,315 --> 01:42:10,050
SEE THE COLORS HERE THAT WHILE

2616
01:42:10,050 --> 01:42:11,918
SHE'S AWAKE SHE'S ACTUALLY

2617
01:42:11,918 --> 01:42:13,887
CONSIDERABLY LESS ACTIVE DURING

2618
01:42:13,887 --> 01:42:15,989
HER DAY THAN WHEN FEELING WELL.

2619
01:42:15,989 --> 01:42:18,492
THIS AGAIN CONFORMS TO OUR

2620
01:42:18,492 --> 01:42:19,159
CLINICAL UNDERSTANDING OF

2621
01:42:19,159 --> 01:42:21,561
DEPRESSION WHICH IS THE IDEA OF

2622
01:42:21,561 --> 01:42:23,430
PSYCHOMOTOR SLOWING SO WHEN

2623
01:42:23,430 --> 01:42:24,931
INDIVIDUALS BECOME PROFOUNDLY

2624
01:42:24,931 --> 01:42:27,667
DEPRESSED THEY OFTEN HAVE VERY

2625
01:42:27,667 --> 01:42:29,302
REDUCED LEVELS OF ENERGY AND

2626
01:42:29,302 --> 01:42:31,738
JUST SIMPLY DON'T MOVE AS MUCH.

2627
01:42:31,738 --> 01:42:34,574
THEY MAY APPEAR TO SPEAK SLOWER,

2628
01:42:34,574 --> 01:42:36,743
MAY FEEL SLUGGISH DESPITE AS IN

2629
01:42:36,743 --> 01:42:38,512
THIS WOMAN'S CASE DESPITE

2630
01:42:38,512 --> 01:42:40,680
SLEEPING LONGER PERIODS OF TIME.

2631
01:42:40,680 --> 01:42:43,550
SO FROM THESE KINDS OF DATA WE

2632
01:42:43,550 --> 01:42:47,154
CAN THEN EXTRACT LATENT

2633
01:42:47,154 --> 01:42:50,023
CONSTRUCTS SUCH AS OVERALL LEVEL

2634
01:42:50,023 --> 01:42:53,193
OF ENERGY THAT THIS PERSON'S

2635
01:42:53,193 --> 01:42:55,495
MANIFESTING, SO WE CAN TAKE

2636
01:42:55,495 --> 01:42:57,030
SIMPLE DIFFERENCE BETWEEN HER

2637
01:42:57,030 --> 01:42:58,331
SLEEPING ACTIVITY AND WAKING

2638
01:42:58,331 --> 01:43:00,634
ACTIVITY TO COMPUTE ESSENTIALLY

2639
01:43:00,634 --> 01:43:02,035
LIKE A DAY-BY-DAY ENERGY

2640
01:43:02,035 --> 01:43:04,104
BALANCE, SO FOR INSTANCE IF

2641
01:43:04,104 --> 01:43:06,006
YOU'RE SLEEPING A LOT AND NOT

2642
01:43:06,006 --> 01:43:07,641
MOVING VERY MUCH, NORMALLY THAT

2643
01:43:07,641 --> 01:43:10,911
WOULD MEAN YOU HAVE AN EXCESS OF

2644
01:43:10,911 --> 01:43:12,679
ENERGY, WHEREAS IF YOU'RE NOT

2645
01:43:12,679 --> 01:43:14,314
SLEEPING VERY MUCH AND YOU ARE

2646
01:43:14,314 --> 01:43:16,049
VERY ACTIVE, THAT WOULD TEND TO

2647
01:43:16,049 --> 01:43:18,218
BUILD UP A DEFICIT IN YOUR

2648
01:43:18,218 --> 01:43:18,919
ENERGY BALANCE.

2649
01:43:18,919 --> 01:43:21,188
AND SO BASED ON THIS WE CAN

2650
01:43:21,188 --> 01:43:23,623
COMPUTE THIS LATENT ENERGY LEVEL

2651
01:43:23,623 --> 01:43:25,692
OR ENERGY BALANCE WHICH

2652
01:43:25,692 --> 01:43:27,561
INTERESTINGLY WHEN WE DO THIS IT

2653
01:43:27,561 --> 01:43:29,329
SHOWS THERE ARE THESE INFLECTION

2654
01:43:29,329 --> 01:43:31,498
POINTS IN HER ENERGY LEVEL BASED

2655
01:43:31,498 --> 01:43:34,668
EXCLUSIVELY ON THE ACTIVITY AND

2656
01:43:34,668 --> 01:43:35,769
SLEEP PATTERNS BUT INTERESTINGLY

2657
01:43:35,769 --> 01:43:37,804
ARE MATCHED CLOSELY WITH THE

2658
01:43:37,804 --> 01:43:41,608
SUBJECTIVE CHANGE SO WHEN YOU

2659
01:43:41,608 --> 01:43:43,844
SEE INFLECTION POINT IN ENERGY

2660
01:43:43,844 --> 01:43:46,346
BALANCE WE'VE COMPUTED BASED ON

2661
01:43:46,346 --> 01:43:47,881
ACTIVITY, IT CORRESPONDS

2662
01:43:47,881 --> 01:43:50,717
DIRECTLY WITH WHEN THIS PERSON

2663
01:43:50,717 --> 01:43:52,152
WAS EXPERIENCING THIS SUBJECTIVE

2664
01:43:52,152 --> 01:43:53,687
CHANGE.

2665
01:43:53,687 --> 01:43:56,957
YOU CAN IMAGINE THAT WHILE THIS

2666
01:43:56,957 --> 01:43:58,625
IS STILL VERY MUCH THEORETICAL,

2667
01:43:58,625 --> 01:44:01,027
YOU CAN IMAGINE THIS COULD BE

2668
01:44:01,027 --> 01:44:02,996
USED EVENTUALLY TO PREDICT WHEN

2669
01:44:02,996 --> 01:44:07,367
SOMEONE IS ABOUT TO ENCOUNTER A

2670
01:44:07,367 --> 01:44:08,902
CHANGE IN THEIR SUBJECTIVE

2671
01:44:08,902 --> 01:44:11,071
REPORTS AND POTENTIALLY USE THAT

2672
01:44:11,071 --> 01:44:14,141
TO PROACTIVELY TREAT SOMEONE FOR

2673
01:44:14,141 --> 01:44:16,776
DEPRESSION.

2674
01:44:16,776 --> 01:44:19,079
SO ANOTHER SOURCE OF DATA WHICH

2675
01:44:19,079 --> 01:44:21,248
WE THINK IS PROFOUNDLY USEFUL IN

2676
01:44:21,248 --> 01:44:22,682
THE COURSE OF EVALUATING

2677
01:44:22,682 --> 01:44:24,017
PATIENTS WITH MENTAL HEALTH

2678
01:44:24,017 --> 01:44:25,886
CONDITIONS IS LOCATION.

2679
01:44:25,886 --> 01:44:28,922
WE ALL MOVE THROUGH OUR LIVES

2680
01:44:28,922 --> 01:44:30,557
AND WHERE WE GO OFTEN DETERMINES

2681
01:44:30,557 --> 01:44:32,659
NOT JUST WHO WE'RE WITH BUT WHAT

2682
01:44:32,659 --> 01:44:35,362
KIND OF ACTIVITIES WE'RE

2683
01:44:35,362 --> 01:44:35,695
EXPERIENCING.

2684
01:44:35,695 --> 01:44:37,998
AND SO AS YOU MIGHT EXPECT WHEN

2685
01:44:37,998 --> 01:44:39,766
WE BEGIN TO MEASURE PEOPLE'S

2686
01:44:39,766 --> 01:44:40,967
LOCATION IN REAL TIME THAT

2687
01:44:40,967 --> 01:44:42,936
BRINGS UP A HOST OF POTENTIAL

2688
01:44:42,936 --> 01:44:43,703
ETHICAL ISSUES.

2689
01:44:43,703 --> 01:44:46,339
SO HERE I'M SHOWING ONE EXAMPLE

2690
01:44:46,339 --> 01:44:48,408
OF A PARTICIPANT WHO WAS

2691
01:44:48,408 --> 01:44:50,810
FOLLOWED FOR OVER TWO YEARS WITH

2692
01:44:50,810 --> 01:44:52,812
ALL THESE SAME MULTI-MODAL

2693
01:44:52,812 --> 01:44:55,515
MEASUREMENTS, AND IN THIS

2694
01:44:55,515 --> 01:44:56,716
EXAMPLE WHAT I'M SHOWING IS,

2695
01:44:56,716 --> 01:44:58,952
AGAIN, AS WITH THE PRIOR PLOTS,

2696
01:44:58,952 --> 01:45:01,321
WE'RE SEEING A 24-HOUR PERIOD OF

2697
01:45:01,321 --> 01:45:03,857
THIS PERSON'S LOCATION.

2698
01:45:03,857 --> 01:45:06,793
AND THEN EACH COLUMN SHOWS A

2699
01:45:06,793 --> 01:45:08,128
SINGLE DAY OF DATA.

2700
01:45:08,128 --> 01:45:09,863
WE'RE SHOWING HERE THE GRAY

2701
01:45:09,863 --> 01:45:11,164
LOCATION IS THE MOST COMMON

2702
01:45:11,164 --> 01:45:12,599
LOCATION THAT THIS PERSON

2703
01:45:12,599 --> 01:45:14,467
EXPERIENCES, WHICH IS THEIR HOME

2704
01:45:14,467 --> 01:45:15,135
LOCATION.

2705
01:45:15,135 --> 01:45:16,336
AND SO NOT SURPRISINGLY THIS

2706
01:45:16,336 --> 01:45:18,939
HOME LOCATION IS WHERE THEY ARE

2707
01:45:18,939 --> 01:45:19,739
SPENDING MOST NIGHTS.

2708
01:45:19,739 --> 01:45:22,075
IT'S WHERE THEY ARE SLEEPING.

2709
01:45:22,075 --> 01:45:24,244
YOU'LL NOTICE A NUMBER OF OTHER

2710
01:45:24,244 --> 01:45:26,846
LOCATIONS WHICH ARE -- WE CAN

2711
01:45:26,846 --> 01:45:27,948
COMPUTE DIRECTLY FROM GEOSPATIAL

2712
01:45:27,948 --> 01:45:28,815
DATA.

2713
01:45:28,815 --> 01:45:30,784
SO THERE'S A NUMBER OF

2714
01:45:30,784 --> 01:45:31,985
INTERESTING LOCATIONS THAT COME

2715
01:45:31,985 --> 01:45:34,187
OUT OF THIS PERSON'S LOCATION

2716
01:45:34,187 --> 01:45:34,521
TRACES.

2717
01:45:34,521 --> 01:45:37,157
ONE IS YOU'LL SEE THIS PATTERN

2718
01:45:37,157 --> 01:45:40,860
OF BLUE THAT WE SEE, SO A FEW

2719
01:45:40,860 --> 01:45:43,496
WEEKS INTO THE STUDY THIS PERSON

2720
01:45:43,496 --> 01:45:45,565
GOES TO ANOTHER LOCATION AND

2721
01:45:45,565 --> 01:45:47,234
STAYS THERE, SLEEPS THERE FOR

2722
01:45:47,234 --> 01:45:49,836
AROUND TEN DAYS, AND THEY GO

2723
01:45:49,836 --> 01:45:51,972
BACK TO THAT SAME LOCATION,

2724
01:45:51,972 --> 01:45:53,473
SEVERAL MONTHS LATER, AND YOU

2725
01:45:53,473 --> 01:45:56,309
CAN SEE THAT SAME LOCATION, THE

2726
01:45:56,309 --> 01:46:01,348
BLUE LOCATION, IS OCCURRING

2727
01:46:01,348 --> 01:46:02,782
SPORADICALLY THROUGH THE TIME

2728
01:46:02,782 --> 01:46:03,016
SERIES.

2729
01:46:03,016 --> 01:46:05,185
I WOULD NORMALLY OPEN UP IN CASE

2730
01:46:05,185 --> 01:46:07,487
ANYONE HAS GUESSES ABOUT THAT

2731
01:46:07,487 --> 01:46:08,822
LOCATION MIGHT REPRESENT, YOU

2732
01:46:08,822 --> 01:46:10,223
MAY HAVE THOUGHT ABOUT CLINICAL

2733
01:46:10,223 --> 01:46:11,992
CONTEXT WHAT A PLACE LIKE THIS

2734
01:46:11,992 --> 01:46:16,696
MIGHT MEAN FOR A PATIENT LIKE

2735
01:46:16,696 --> 01:46:16,896
THIS.

2736
01:46:16,896 --> 01:46:19,666
I'LL JUST SAY THIS IS A

2737
01:46:19,666 --> 01:46:21,101
HOSPITALIZATION, SO WE CAN

2738
01:46:21,101 --> 01:46:23,703
ACTUALLY INFER THAT NOT ONLY

2739
01:46:23,703 --> 01:46:24,838
FROM THE EXACT GEOSPATIAL

2740
01:46:24,838 --> 01:46:26,473
LOCATION, THE FACT THAT WE KNOW

2741
01:46:26,473 --> 01:46:28,108
WHERE THEY WERE GOING BUT EVEN

2742
01:46:28,108 --> 01:46:30,277
THE PATTERN OF WHERE THIS PERSON

2743
01:46:30,277 --> 01:46:33,680
IS GOING WE KNOW THAT THIS

2744
01:46:33,680 --> 01:46:35,215
PERSON IS ATTENDING THE HOSPITAL

2745
01:46:35,215 --> 01:46:37,851
DURING THE WEEK FOR THERAPY AND

2746
01:46:37,851 --> 01:46:38,285
CLINIC VISITS.

2747
01:46:38,285 --> 01:46:41,655
WE KNOW THEY ARE GOING TO THEIR

2748
01:46:41,655 --> 01:46:42,889
ALCOHOLICS ANONYMOUS MEETINGS IN

2749
01:46:42,889 --> 01:46:46,693
THE THINKS, AT THAT SAME

2750
01:46:46,693 --> 01:46:48,795
LOCATION, AND WE CANNOT JUST

2751
01:46:48,795 --> 01:46:50,530
INFER BUT ACTUALLY DEDUCE

2752
01:46:50,530 --> 01:46:51,831
DIRECTLY FROM THE GEOSPATIAL

2753
01:46:51,831 --> 01:46:53,933
DATA THAT THIS PERSON WAS AT THE

2754
01:46:53,933 --> 01:46:57,203
HOSPITAL AND WAS HOSPITALIZED

2755
01:46:57,203 --> 01:46:59,572
DURING THE TWO PERIODS.

2756
01:46:59,572 --> 01:47:00,840
THE NEXT LOCATION, SET OF

2757
01:47:00,840 --> 01:47:02,142
LOCATIONS TO DRAW YOUR ATTENTION

2758
01:47:02,142 --> 01:47:05,011
TO WAS THE TEAL LOCATIONS AT THE

2759
01:47:05,011 --> 01:47:05,612
BEGINNING.

2760
01:47:05,612 --> 01:47:08,615
SO, THIS IS A LOCATION, THIS

2761
01:47:08,615 --> 01:47:11,251
PERSON IS SPENDING TIME AT

2762
01:47:11,251 --> 01:47:12,118
PERIODICALLY THROUGH THE WEEK.

2763
01:47:12,118 --> 01:47:14,654
AND IF I SHOW YOU THIS AS A

2764
01:47:14,654 --> 01:47:16,523
FUNCTION OF DAY OF WEEK, WE CAN

2765
01:47:16,523 --> 01:47:18,024
SEE THIS IS A LOCATION THIS

2766
01:47:18,024 --> 01:47:20,126
PERSON IS SPENDING THEIR TIME ON

2767
01:47:20,126 --> 01:47:21,995
MONDAYS AND FRIDAYS FOR THE

2768
01:47:21,995 --> 01:47:24,397
FIRST PART OF THE STUDY, AND

2769
01:47:24,397 --> 01:47:26,700
THEN SORT OF DROPS OFF IN THE

2770
01:47:26,700 --> 01:47:28,902
MIDDLE OF THE STUDY, AND SO

2771
01:47:28,902 --> 01:47:30,770
BECAUSE OF THE TIMING THE PERSON

2772
01:47:30,770 --> 01:47:34,607
GETS TO THAT LOCATION AROUND

2773
01:47:34,607 --> 01:47:37,010
9:00 A.M., LEAVES AT 4:00 P.M.,

2774
01:47:37,010 --> 01:47:41,815
NOT HARD TO IMAGINE THIS IS A

2775
01:47:41,815 --> 01:47:44,884
PART-TIME WORK LOCATION.

2776
01:47:44,884 --> 01:47:47,087
WE CAN DEDUCE WHEN AND HOW OFTEN

2777
01:47:47,087 --> 01:47:48,355
THIS PERSON MAKES IT TO WORK,

2778
01:47:48,355 --> 01:47:49,689
CAN TELL WHEN THEY ARE LATE TO

2779
01:47:49,689 --> 01:47:51,791
WORK, WHEN THEY ARE MISSING WORK

2780
01:47:51,791 --> 01:47:52,659
FOR SOME REASON.

2781
01:47:52,659 --> 01:47:55,261
WE CAN ALSO SEE WHEN THIS PERSON

2782
01:47:55,261 --> 01:47:57,897
STOPS GOING TO WORK, SO WE CAN

2783
01:47:57,897 --> 01:47:58,898
DIRECTLY DETERMINE TIMES WHEN

2784
01:47:58,898 --> 01:48:01,935
THIS PERSON IS NOT ENGAGED IN

2785
01:48:01,935 --> 01:48:03,169
WORK OR SCHOOL.

2786
01:48:03,169 --> 01:48:05,038
WE CAN ALSO SEE THAT THERE'S

2787
01:48:05,038 --> 01:48:06,573
THIS PERIOD OF TIME IN THE

2788
01:48:06,573 --> 01:48:08,007
MIDDLE WHEN THIS PERSON STOPS

2789
01:48:08,007 --> 01:48:09,776
GOING TO WORK AT ALL AND THEY

2790
01:48:09,776 --> 01:48:12,379
ARE GOING BACK AND FORTH BETWEEN

2791
01:48:12,379 --> 01:48:14,147
THE HOME LOCATION AND THIS

2792
01:48:14,147 --> 01:48:15,014
HOSPITAL LOCATION.

2793
01:48:15,014 --> 01:48:16,783
AND YOU CAN SEE TOWARDS THE END

2794
01:48:16,783 --> 01:48:18,952
OF THE STUDY THIS PERSON STARTS

2795
01:48:18,952 --> 01:48:21,187
TO GO TO A DIFFERENT PART-TIME

2796
01:48:21,187 --> 01:48:25,525
LOCATION, AS SHOWN IN THE BLACK.

2797
01:48:25,525 --> 01:48:27,060
AND SO THE FINAL SET OF

2798
01:48:27,060 --> 01:48:28,461
LOCATIONS I WANT TO DRAW YOUR

2799
01:48:28,461 --> 01:48:29,562
ATTENTION TO IS THIS ORANGE

2800
01:48:29,562 --> 01:48:31,531
LOCATION.

2801
01:48:31,531 --> 01:48:33,266
THIS IS A LOCATION THAT ALSO

2802
01:48:33,266 --> 01:48:35,702
THIS PERSON IS GOING TO,

2803
01:48:35,702 --> 01:48:38,438
SPENDING THE NIGHT AT, UNLIKE

2804
01:48:38,438 --> 01:48:39,906
THE HOSPITALIZATION LOCATION

2805
01:48:39,906 --> 01:48:41,508
THEY ARE SPENDING A FEW DAYS AT

2806
01:48:41,508 --> 01:48:43,476
A TIME AND GOING BACK TO THE

2807
01:48:43,476 --> 01:48:46,012
HOME LOCATION AND GOING BACK TO

2808
01:48:46,012 --> 01:48:46,880
THAT OTHER LOCATION.

2809
01:48:46,880 --> 01:48:48,314
AND SO AGAIN YOU MIGHT HAVE

2810
01:48:48,314 --> 01:48:50,150
THOUGHTS ABOUT WHAT THAT

2811
01:48:50,150 --> 01:48:51,384
LOCATION MIGHT REPRESENT.

2812
01:48:51,384 --> 01:48:53,653
BUT THIS IS A SIGNIFICANT OTHER,

2813
01:48:53,653 --> 01:48:56,623
SO THIS PERSON DEVELOPS A

2814
01:48:56,623 --> 01:48:57,490
SIGNIFICANT OTHER RELATIONSHIP,

2815
01:48:57,490 --> 01:48:58,425
WHICH INITIALLY THEY ARE

2816
01:48:58,425 --> 01:49:01,227
SPENDING THAT TIME IN THE

2817
01:49:01,227 --> 01:49:02,529
EVENINGS, AND THEN EVENTUALLY

2818
01:49:02,529 --> 01:49:04,164
THAT RELATIONSHIP GRADUATES TO

2819
01:49:04,164 --> 01:49:05,298
BEING MAYBE MORE SERIOUS

2820
01:49:05,298 --> 01:49:06,699
RELATIONSHIP, SO WE CAN SEE THAT

2821
01:49:06,699 --> 01:49:08,334
PERSON GOING TO THE SIGNIFICANT

2822
01:49:08,334 --> 01:49:09,769
OTHER RELATIONSHIP FOR A NUMBER

2823
01:49:09,769 --> 01:49:13,673
OF WEEKS, AND WE CAN SEE PERIODS

2824
01:49:13,673 --> 01:49:15,208
OF TIME WHEN THE RELATIONSHIP

2825
01:49:15,208 --> 01:49:16,409
MAYBE GOES AWAY FOR A PERIOD OF

2826
01:49:16,409 --> 01:49:18,178
TIME AND TOWARDS THE END OF THE

2827
01:49:18,178 --> 01:49:24,651
STUDY MAYBE PICKS UP AGAIN.

2828
01:49:24,651 --> 01:49:26,052
AGAIN SIMPLY BY REVIEWING

2829
01:49:26,052 --> 01:49:27,720
LOCATION TRACES WE CAN ACTUALLY

2830
01:49:27,720 --> 01:49:30,557
BEGIN TO SEE WHAT THIS PERSON'S

2831
01:49:30,557 --> 01:49:31,658
LIFE MIGHT BE LIKE.

2832
01:49:31,658 --> 01:49:34,594
THE WAY WE CAN THINK ABOUT THIS

2833
01:49:34,594 --> 01:49:40,166
IS THAT THIS IS A WAY OF

2834
01:49:40,166 --> 01:49:42,135
COLLECTING AUTOMATED MINIMALLY

2835
01:49:42,135 --> 01:49:42,802
BURDENSOME CONTINUOUS

2836
01:49:42,802 --> 01:49:46,639
ASSESSMENTS OF COMMUNITY AND

2837
01:49:46,639 --> 01:49:51,911
SOCIAL FUNCTIONINGS IN

2838
01:49:51,911 --> 01:49:53,580
INDIVIDUALS WITH SEVERE MENTAL

2839
01:49:53,580 --> 01:49:59,686
ILLNESS OR SEVERE  PSYCHIATRIC

2840
01:49:59,686 --> 01:50:00,687
CONDITIONS.

2841
01:50:00,687 --> 01:50:02,322
THIS IS SEVEN INDIVIDUALS.

2842
01:50:02,322 --> 01:50:03,623
THIS ONE DIFFERS FROM THE

2843
01:50:03,623 --> 01:50:05,191
INDIVIDUAL BELOW IT WITH A

2844
01:50:05,191 --> 01:50:07,894
SPARSER SET OF LOCATIONS THAT

2845
01:50:07,894 --> 01:50:09,028
THEY VISIT.

2846
01:50:09,028 --> 01:50:11,631
AND SO WE COULD ACTUALLY BE ABLE

2847
01:50:11,631 --> 01:50:14,133
TO TELL REALLY WHAT ANYONE ON

2848
01:50:14,133 --> 01:50:16,002
THIS CALL'S LIFE LOOKS LIKE

2849
01:50:16,002 --> 01:50:17,770
BASED ON THE LOCATIONS THEY

2850
01:50:17,770 --> 01:50:24,010
ATTEND, OVER A PERIOD OF TIME AS

2851
01:50:24,010 --> 01:50:25,211
WELL AS SIGNIFICANT LIFE CHANGES

2852
01:50:25,211 --> 01:50:26,846
SUCH AS THIS PERSON WHO SHOWS

2853
01:50:26,846 --> 01:50:27,947
CHANGES IN WHERE THEY LIVE OVER

2854
01:50:27,947 --> 01:50:34,187
THE COURSE OF THE STUDY.

2855
01:50:34,187 --> 01:50:37,123
BECAUSE WE WERE TALKING ABOUT

2856
01:50:37,123 --> 01:50:38,358
INITIATIVES, WE'VE DONE THE SAME

2857
01:50:38,358 --> 01:50:41,194
WORK WITH INDIVIDUALS WHO HAVE

2858
01:50:41,194 --> 01:50:43,363
IMPLANTED BRAIN RECORDING

2859
01:50:43,363 --> 01:50:43,596
DEVICES.

2860
01:50:43,596 --> 01:50:47,967
THIS WAS A PROJECT PART OF THE

2861
01:50:47,967 --> 01:50:49,168
BRAIN SYNCHRONIZATION EFFORT.

2862
01:50:49,168 --> 01:50:55,708
THIS EFFORT WAS FOLLOWED OVER 18

2863
01:50:55,708 --> 01:50:58,144
MONTHS, SAME SELF-REPORTED AND

2864
01:50:58,144 --> 01:51:01,114
PASSIVE DATA, THEY HAD A DEVICE

2865
01:51:01,114 --> 01:51:02,415
IMPLANTED IN THE VENTRAL

2866
01:51:02,415 --> 01:51:05,485
STRIATUM, AS WELL AS IN THEIR

2867
01:51:05,485 --> 01:51:06,786
MEDIAL FRONTAL CORTEX WHERE WE

2868
01:51:06,786 --> 01:51:08,521
CAN BEGIN TO SEE NOT JUST WHERE

2869
01:51:08,521 --> 01:51:10,823
THEY ARE MOVING AND HOW THEY

2870
01:51:10,823 --> 01:51:12,258
FEEL BUT ALSO GET A

2871
01:51:12,258 --> 01:51:13,126
REPRESENTATION OF WHAT THEIR

2872
01:51:13,126 --> 01:51:16,095
BRAIN IS DOING AT THOSE

2873
01:51:16,095 --> 01:51:17,297
DIFFERENT TIMES.

2874
01:51:17,297 --> 01:51:18,698
AGAIN, THE CHALLENGE AND THE

2875
01:51:18,698 --> 01:51:21,234
OPPORTUNITY WITH THIS IS THAT

2876
01:51:21,234 --> 01:51:23,736
THESE KINDS OF DATA CAN ACTUALLY

2877
01:51:23,736 --> 01:51:27,473
BE DIRECTLY USED TO PREDICT AND

2878
01:51:27,473 --> 01:51:28,841
ESTIMATE CURRENT AND POTENTIALLY

2879
01:51:28,841 --> 01:51:31,177
FUTURE BEHAVIOR SIMPLY ON BASIS

2880
01:51:31,177 --> 01:51:34,447
OF RECOGNIZING PATTERNS AND

2881
01:51:34,447 --> 01:51:36,549
BEING ABLE TO HAVE PATTERNS OF

2882
01:51:36,549 --> 01:51:38,518
SUBJECTIVE AND OBJECTIVE DATA

2883
01:51:38,518 --> 01:51:40,386
AND LINK THOSE UP WITH NEURAL

2884
01:51:40,386 --> 01:51:43,222
DATA TO BE ABLE TO HAVE FULL

2885
01:51:43,222 --> 01:51:45,858
APPRECIATION OF NEURAL SYSTEMS

2886
01:51:45,858 --> 01:51:47,493
DRIVING THOSE BEHAVIORAL

2887
01:51:47,493 --> 01:51:47,827
CHANGES.

2888
01:51:47,827 --> 01:51:50,463
SO SIGNIFICANT PART OF THIS WORK

2889
01:51:50,463 --> 01:51:57,236
IS REALLY GEARED TOWARDS MOVING

2890
01:51:57,236 --> 01:51:59,105
BEYOND THE CARICATURE-BASED

2891
01:51:59,105 --> 01:51:59,939
ASSESSMENTS OF PSYCHIATRIC

2892
01:51:59,939 --> 01:52:05,878
CONDITIONS, SOMETHING THAT'S

2893
01:52:05,878 --> 01:52:08,715
MUCH MORE COMPUTATIONALLY

2894
01:52:08,715 --> 01:52:09,215
DEDEVELOPED, LEVERAGING

2895
01:52:09,215 --> 01:52:10,049
CONSTRUCTS AND ARTIFICIAL

2896
01:52:10,049 --> 01:52:11,250
INTELLIGENCE TO INFER MENTAL

2897
01:52:11,250 --> 01:52:12,652
STATES, BOTH CURRENT AND FUTURE

2898
01:52:12,652 --> 01:52:15,622
MENTAL STATES, IN WAYS THAT

2899
01:52:15,622 --> 01:52:16,556
COULD GUIDE TREATMENT PLANNING

2900
01:52:16,556 --> 01:52:20,226
FOR INDIVIDUALS WITH A RANGE OF

2901
01:52:20,226 --> 01:52:20,994
COMPLEX PSYCHIATRIC CONDITIONS.

2902
01:52:20,994 --> 01:52:26,799
AGAIN THIS BRINGS UP A HOST OF

2903
01:52:26,799 --> 01:52:28,635
ETHICAL CONSIDERATIONS, WHICH

2904
01:52:28,635 --> 01:52:30,103
WE'VE DONE A NUMBER OF PRIOR

2905
01:52:30,103 --> 01:52:31,838
SYMPOSIA ON THIS WORK.

2906
01:52:31,838 --> 01:52:36,209
IF YOU'RE INTERESTED, TO READ

2907
01:52:36,209 --> 01:52:38,845
MORE, I CAN DROP THIS,

2908
01:52:38,845 --> 01:52:41,681
POTENTIALLY SHARE THIS LINK WITH

2909
01:52:41,681 --> 01:52:45,985
OTHERS ON THIS VIDEOCAST.

2910
01:52:45,985 --> 01:52:48,821
BUT ALSO DR. SILVERMAN WILL GO

2911
01:52:48,821 --> 01:52:51,224
THROUGH SOME OF THESE IN MORE

2912
01:52:51,224 --> 01:52:53,459
DETAIL IN THE NEXT PRESENTATION.

2913
01:52:53,459 --> 01:52:57,797
>> THANKS, DR. BAKER, A 
GREAT

2914
01:52:57,797 --> 01:52:59,465
OVERVIEW OF DEEP PHENOTYPING IN

2915
01:52:59,465 --> 01:52:59,766
PSYCHIATRY.

2916
01:52:59,766 --> 01:53:02,835
WE'LL HOLD OFF ON QUESTIONS

2917
01:53:02,835 --> 01:53:04,037
UNTIL AFTER DR. SILVERMAN, NOW

2918
01:53:04,037 --> 01:53:07,974
GIVING A SUMMARY OF ETHICAL

2919
01:53:07,974 --> 01:53:10,610
ISSUES. 

2920
01:53:10,610 --> 01:53:12,945
SENIOR IRBs CHAIR OF MASS

2921
01:53:12,945 --> 01:53:15,214
GENERAL BRIGHAM, ALSO THE

2922
01:53:15,214 --> 01:53:17,750
DIRECTOR OF ETHICS FOR MCLEAN

2923
01:53:17,750 --> 01:53:20,353
INSTITUTE OF INDUSTRY, CHAIR OF

2924
01:53:20,353 --> 01:53:24,424
THE EMBRYONIC STEM CELL RESEARCH

2925
01:53:24,424 --> 01:53:25,525
OVERSIGHT SUBCOMMITTEE,

2926
01:53:25,525 --> 01:53:26,726
ASSISTANT PROFESSOR OF

2927
01:53:26,726 --> 01:53:31,030
PSYCHIATRY AT HARVARD, RESEARCH

2928
01:53:31,030 --> 01:53:34,934
ETHICS IN PARTICULAR PERTAINING

2929
01:53:34,934 --> 01:53:36,202
TO RESEARCH CONDUCTED WITH

2930
01:53:36,202 --> 01:53:46,245
VULNERABLE POPULATIONS.

2931
01:53:46,245 --> 01:53:48,414
>> THANK YOU FOR THE

2932
01:53:48,414 --> 01:53:49,849
INTRODUCTION AND HAVING US TALK

2933
01:53:49,849 --> 01:53:51,250
ABOUT THIS, APPRECIATE THE

2934
01:53:51,250 --> 01:53:52,885
OPPORTUNITY, HOPE THIS DOVETAILS

2935
01:53:52,885 --> 01:53:56,989
WITH THE OTHER FURTHER TALKS.

2936
01:53:56,989 --> 01:53:58,858
PSOAS DR. BAKER DESCRIBED, I'M

2937
01:53:58,858 --> 01:54:00,727
GOING TO TALK MORE ABOUT SOME OF

2938
01:54:00,727 --> 01:54:07,834
THE ETHICAL ISSUES THAT COME UP

2939
01:54:07,834 --> 01:54:09,035
WITH DEEP PHENOTYPING, THIS

2940
01:54:09,035 --> 01:54:09,936
PARTICULAR PSYCHIATRIC RESEARCH.

2941
01:54:09,936 --> 01:54:12,538
WE'VE HEARD ABOUT SOME ALREADY

2942
01:54:12,538 --> 01:54:12,739
TODAY.

2943
01:54:12,739 --> 01:54:14,741
BUT JUST FOR THE SAKE OF TIME TO

2944
01:54:14,741 --> 01:54:15,908
HAVE SOME DISCUSSION I'M GOING

2945
01:54:15,908 --> 01:54:18,678
TO KEEP THIS AT A FAIRLY HIGH

2946
01:54:18,678 --> 01:54:18,878
LEVEL.

2947
01:54:18,878 --> 01:54:21,614
BUT THE BIG ISSUES THAT COME UP

2948
01:54:21,614 --> 01:54:27,653
REALLY RELATE TO INFORMED

2949
01:54:27,653 --> 01:54:28,421
CONSENT, PRIVACY AND

2950
01:54:28,421 --> 01:54:29,288
CONFIDENTIALITY, EQUITY

2951
01:54:29,288 --> 01:54:30,823
DIVERSITY AND ACCESS

2952
01:54:30,823 --> 01:54:31,457
PARTICULARLY WITH ALGORITHMIC

2953
01:54:31,457 --> 01:54:34,427
BIAS IN USING A.I. AND ML

2954
01:54:34,427 --> 01:54:36,395
MODELS, RETURN OF INDIVIDUAL

2955
01:54:36,395 --> 01:54:37,697
RESEARCH RESULTS, DUTY TO WARN

2956
01:54:37,697 --> 01:54:41,534
OR DUTY TO REPORT, AMONG MANY

2957
01:54:41,534 --> 01:54:41,768
OTHERS.

2958
01:54:41,768 --> 01:54:43,836
I WANT TO MAKE A QUICK NOTE HERE

2959
01:54:43,836 --> 01:54:45,071
ON TERMINOLOGY.

2960
01:54:45,071 --> 01:54:47,106
WE FOUND AS WE'VE DONE THIS WORK

2961
01:54:47,106 --> 01:54:49,642
THAT THERE'S A LOT OF DIFFERENT

2962
01:54:49,642 --> 01:54:51,711
PHRASES PEOPLE USE THAT I THINK

2963
01:54:51,711 --> 01:54:54,313
FOR OUR PURPOSES TODAY WE CAN

2964
01:54:54,313 --> 01:54:56,415
SAY ARE MEANING THE SAME THING.

2965
01:54:56,415 --> 01:54:58,184
WE USE THE PHRASE DEEP

2966
01:54:58,184 --> 01:54:59,619
PHENOTYPING, YOU SEE IN ONE OF

2967
01:54:59,619 --> 01:55:02,121
THE PAPERS WE USE THE PHRASE

2968
01:55:02,121 --> 01:55:03,022
DIGITAL PHENOTYPING.

2969
01:55:03,022 --> 01:55:05,958
WE ALSO HEAR THE PHRASE AND USE

2970
01:55:05,958 --> 01:55:06,726
THE PHRASE COMPUTATIONAL

2971
01:55:06,726 --> 01:55:07,059
PHENOTYPING.

2972
01:55:07,059 --> 01:55:09,896
THERE ARE PEOPLE WHO MIGHT ARGUE

2973
01:55:09,896 --> 01:55:12,832
THAT ONE OF THESE WORDS, ONE IS

2974
01:55:12,832 --> 01:55:14,600
JUST PHONE DATA OR ONE IS JUST A

2975
01:55:14,600 --> 01:55:19,539
DIFFERENT TYPE OF DATA BUT FOR

2976
01:55:19,539 --> 01:55:21,841
THE PURPOSE OF ETHICS TALKS

2977
01:55:21,841 --> 01:55:27,346
THINK OF THESE AS FAIRLY

2978
01:55:27,346 --> 01:55:28,281
INTERCHANGEABLE.

2979
01:55:28,281 --> 01:55:31,617
WE'VE WRITTEN A COUPLE PAPERS

2980
01:55:31,617 --> 01:55:33,152
ABOUT THIS, REALLY HAVE HAD THE

2981
01:55:33,152 --> 01:55:34,787
GREAT PLEASURE OF WORKING WITH

2982
01:55:34,787 --> 01:55:39,392
DR. BAKER AND ALSO CLOSELY WITH

2983
01:55:39,392 --> 01:55:41,260
FRANCIS SHEN WHO YOU MAY KNOW,

2984
01:55:41,260 --> 01:55:45,097
AND THE PLEASURE OF HAVING DR.

2985
01:55:45,097 --> 01:55:48,568
GRADY AND OTHERS PROBABLY HERE

2986
01:55:48,568 --> 01:55:51,637
TODAY PARTICIPATING IN SYMPOSIA

2987
01:55:51,637 --> 01:55:52,972
AND WORKSHOPS AND FUNDING FROM

2988
01:55:52,972 --> 01:55:53,439
NIH.

2989
01:55:53,439 --> 01:55:55,708
WE'VE PUBLISHED A COUPLE PAPERS.

2990
01:55:55,708 --> 01:55:57,543
THIS IS ONE WHICH CAME OUT OF

2991
01:55:57,543 --> 01:55:59,979
SOME OF OUR WORK THAT WAS TRYING

2992
01:55:59,979 --> 01:56:03,683
TO CREATE ETHICS CHECK LIST FOR

2993
01:56:03,683 --> 01:56:05,852
PEOPLE DOING THIS TYPE OF WORK

2994
01:56:05,852 --> 01:56:08,154
WITH THE IDEA THAT THERE REALLY

2995
01:56:08,154 --> 01:56:09,488
ARE NOT NECESSARILY RIGHT OR

2996
01:56:09,488 --> 01:56:10,790
WRONG ANSWERS AT THIS STAGE OF

2997
01:56:10,790 --> 01:56:13,092
THE GAME BUT THE KEY WAS TRYING

2998
01:56:13,092 --> 01:56:14,627
TO GET RESEARCHERS WHO WANTED TO

2999
01:56:14,627 --> 01:56:17,029
DO THIS TYPE OF WORK TO BE

3000
01:56:17,029 --> 01:56:18,564
THINKING ABOUT THESE ISSUES

3001
01:56:18,564 --> 01:56:20,066
BEFORE STARTING THEIR RESEARCH

3002
01:56:20,066 --> 01:56:20,299
STUDIES.

3003
01:56:20,299 --> 01:56:22,134
SO WE CREATED A SERIES OF

3004
01:56:22,134 --> 01:56:24,470
QUESTIONS IN EACH OF THE DOMAINS

3005
01:56:24,470 --> 01:56:26,038
I MENTIONED EARLIER, I PUT IT UP

3006
01:56:26,038 --> 01:56:26,239
HERE.

3007
01:56:26,239 --> 01:56:28,407
THIS IS ONE OF THE PAPERS THAT

3008
01:56:28,407 --> 01:56:30,376
WAS CIRCULATED TO YOU IN

3009
01:56:30,376 --> 01:56:30,710
ADVANCE.

3010
01:56:30,710 --> 01:56:32,345
I APPRECIATE THERE'S TOO MANY

3011
01:56:32,345 --> 01:56:33,980
WORDS ON THIS SLIDE FOR YOU TO

3012
01:56:33,980 --> 01:56:34,647
POTENTIALLY READ IT.

3013
01:56:34,647 --> 01:56:37,383
I WANT TO SHOW WHAT IT LOOKS

3014
01:56:37,383 --> 01:56:38,684
LIKE, ESSENTIALLY WHAT WE'RE

3015
01:56:38,684 --> 01:56:40,553
ASKING RESEARCHERS TO DO, TO

3016
01:56:40,553 --> 01:56:41,888
THINK ABOUT EACH OF THESE

3017
01:56:41,888 --> 01:56:43,522
DOMAINS AND GO THROUGH A SERIES

3018
01:56:43,522 --> 01:56:46,359
OF QUESTIONS TO SAY YES OR NO,

3019
01:56:46,359 --> 01:56:47,894
HAVE I CONSIDERED THESE ISSUES,

3020
01:56:47,894 --> 01:56:50,730
WITH THE GOAL THAT REALLY YOU

3021
01:56:50,730 --> 01:56:52,832
GET TO YES FOR ALL OF THESE OR

3022
01:56:52,832 --> 01:56:55,001
AT LEAST GET TO PENDING BEFORE

3023
01:56:55,001 --> 01:56:56,402
YOU COULD FEEL COMFORTABLE KIND

3024
01:56:56,402 --> 01:56:58,271
OF PROCEEDING WITH YOUR STUDY IN

3025
01:56:58,271 --> 01:57:00,339
THE MOST ETHICAL WAY POSSIBLE.

3026
01:57:00,339 --> 01:57:02,541
I'LL GO THROUGH SOME IN MORE

3027
01:57:02,541 --> 01:57:03,743
DETAIL TODAY, ALTHOUGH AS I SAID

3028
01:57:03,743 --> 01:57:06,913
WE WON'T HAVE TIME TO TOUCH ON

3029
01:57:06,913 --> 01:57:07,780
EVERY SINGLE ONE.

3030
01:57:07,780 --> 01:57:10,650
I THINK INFORMED CONSENT IS THE

3031
01:57:10,650 --> 01:57:12,051
ONE THAT WE SPEND THE MOST TIME

3032
01:57:12,051 --> 01:57:14,387
TALKING ABOUT, THIS CAME UP IN

3033
01:57:14,387 --> 01:57:16,789
EARLIER TALKS THIS MORNING.

3034
01:57:16,789 --> 01:57:20,092
JUST TO SAY QUICKLY THAT IF WE

3035
01:57:20,092 --> 01:57:22,495
THINK ABOUT THE HISTORY OF

3036
01:57:22,495 --> 01:57:23,462
PSYCHIATRIC ETHICS OR BASIC

3037
01:57:23,462 --> 01:57:26,198
FRAME FOR HOW WE THINK ABOUT

3038
01:57:26,198 --> 01:57:30,236
PSYCHIATRIC CARE OR

3039
01:57:30,236 --> 01:57:32,338
PSYCHOTHERAPY, THERE'S A

3040
01:57:32,338 --> 01:57:33,639
TRADITION IN PSYCHIATRY,

3041
01:57:33,639 --> 01:57:34,407
PSYCHOLOGY, PSYCHOANALYSIS, KIND

3042
01:57:34,407 --> 01:57:37,810
OF OUR TRADITION IN OUR

3043
01:57:37,810 --> 01:57:38,344
PSYCHIATRIC ETHICS, THAT

3044
01:57:38,344 --> 01:57:39,979
BASICALLY ALL OF THE INFORMATION

3045
01:57:39,979 --> 01:57:41,747
THAT WE OBTAIN CAME FROM THE

3046
01:57:41,747 --> 01:57:43,616
PERSON SITTING ACROSS FROM US.

3047
01:57:43,616 --> 01:57:45,885
AND YOU CREATED THIS FRAME OF

3048
01:57:45,885 --> 01:57:47,053
TREATMENT WHERE EVERYTHING WAS

3049
01:57:47,053 --> 01:57:50,589
SUPPOSED TO HAPPEN IN THE ROOM,

3050
01:57:50,589 --> 01:57:52,591
WITH THE PATIENT, AND THAT TIMES

3051
01:57:52,591 --> 01:57:54,226
OF REENACTMENTS YOU WOULD HAVE

3052
01:57:54,226 --> 01:57:57,063
IN THE ROOM WERE REFLECTIVE OF

3053
01:57:57,063 --> 01:57:59,465
THEIR LIFE OUTSIDE THE ROOM, AND

3054
01:57:59,465 --> 01:58:02,535
THAT WAS BY HAVING THOSE

3055
01:58:02,535 --> 01:58:04,837
REENACTMENTS AND TRANSFERENCE

3056
01:58:04,837 --> 01:58:05,838
AND COUNTER-TRANSFERENCE YOU

3057
01:58:05,838 --> 01:58:07,373
WOULD HELP THEM CHANGE THEIR

3058
01:58:07,373 --> 01:58:07,640
BEHAVIOR.

3059
01:58:07,640 --> 01:58:10,409
SO WE STARTED FROM A PLACE WHERE

3060
01:58:10,409 --> 01:58:12,712
BASICALLY ESSENTIALLY THERE WAS

3061
01:58:12,712 --> 01:58:13,813
A FORBIDDANCE OF EXTERNAL

3062
01:58:13,813 --> 01:58:15,781
INFORMATION I WOULD SAY.

3063
01:58:15,781 --> 01:58:17,984
OVER TIME OBVIOUSLY IT BECAME

3064
01:58:17,984 --> 01:58:19,852
RELEVANT AND IMPORTANT TO HAVE

3065
01:58:19,852 --> 01:58:21,287
COLLATERAL INFORMATION ABOUT OUR

3066
01:58:21,287 --> 01:58:21,821
PATIENTS.

3067
01:58:21,821 --> 01:58:24,123
SO WE WOULD OFTEN GET EXTERNAL

3068
01:58:24,123 --> 01:58:26,525
INFORMATION ABOUT OUR PATIENTS.

3069
01:58:26,525 --> 01:58:27,727
AND PSYCHIATRY AND PSYCHOLOGY

3070
01:58:27,727 --> 01:58:28,494
AND PSYCHIATRIC ETHICS HAD TO

3071
01:58:28,494 --> 01:58:29,595
ADAPT TO THAT.

3072
01:58:29,595 --> 01:58:31,797
HEY, WHAT DO YOU DO WHEN SOMEONE

3073
01:58:31,797 --> 01:58:34,100
CALLS YOU AND PROVIDES YOU WITH

3074
01:58:34,100 --> 01:58:35,501
INFORMATION ABOUT A PATIENT OR

3075
01:58:35,501 --> 01:58:38,037
WHAT DO DO YOU, ANY OF US WHO

3076
01:58:38,037 --> 01:58:39,472
TREATED PATIENTS LONG ENOUGH

3077
01:58:39,472 --> 01:58:41,207
HAVE BEEN IN THESE SITUATIONS,

3078
01:58:41,207 --> 01:58:44,043
WHEN YOU GET A FIVE-PAGE LETTER

3079
01:58:44,043 --> 01:58:45,578
FROM A PATIENT'S FAMILY MEMBER

3080
01:58:45,578 --> 01:58:46,545
THAT'S DESCRIBING ALL OF THE

3081
01:58:46,545 --> 01:58:48,948
STUFF YOU DIDN'T KNOW ABOUT WHAT

3082
01:58:48,948 --> 01:58:49,849
WAS GOING ON.

3083
01:58:49,849 --> 01:58:52,385
SO WE HAD TO GRAPPLE WITH HOW DO

3084
01:58:52,385 --> 01:58:54,053
WE DEAL WITH THIS EXTERNAL

3085
01:58:54,053 --> 01:58:54,787
COLLATERAL INFORMATION.

3086
01:58:54,787 --> 01:58:57,523
I SAY THAT JUST TO SAY I THINK

3087
01:58:57,523 --> 01:58:59,158
WHAT WE'RE TALKING ABOUT IN THE

3088
01:58:59,158 --> 01:59:00,593
FRAME OF PSYCHIATRIC CLINICAL

3089
01:59:00,593 --> 01:59:02,228
ETHICS IS REALLY JUST AN

3090
01:59:02,228 --> 01:59:04,096
EXPANSION OF THAT, HOW DO WE

3091
01:59:04,096 --> 01:59:06,932
DEAL WITH ALL OF THIS EXTERNAL

3092
01:59:06,932 --> 01:59:08,034
INFORMATION WE'RE NOW GETTING,

3093
01:59:08,034 --> 01:59:10,336
SOME OF WHICH IS COLLECTED

3094
01:59:10,336 --> 01:59:11,737
PASSIVELY, AS DR. BAKER SAID.

3095
01:59:11,737 --> 01:59:13,072
THE MAJOR DIFFERENCE THAT I

3096
01:59:13,072 --> 01:59:15,374
THINK ABOUT FROM LIKE A PURE

3097
01:59:15,374 --> 01:59:17,443
ETHICS STANDPOINT IS BASICALLY

3098
01:59:17,443 --> 01:59:19,311
THE IDEA OF AWARENESS OF

3099
01:59:19,311 --> 01:59:20,980
INFORMATION SHARING, RIGHT?

3100
01:59:20,980 --> 01:59:23,416
SO FOR MOST OF THESE EXAMPLES

3101
01:59:23,416 --> 01:59:25,418
THAT I'M GIVING IN THE PAST YOU

3102
01:59:25,418 --> 01:59:27,853
WOULD ASK FOR RELEASE OF

3103
01:59:27,853 --> 01:59:28,721
INFORMATION, YOU WOULD TELL

3104
01:59:28,721 --> 01:59:30,456
SOMEONE I'M GOING TO TALK TO

3105
01:59:30,456 --> 01:59:31,590
SO-AND-SO, I'M GOING TO GET

3106
01:59:31,590 --> 01:59:32,892
INFORMATION, MAYBE YOU WOULD TRY

3107
01:59:32,892 --> 01:59:35,728
TO HAVE THE PATIENT THERE TO BE

3108
01:59:35,728 --> 01:59:37,363
ABLE TO HEAR EXACTLY WHAT'S

3109
01:59:37,363 --> 01:59:37,596
SAID.

3110
01:59:37,596 --> 01:59:44,270
THERE'S REALLY A DIRECT HANDOFF.

3111
01:59:44,270 --> 01:59:46,238
IN OUR NEW WORLD THERE'S NO

3112
01:59:46,238 --> 01:59:47,339
DIRECT HANDOFF, INFORMATION

3113
01:59:47,339 --> 01:59:51,377
COMING FROM EVERYWHERE, ALL AT

3114
01:59:51,377 --> 01:59:53,012
ONCE, SO THE ABILITY TO KNOW

3115
01:59:53,012 --> 01:59:54,780
WHAT IS SHARED AND HOW THE

3116
01:59:54,780 --> 01:59:55,981
INFORMATION IS OBTAINED AND HOW

3117
01:59:55,981 --> 01:59:59,452
IT'S BEING USED IS VERY, VERY

3118
01:59:59,452 --> 02:00:00,986
DIFFERENT FROM THE PAST.

3119
02:00:00,986 --> 02:00:02,655
SOME RISKS I THINK THAT COME

3120
02:00:02,655 --> 02:00:04,857
FROM THAT, THE POTENTIAL RISKS

3121
02:00:04,857 --> 02:00:07,793
FROM DEEP PHENOTYPING, IS THE

3122
02:00:07,793 --> 02:00:09,728
IDEA OF DISCLOSING UNWANTED OR

3123
02:00:09,728 --> 02:00:10,396
UNINTENDED INFORMATION.

3124
02:00:10,396 --> 02:00:12,364
SO ANYONE WHO WALKS INTO A

3125
02:00:12,364 --> 02:00:17,503
DOCTOR'S OFFICE OR PSYCHIATRIC'S

3126
02:00:17,503 --> 02:00:19,371
OFFICE OR PSYCHOLOGISTS OFFICES

3127
02:00:19,371 --> 02:00:21,207
MAKE CHOICES CONSCIOUSLY OR

3128
02:00:21,207 --> 02:00:22,541
UNCONSCIOUSLY ABOUT WHAT THEY

3129
02:00:22,541 --> 02:00:23,542
CHOOSE TO SHARE.

3130
02:00:23,542 --> 02:00:25,845
MAYBE THERE'S INFORMATION THEY

3131
02:00:25,845 --> 02:00:27,246
DIDN'T WANT SHARED.

3132
02:00:27,246 --> 02:00:28,781
THERE COULD BE INACCURATE

3133
02:00:28,781 --> 02:00:29,115
INFORMATION.

3134
02:00:29,115 --> 02:00:30,950
YOU KNOW, DR. BAKER DIDN'T TALK

3135
02:00:30,950 --> 02:00:34,453
ABOUT IT AS MUCH BUT OFTEN THIS

3136
02:00:34,453 --> 02:00:37,857
DEEP PHENOTYPING INCLUDES

3137
02:00:37,857 --> 02:00:39,058
PULLING INFORMATION FROM SOCIAL

3138
02:00:39,058 --> 02:00:44,530
MEDIA, OTHER TIMES OF

3139
02:00:44,530 --> 02:00:46,165
INFORMATION, ESPECIALLY NOW MAY

3140
02:00:46,165 --> 02:00:47,700
BE WRONG, OR LIKE THE PUPPIES

3141
02:00:47,700 --> 02:00:53,839
PLAYING IN THE SNOW MAY BE FAKE.

3142
02:00:53,839 --> 02:00:54,707
PERMITTING UNINTENDED OR

3143
02:00:54,707 --> 02:00:55,574
UNWANTED SHARING OF INFORMATION

3144
02:00:55,574 --> 02:00:57,977
SO, AGAIN, YOU MAY NOT ACTUALLY

3145
02:00:57,977 --> 02:00:59,745
WANT CERTAIN INFORMATION SHARED

3146
02:00:59,745 --> 02:01:02,815
IN DIFFERENT WAYS, AND ONCE YOU

3147
02:01:02,815 --> 02:01:04,350
OPEN THIS BOX IT'S HARD TO

3148
02:01:04,350 --> 02:01:06,886
CONTROL ALL OF THAT.

3149
02:01:06,886 --> 02:01:08,220
THERE'S SOME FEAR POTENTIALLY

3150
02:01:08,220 --> 02:01:10,589
ABOUT BECOMING DEPENDENT OR

3151
02:01:10,589 --> 02:01:13,025
OVERRELIANT ON USING THESE TYPES

3152
02:01:13,025 --> 02:01:15,427
OF ELECTRONIC COMMUNICATIONS TO

3153
02:01:15,427 --> 02:01:17,730
COMMUNICATE WITH PROVIDERS THAT

3154
02:01:17,730 --> 02:01:20,466
MAYBE YOU WOULD ASSUME YOUR

3155
02:01:20,466 --> 02:01:21,433
PROVIDER IS GOING TO CATCH

3156
02:01:21,433 --> 02:01:22,768
SOMETHING BECAUSE THEY ARE

3157
02:01:22,768 --> 02:01:24,403
COLLECTING INFORMATION SO YOU

3158
02:01:24,403 --> 02:01:25,271
DON'T NECESSARILY COMMUNICATE

3159
02:01:25,271 --> 02:01:27,473
CONCERNS THE WAY YOU

3160
02:01:27,473 --> 02:01:29,241
TRADITIONALLY WOULD.

3161
02:01:29,241 --> 02:01:30,442
CONFIDENTIALITY AND PRIVACY

3162
02:01:30,442 --> 02:01:32,411
BREACHES OF COURSE, AND THEN

3163
02:01:32,411 --> 02:01:34,280
POTENTIAL TO IMPACT TREATMENT OR

3164
02:01:34,280 --> 02:01:35,581
THERAPEUTIC RELATIONSHIP AND

3165
02:01:35,581 --> 02:01:38,984
WE'LL TALK MORE ABOUT THAT.

3166
02:01:38,984 --> 02:01:42,054
I WILL SAY AS WAS SAID IN MY

3167
02:01:42,054 --> 02:01:43,589
INTRODUCTION, MY DAY JOB IS AS

3168
02:01:43,589 --> 02:01:44,890
AN IRB CHAIR.

3169
02:01:44,890 --> 02:01:46,425
I SPEND TIME THINKING ABOUT

3170
02:01:46,425 --> 02:01:49,495
RISKS AND RESEARCH STUDIES AND

3171
02:01:49,495 --> 02:01:51,263
HOW WE CATEGORIZE RISKS.

3172
02:01:51,263 --> 02:01:53,332
ONE THING I SAY AS I TALK ABOUT

3173
02:01:53,332 --> 02:01:55,734
THESE ISSUES, I BRING UP

3174
02:01:55,734 --> 02:01:57,570
POTENTIAL RISKS AND THEORETICAL

3175
02:01:57,570 --> 02:01:59,138
PROBLEMS, WHAT WE DO GENERALLY I

3176
02:01:59,138 --> 02:02:01,540
WOULD SAY CONSIDER MOST OF WHAT

3177
02:02:01,540 --> 02:02:06,145
WE'RE TALKING ABOUT  TO BE IN TE

3178
02:02:06,145 --> 02:02:08,614
CATEGORY OF MINIMAL RISK.

3179
02:02:08,614 --> 02:02:11,016
THE REASON IS THE FEDERAL

3180
02:02:11,016 --> 02:02:12,251
DEFINITION OF MINIMAL RISK

3181
02:02:12,251 --> 02:02:13,886
POSTED HERE IS THOSE THE SAME

3182
02:02:13,886 --> 02:02:15,287
ENCOUNTERED IN DAILY LIFE.

3183
02:02:15,287 --> 02:02:18,257
AT THIS POINT, MOST OF US ARE

3184
02:02:18,257 --> 02:02:20,125
CARRYING AROUND SMARTPHONE

3185
02:02:20,125 --> 02:02:23,395
DEVICES, OR HAVE OTHER SMART

3186
02:02:23,395 --> 02:02:25,030
APPLIANCES IN YOUR HOUSE THAT

3187
02:02:25,030 --> 02:02:27,666
ARE LISTENING TO YOU 24/7, THAT

3188
02:02:27,666 --> 02:02:29,535
THE IDEA OF COLLECTING THIS DATA

3189
02:02:29,535 --> 02:02:31,503
OF NOT NECESSARILY KNOWING WHERE

3190
02:02:31,503 --> 02:02:33,939
THE DATA IS GOING, AND OF GIVING

3191
02:02:33,939 --> 02:02:36,108
PERMISSION TO USE THAT DATA WHEN

3192
02:02:36,108 --> 02:02:38,210
YOU SIGN, WHEN YOU AGREE TO USE

3193
02:02:38,210 --> 02:02:40,045
A PHONE FOR THE FIRST TIME, YOU

3194
02:02:40,045 --> 02:02:42,047
KNOW, WE ALL DO THAT ON A

3195
02:02:42,047 --> 02:02:43,549
DAY-TO-DAY BASIS WITHOUT I WOULD

3196
02:02:43,549 --> 02:02:46,518
SAY NECESSARILY WHAT WE WOULD

3197
02:02:46,518 --> 02:02:47,953
TRADITIONALLY CONSIDER TRUE

3198
02:02:47,953 --> 02:02:49,021
INFORMED CONSENT.

3199
02:02:49,021 --> 02:02:50,990
SO WE'VE GENERALLY TAKEN THE

3200
02:02:50,990 --> 02:02:52,658
POSITION MOST OF THIS DATA

3201
02:02:52,658 --> 02:02:54,593
COLLECTION AND USE IS IN THE

3202
02:02:54,593 --> 02:02:56,595
REALM OF MINIMAL RISK, DOESN'T

3203
02:02:56,595 --> 02:02:58,030
MEAN WE SHOULD IGNORE IT.

3204
02:02:58,030 --> 02:02:59,131
THAT'S WHY WE'RE HERE TALKING

3205
02:02:59,131 --> 02:03:01,066
ABOUT IT BUT JUST TO FRAME IT

3206
02:03:01,066 --> 02:03:02,835
FOR THE GROUP IN TERMS OF HOW WE

3207
02:03:02,835 --> 02:03:04,903
THINK ABOUT IT.

3208
02:03:04,903 --> 02:03:09,942
FROM OUR CHECKLIST SOME 

3209
02:03:09,942 --> 02:03:15,648
QUESTIONS, HOW CAN WE BE

3210
02:03:15,648 --> 02:03:16,949
MEANINGFUL COMMUNICATE AND

3211
02:03:16,949 --> 02:03:21,186
TRANSPARENT ABOUT ARTIFICIAL

3212
02:03:21,186 --> 02:03:22,288
INTELLIGENCE ANALYSIS AND

3213
02:03:22,288 --> 02:03:23,722
INTERPRETATION, SO WE'RE BAD AT

3214
02:03:23,722 --> 02:03:25,024
INFORMED CONSENT FOR RESEARCH IN

3215
02:03:25,024 --> 02:03:26,659
GENERAL I WOULD SAY.

3216
02:03:26,659 --> 02:03:29,295
YOU HAND SOMEONE A 50-PAGE

3217
02:03:29,295 --> 02:03:29,962
CONSENT FORM.

3218
02:03:29,962 --> 02:03:31,830
THERE IS A LIMIT TO WHAT PEOPLE

3219
02:03:31,830 --> 02:03:33,699
ARE GOING TO BE ABLE TO

3220
02:03:33,699 --> 02:03:33,999
UNDERSTAND.

3221
02:03:33,999 --> 02:03:38,370
AND I THINK THIS TYPE OF WORK

3222
02:03:38,370 --> 02:03:41,006
POSES UNIQUE PROBLEMS FOR

3223
02:03:41,006 --> 02:03:42,675
CONSENT IN THAT PEOPLE MAY NOT

3224
02:03:42,675 --> 02:03:44,176
KNOW WHAT THEIR PHONES ARE

3225
02:03:44,176 --> 02:03:46,045
ALREADY COLLECTING ON THEIR OWN,

3226
02:03:46,045 --> 02:03:46,445
RIGHT?

3227
02:03:46,445 --> 02:03:47,112
WHAT'S ALREADY HAPPENING.

3228
02:03:47,112 --> 02:03:50,082
AND THAT'S ONE OF THE OTHER

3229
02:03:50,082 --> 02:03:50,849
QUESTIONS HERE.

3230
02:03:50,849 --> 02:03:53,686
HOW MUCH ARE WE REQUIRED TO

3231
02:03:53,686 --> 02:03:54,787
PROVIDE BACKGROUND INFORMATION

3232
02:03:54,787 --> 02:03:56,989
ON WHAT'S ALREADY HAPPENING WITH

3233
02:03:56,989 --> 02:03:57,856
SOMEONE'S DATA, IF SOMEBODY

3234
02:03:57,856 --> 02:03:58,957
DOESN'T KNOW IT.

3235
02:03:58,957 --> 02:04:01,593
IN SOME WAYS IN ORDER TO ACCEPT

3236
02:04:01,593 --> 02:04:02,795
MINIMAL RISK DETERMINATION THAT

3237
02:04:02,795 --> 02:04:04,530
I'M TALKING ABOUT SOMEONE WOULD

3238
02:04:04,530 --> 02:04:06,031
HAVE TO UNDERSTAND WHAT'S

3239
02:04:06,031 --> 02:04:07,266
HAPPENING ALREADY, AND SO HOW

3240
02:04:07,266 --> 02:04:10,235
MUCH EDUCATION DO WE HAVE TO

3241
02:04:10,235 --> 02:04:10,469
PROVIDE.

3242
02:04:10,469 --> 02:04:13,205
AND THEN AS DR. BAKER ALLUDED TO

3243
02:04:13,205 --> 02:04:16,809
A LOT IS PASSIVE DATA

3244
02:04:16,809 --> 02:04:17,142
COLLECTION.

3245
02:04:17,142 --> 02:04:18,544
SO WE CONSENT PEOPLE UP FRONT

3246
02:04:18,544 --> 02:04:21,613
BUT WHAT DOES THAT MEAN IN TERMS

3247
02:04:21,613 --> 02:04:24,216
OF ONGOING KNOWLEDGE ABOUT

3248
02:04:24,216 --> 02:04:26,552
WHAT'S BEING COLLECTED.

3249
02:04:26,552 --> 02:04:28,821
WE THINK ABOUT OBVIOUSLY THE

3250
02:04:28,821 --> 02:04:29,822
STUDY POPULATION, STUDY

3251
02:04:29,822 --> 02:04:32,458
POPULATIONS WE WORK WITH OFTEN

3252
02:04:32,458 --> 02:04:33,859
AT TIMES HAVE IMPAIRED DECISION

3253
02:04:33,859 --> 02:04:35,294
MAKING CAPACITY AND WE HAVE TO

3254
02:04:35,294 --> 02:04:37,596
THINK ABOUT HOW WE HANDLE

3255
02:04:37,596 --> 02:04:41,400
CONSENT OVER TIME, WHEN

3256
02:04:41,400 --> 02:04:42,634
SOMEBODY'S CAPACITY MAY

3257
02:04:42,634 --> 02:04:46,338
FLUCTUATE, WHETHER WE NEED

3258
02:04:46,338 --> 02:04:48,674
SURROGATE CONSENT, AND WHAT A

3259
02:04:48,674 --> 02:04:49,942
REASONABLE PERSON WOULD WANT TO

3260
02:04:49,942 --> 02:04:52,044
KNOW AND HOW DO WE REACH THAT.

3261
02:04:52,044 --> 02:04:53,912
I WILL SAY THAT WE HAVE BEEN

3262
02:04:53,912 --> 02:04:54,980
THINKING ABOUT THESE ISSUES FOR

3263
02:04:54,980 --> 02:04:57,383
A WHILE NOW, AND MOST OF THIS IS

3264
02:04:57,383 --> 02:04:58,384
STILL NOT EMPIRIC.

3265
02:04:58,384 --> 02:05:00,452
SO MOST OF THIS IS STILL

3266
02:05:00,452 --> 02:05:03,322
QUESTIONS THAT WE'RE ASKING BUT

3267
02:05:03,322 --> 02:05:05,057
WE DON'T REALLY KNOW EXACTLY

3268
02:05:05,057 --> 02:05:08,026
WHAT THE EVIDENCE BASED ANSWERS

3269
02:05:08,026 --> 02:05:09,962
WOULD BE.

3270
02:05:09,962 --> 02:05:10,763
SOME ADDITIONAL QUESTIONS THESE

3271
02:05:10,763 --> 02:05:12,598
ARE BEYOND THE SCOPE OF WHAT WAS

3272
02:05:12,598 --> 02:05:14,133
IN OUR PAPER BUT THINGS THAT

3273
02:05:14,133 --> 02:05:23,876
COME UP FOR ME, YOU KNOW, HOW WE

3274
02:05:23,876 --> 02:05:25,277
HANDLE EXCULPATORY LANGUAGE,

3275
02:05:25,277 --> 02:05:28,013
OFTEN COLLECTING IN DEEP

3276
02:05:28,013 --> 02:05:30,549
PHENOTYPING REQUIRES USE OF

3277
02:05:30,549 --> 02:05:31,550
THIRD-PARTY APPLICATIONS OR

3278
02:05:31,550 --> 02:05:33,585
DEVICES, ANDS SO THOSE OFTEN

3279
02:05:33,585 --> 02:05:38,323
HAVE TERMS OF USE, END USER

3280
02:05:38,323 --> 02:05:39,625
LICENSE AGREEMENT WAS

3281
02:05:39,625 --> 02:05:41,927
EXCULPATORY LANGUAGE AND WE'RE

3282
02:05:41,927 --> 02:05:45,964
IN A CATCH 22, WE STRUCKEL  --

3283
02:05:45,964 --> 02:05:46,965
STRUGGLE WITH THAT.

3284
02:05:46,965 --> 02:05:49,468
AS DR. BAKER SHOWED WITH THE

3285
02:05:49,468 --> 02:05:51,303
DATA, YOU KNOW, HE CAN TELL YOU

3286
02:05:51,303 --> 02:05:52,538
BASICALLY YOU'RE GOING TO THE

3287
02:05:52,538 --> 02:05:53,639
BATHROOM, THIS PERSON IS GOING

3288
02:05:53,639 --> 02:05:55,908
TO THE BATHROOM AT THIS TIME.

3289
02:05:55,908 --> 02:05:56,809
THIS PERSON IS PROBABLY HAVING

3290
02:05:56,809 --> 02:05:58,343
SEX AT THIS TIME.

3291
02:05:58,343 --> 02:06:00,212
SHOULD WE BE ACTUALLY MORE

3292
02:06:00,212 --> 02:06:03,715
EXPLICIT IN CONSENT FORMS ABOUT

3293
02:06:03,715 --> 02:06:04,583
WHAT WE CAN SEE?

3294
02:06:04,583 --> 02:06:06,018
WHERE THE DATA IS STORED AND HOW

3295
02:06:06,018 --> 02:06:08,420
IT'S GOING TO BE SHARED I THINK

3296
02:06:08,420 --> 02:06:09,721
IS OBVIOUSLY CRITICAL.

3297
02:06:09,721 --> 02:06:13,559
THERE ARE REAL RISKS I THINK TO

3298
02:06:13,559 --> 02:06:14,426
ENSURABILITY AND POTENTIALLY

3299
02:06:14,426 --> 02:06:15,661
CRIMINAL LIABILITY FROM SOME OF

3300
02:06:15,661 --> 02:06:16,628
THIS DATA.

3301
02:06:16,628 --> 02:06:18,297
YOU KNOW, BEING ABLE TO IDENTIFY

3302
02:06:18,297 --> 02:06:21,133
WHERE SOMEONE WAS AT A

3303
02:06:21,133 --> 02:06:22,668
PARTICULAR TIME, BEING ABLE TO

3304
02:06:22,668 --> 02:06:24,203
IDENTIFY THAT SOMEBODY IS

3305
02:06:24,203 --> 02:06:25,504
SMOKING CIGARETTES OR USING

3306
02:06:25,504 --> 02:06:26,505
ALCOHOL, OTHER THINGS THAT MAY

3307
02:06:26,505 --> 02:06:28,574
NOT HAVE PREVIOUSLY BEEN IN THE

3308
02:06:28,574 --> 02:06:29,875
MEDICAL RECORD THAT COULD AFFECT

3309
02:06:29,875 --> 02:06:31,643
INSURABILITY, ET CETERA.

3310
02:06:31,643 --> 02:06:33,812
ONE OF THE BIG ONES IS HOW OFTEN

3311
02:06:33,812 --> 02:06:36,148
DO WE HAVE TO RECONSENT PEOPLE.

3312
02:06:36,148 --> 02:06:39,751
WE KNOW WE HAVE DATA THAT SHOWS

3313
02:06:39,751 --> 02:06:41,954
THAT WHEN YOU ENROLL PEOPLE IN

3314
02:06:41,954 --> 02:06:44,122
THESE RESEARCH STUDIES THEIR

3315
02:06:44,122 --> 02:06:45,657
BEHAVIOR DOES CHANGE BRIEFLY

3316
02:06:45,657 --> 02:06:47,192
BASICALLY BUT THEN IT

3317
02:06:47,192 --> 02:06:47,526
ACCOMMODATES.

3318
02:06:47,526 --> 02:06:50,362
IT GOES BACK TO KIND OF BEING,

3319
02:06:50,362 --> 02:06:51,763
YOU KNOW, NORMAL AFTER PEOPLE

3320
02:06:51,763 --> 02:06:53,432
ADJUST TO A PERIOD OF BEING

3321
02:06:53,432 --> 02:06:53,699
MONITORED.

3322
02:06:53,699 --> 02:06:55,167
WHAT DOES THAT MEAN?

3323
02:06:55,167 --> 02:06:56,802
DOES THAT MEAN THEY FORGOT THE

3324
02:06:56,802 --> 02:07:01,540
DATA IS BEING COLLECTED AND NEED

3325
02:07:01,540 --> 02:07:03,709
TO BE RECONSENTED OR IS THAT A

3326
02:07:03,709 --> 02:07:05,677
NORMAL HUMAN ACCOMMODATION TO

3327
02:07:05,677 --> 02:07:07,446
BEING MONITORED AND THEY

3328
02:07:07,446 --> 02:07:09,181
UNDERSTAND THAT IT'S THERE AND

3329
02:07:09,181 --> 02:07:11,083
THEY HAVE JUST GOTTEN USED TO IT

3330
02:07:11,083 --> 02:07:12,918
AND GOTTEN BACK TO NORMAL

3331
02:07:12,918 --> 02:07:14,319
BEHAVIOR, AND THEN, YOU KNOW,

3332
02:07:14,319 --> 02:07:16,088
WHAT WE DO, YOU KNOW, CHANGES,

3333
02:07:16,088 --> 02:07:18,724
RIGHT? 

3334
02:07:18,724 --> 02:07:20,259
YOUR PHONE WILL TELL YOU I NEED

3335
02:07:20,259 --> 02:07:24,530
TO DO AN UPDATE NOW, RIGHT?

3336
02:07:24,530 --> 02:07:26,398
AND SO SOFTWARE PROGRAMS ARE

3337
02:07:26,398 --> 02:07:27,833
CHANGING, THE WAY WE CAPTURE

3338
02:07:27,833 --> 02:07:29,668
DATA CHANGES, HOW WE INTERPRET

3339
02:07:29,668 --> 02:07:34,172
THE DATA CHANGES, IT'S OFTEN NOT

3340
02:07:34,172 --> 02:07:43,815
PER 

3341
02:07:43,815 --> 02:07:45,784
PERCEPTIBLE TO THE PERSON.

3342
02:07:45,784 --> 02:07:48,854
PRIVACY AND CONFIDENTIAL, THESE

3343
02:07:48,854 --> 02:07:50,822
ARE SOME QUESTIONS WE ASK BEFORE

3344
02:07:50,822 --> 02:07:52,791
STARTING A RESEARCH STUDY.

3345
02:07:52,791 --> 02:07:55,193
HOW DO WE BALANCE ROBUST DATA

3346
02:07:55,193 --> 02:07:56,628
COLLECTION WITH TRYING TO

3347
02:07:56,628 --> 02:08:02,100
PROTECT PRIVACY,  HAVE WE

3348
02:08:02,100 --> 02:08:03,068
CONSULTED WITH SECURITY EXPERTS

3349
02:08:03,068 --> 02:08:04,403
ABOUT DATA FLOW?

3350
02:08:04,403 --> 02:08:07,139
A LARGE PART OF I WOULD SAY

3351
02:08:07,139 --> 02:08:10,409
ETHICS OF PRIVACY AND

3352
02:08:10,409 --> 02:08:11,176
CONFIDENTIALITY IS KNOWING WHERE

3353
02:08:11,176 --> 02:08:13,478
THE DATA IS AT ALL TIMES

3354
02:08:13,478 --> 02:08:13,812
ESSENTIALLY.

3355
02:08:13,812 --> 02:08:16,315
SO WHEN WE CONSULT WITH OUR

3356
02:08:16,315 --> 02:08:17,849
INFORMATION SECURITY TEAMS ABOUT

3357
02:08:17,849 --> 02:08:19,718
THESE RESEARCH STUDIES, THAT'S

3358
02:08:19,718 --> 02:08:20,586
WHAT THEY DO, THEY BASICALLY

3359
02:08:20,586 --> 02:08:21,887
FOLLOW AND SAY THE DATA GOES

3360
02:08:21,887 --> 02:08:24,089
FROM HERE TO HERE TO HERE TO

3361
02:08:24,089 --> 02:08:25,624
HERE, AND SOMETIMES THAT'S LESS

3362
02:08:25,624 --> 02:08:26,925
CLEAR THAN OTHERS, RIGHT?

3363
02:08:26,925 --> 02:08:28,327
BUT REALLY TRYING TO UNDERSTAND

3364
02:08:28,327 --> 02:08:30,529
THAT AS BEST AS POSSIBLE.

3365
02:08:30,529 --> 02:08:32,831
MAKING SURE WE HAVE A POLICY

3366
02:08:32,831 --> 02:08:34,366
ABOUT DE-IDENTIFICATION IF

3367
02:08:34,366 --> 02:08:36,902
THAT'S EVEN POSSIBLE ANYMORE.

3368
02:08:36,902 --> 02:08:39,271
AND PRIVACY THAT'S CONSISTENT

3369
02:08:39,271 --> 02:08:41,773
WITH BEST PRACTICES, THINKING

3370
02:08:41,773 --> 02:08:44,543
ABOUT THIRD PARTIES AND HIPAA

3371
02:08:44,543 --> 02:08:44,977
COMPLIANCE.

3372
02:08:44,977 --> 02:08:47,045
I DON'T SPEND A LOT OF TIMES

3373
02:08:47,045 --> 02:08:47,679
SPEAKING ABOUT REGULATIONS

3374
02:08:47,679 --> 02:08:50,449
BECAUSE TODAY OF TIME BUT

3375
02:08:50,449 --> 02:08:51,850
THERE'S REGULATORY ISSUES

3376
02:08:51,850 --> 02:08:52,384
ESPECIALLY IF DOING

3377
02:08:52,384 --> 02:08:54,052
INTERNATIONAL WORK, YOU GET INTO

3378
02:08:54,052 --> 02:08:55,821
OTHER REGULATIONS AS WELL.

3379
02:08:55,821 --> 02:08:57,222
AND THINKING ABOUT MAKING SURE

3380
02:08:57,222 --> 02:08:59,458
THAT THE RESEARCH IS COMPLIANT

3381
02:08:59,458 --> 02:09:03,028
IS ALSO CRITICAL.

3382
02:09:03,028 --> 02:09:04,229
EQUITY, DIVERSITY, AND ACCESS,

3383
02:09:04,229 --> 02:09:07,499
SO REALLY THINKING ABOUT IF THE

3384
02:09:07,499 --> 02:09:10,302
RESEARCH IS GOING TO REPLICATE

3385
02:09:10,302 --> 02:09:15,407
EXISTING OR GENERATE NEW BIASED

3386
02:09:15,407 --> 02:09:16,375
RESULTS, CONTRIBUTE TO

3387
02:09:16,375 --> 02:09:18,777
INEQUITIES, BASE ON LEGALLY

3388
02:09:18,777 --> 02:09:20,846
PROTECTED CLASS, CRITICALLY

3389
02:09:20,846 --> 02:09:21,113
IMPORTANT.

3390
02:09:21,113 --> 02:09:23,649
WE THINK THAT REALLY ONE OF THE

3391
02:09:23,649 --> 02:09:27,219
KEYS HERE IS AT A VERY EARLY

3392
02:09:27,219 --> 02:09:28,854
STAGE INVOLVING DIVERSE

3393
02:09:28,854 --> 02:09:30,055
COMMUNITY OF STAKEHOLDERS, AND

3394
02:09:30,055 --> 02:09:31,590
SO REALLY ASKING PEOPLE TO PUT

3395
02:09:31,590 --> 02:09:34,426
THEIR NICKEL DOWN ON WHETHER

3396
02:09:34,426 --> 02:09:36,728
THEY HAVE SOUGHT INPUT ON

3397
02:09:36,728 --> 02:09:39,231
RESEARCH DESIGN FROM APPROPRIATE

3398
02:09:39,231 --> 02:09:41,233
STAKEHOLDERS TO ADDRESS EQUITY

3399
02:09:41,233 --> 02:09:42,634
AND JUSTICE CONCERNS.

3400
02:09:42,634 --> 02:09:45,370
MAKING SURE THE RESEARCH PLAN

3401
02:09:45,370 --> 02:09:48,006
HAS ADDRESSED POSSIBLE PROBLEMS

3402
02:09:48,006 --> 02:09:49,975
WITH ACCESS, TO THE

3403
02:09:49,975 --> 02:09:51,276
TECHNOLOGIES, AND MAKING SURE

3404
02:09:51,276 --> 02:09:56,314
THE RESEARCH TEAM IS FULLY

3405
02:09:56,314 --> 02:09:57,616
TRAINED AND KNOWLEDGEABLE ABOUT

3406
02:09:57,616 --> 02:09:58,817
THESE ISSUES.

3407
02:09:58,817 --> 02:10:00,185
ALGORITHMIC BIAS WAS MENTIONED

3408
02:10:00,185 --> 02:10:02,888
EARLIER SO BUT I THINK RELEVANT

3409
02:10:02,888 --> 02:10:04,056
TO MENTION AGAIN, YOU KNOW, ONE

3410
02:10:04,056 --> 02:10:06,591
OF THE BIG QUESTIONS HERE IS

3411
02:10:06,591 --> 02:10:08,860
ESSENTIALLY ARE DATASETS BEING

3412
02:10:08,860 --> 02:10:09,761
USED, ADEQUATE REPRESENTATIVE OF

3413
02:10:09,761 --> 02:10:11,630
THE POPULATION THAT'S GOING TO

3414
02:10:11,630 --> 02:10:13,131
USE THE MODEL THAT'S BEING

3415
02:10:13,131 --> 02:10:16,668
DEVELOPED, I THINK THIS IS WHERE

3416
02:10:16,668 --> 02:10:18,403
ALGORITHMIC BIAS CAUSES THE

3417
02:10:18,403 --> 02:10:20,172
BIGGEST PROBLEMS, WHEN YOU TRAIN

3418
02:10:20,172 --> 02:10:21,807
AN ALGORITHM ON ONE POPULATION

3419
02:10:21,807 --> 02:10:24,309
AND THEN YOU APPLY IT TO ANOTHER

3420
02:10:24,309 --> 02:10:26,211
POPULATION THAT WAS NOT INCLUDED

3421
02:10:26,211 --> 02:10:28,714
IN YOUR DATASET, THERE'S A HIGH

3422
02:10:28,714 --> 02:10:31,516
LIKELIHOOD THERE'S GOING TO BE

3423
02:10:31,516 --> 02:10:32,451
PROBLEMS THERE.

3424
02:10:32,451 --> 02:10:34,619
THIS IS A TANGENT BUT I'M GLAD I

3425
02:10:34,619 --> 02:10:36,922
INCLUDED THIS BECAUSE THIS CAME

3426
02:10:36,922 --> 02:10:38,023
UP EARLIER.

3427
02:10:38,023 --> 02:10:39,791
I THINK THERE'S A REAL TENSION,

3428
02:10:39,791 --> 02:10:43,929
YOU KNOW, ON THE ONE HAND WE I

3429
02:10:43,929 --> 02:10:46,131
THINK AS KIND OF CERTAINLY AS AN

3430
02:10:46,131 --> 02:10:51,503
IRB PERSON, AS ETHICS PEOPLE,

3431
02:10:51,503 --> 02:10:53,605
YOU KNOW, GENERALLY PROMOTE

3432
02:10:53,605 --> 02:10:55,440
CONSENT AS KIND OF BEING, YOU

3433
02:10:55,440 --> 02:10:57,743
KNOW, THE KEY FEATURE TO HOW TO

3434
02:10:57,743 --> 02:11:01,113
DO THIS WORK MOST ETHICALLY.

3435
02:11:01,113 --> 02:11:04,983
YOU KNOW, CONSENT, I THINK THE

3436
02:11:04,983 --> 02:11:06,384
CAM EXAMPLE GIVEN EARLIER SHOULD

3437
02:11:06,384 --> 02:11:07,819
PEOPLE CONSENT TO EXACTLY WHICH

3438
02:11:07,819 --> 02:11:08,787
MODELS THEIR DATA ARE GOING TO

3439
02:11:08,787 --> 02:11:12,190
BE USED TO TRAIN ON, AND WAS

3440
02:11:12,190 --> 02:11:14,159
ALSO SAID EARLIER, AND YOU NEED

3441
02:11:14,159 --> 02:11:16,795
ENORMOUS AMOUNTS OF DATA TO

3442
02:11:16,795 --> 02:11:20,832
REALLY TRAIN THESE ALGORITHMS

3443
02:11:20,832 --> 02:11:22,267
WHILE AGAIN ENORMOUS AMOUNTS OF

3444
02:11:22,267 --> 02:11:24,569
DIVERSE DATA TO BE ADEQUATELY

3445
02:11:24,569 --> 02:11:25,437
REPRESENTED AND NOT BIASED BUT

3446
02:11:25,437 --> 02:11:27,606
AS SOON AS YOU IMPLEMENT

3447
02:11:27,606 --> 02:11:31,877
REQUIREMENTS FOR CONSENT THERE'S

3448
02:11:31,877 --> 02:11:33,245
A LIKELIHOOD THAT YOUR DATABASE

3449
02:11:33,245 --> 02:11:35,413
IS POTENTIALLY GOING TO BE LESS

3450
02:11:35,413 --> 02:11:36,615
REPRESENTATIVE, AND THAT YOU'RE

3451
02:11:36,615 --> 02:11:39,117
ACTUALLY GOING TO END UP WITH

3452
02:11:39,117 --> 02:11:41,620
MORE BIAS DATA, MORE BIASED

3453
02:11:41,620 --> 02:11:42,621
ALGORITHMS, MORE DISCRIMINATION,

3454
02:11:42,621 --> 02:11:45,357
SO THERE'S A REAL TENSION THERE

3455
02:11:45,357 --> 02:11:48,994
AROUND HOY DO YOU DEVELOP

3456
02:11:48,994 --> 02:11:49,861
THOUGHTFUL UNBIASED ALGORITHMS

3457
02:11:49,861 --> 02:11:53,799
THAT ARE TRAINED ON EVERYONE'S

3458
02:11:53,799 --> 02:11:56,301
DATA, WHEN, YOU KNOW, IT'S A

3459
02:11:56,301 --> 02:11:57,936
CHALLENGE TO GET EVERYONE TO BE

3460
02:11:57,936 --> 02:11:59,771
WILLING TO AGREE TO PARTICIPATE

3461
02:11:59,771 --> 02:12:03,108
IN RESEARCH GIVEN, YOU KNOW, THE

3462
02:12:03,108 --> 02:12:05,510
HISTORY THAT WE HAVE.

3463
02:12:05,510 --> 02:12:07,279
AND HAPPY TO TALK MORE ABOUT

3464
02:12:07,279 --> 02:12:08,146
THAT INTERESTING ISSUE.

3465
02:12:08,146 --> 02:12:09,347
ONE OF THE QUESTIONS HAVE THE

3466
02:12:09,347 --> 02:12:11,550
TOOLS THAT ARE BEING USED FOR

3467
02:12:11,550 --> 02:12:13,051
DATA COLLECTION BEEN ASSESSED

3468
02:12:13,051 --> 02:12:15,053
FOR PREEXISTING BIAS SO IF

3469
02:12:15,053 --> 02:12:16,254
YOU'RE USING FACIAL RECOGNITION

3470
02:12:16,254 --> 02:12:19,324
TECHNOLOGY WHICH WAS MENTIONED

3471
02:12:19,324 --> 02:12:28,567
EARLIER, ARE YOU  PULSE OXIMETR,

3472
02:12:28,567 --> 02:12:30,202
CHANCES ARE THAT DATA IS GOING

3473
02:12:30,202 --> 02:12:31,603
TO BE BIASED AND CAUSE PROBLEMS

3474
02:12:31,603 --> 02:12:33,939
IN YOUR ALGORITHM.

3475
02:12:33,939 --> 02:12:37,442
FINALLY HAVE YOU BUILT IN DATA

3476
02:12:37,442 --> 02:12:39,411
BIAS MONITORING, TESTING,

3477
02:12:39,411 --> 02:12:44,216
MITIGATION STRATEGIES THROUGHOUT

3478
02:12:44,216 --> 02:12:46,318
YOUR AI/ML DEVELOPMENT LIFE

3479
02:12:46,318 --> 02:12:46,618
CYCLE.

3480
02:12:46,618 --> 02:12:48,153
I'LL SPEND THE REST OF THE TIME,

3481
02:12:48,153 --> 02:12:49,688
WATCHING THE CLOCK HERE, TRYING

3482
02:12:49,688 --> 02:12:51,857
TO MOVE QUICKLY, RETURN OF

3483
02:12:51,857 --> 02:12:52,557
INDIVIDUAL RESEARCH RESULTS

3484
02:12:52,557 --> 02:12:54,292
WHICH IS AN INTERESTING TOPIC.

3485
02:12:54,292 --> 02:12:56,795
THIS IS AGAIN A PLACE WHERE I

3486
02:12:56,795 --> 02:12:59,664
HAD THE PLEASURE OF WORKING WITH

3487
02:12:59,664 --> 02:13:07,405
DR. SHEN, DR. BAKER, DR. GRADY

3488
02:13:07,405 --> 02:13:10,709
AND OTHERS HERE, TALKING ABOUT

3489
02:13:10,709 --> 02:13:14,646
RETURNING RESULTS, IN PSYCHIATRY

3490
02:13:14,646 --> 02:13:16,815
FROM WHAT DR. BAKER PRESENTED

3491
02:13:16,815 --> 02:13:20,118
THERE'S A DESIRE TO RETURN

3492
02:13:20,118 --> 02:13:21,519
RESULTS, YOU CAN ESSENTIALLY

3493
02:13:21,519 --> 02:13:24,256
PREDICT POTENTIALLY WHEN

3494
02:13:24,256 --> 02:13:26,658
SOMEONE'S ABOUT TO HAVE A MANIC

3495
02:13:26,658 --> 02:13:28,760
OR DEPRESSIVE EPISODE SO

3496
02:13:28,760 --> 02:13:31,496
WOULDN'T IT BE NICE TO USE THAT

3497
02:13:31,496 --> 02:13:34,432
DATA TO INTERVENE SOONER,

3498
02:13:34,432 --> 02:13:36,735
ESSENTIALLY THE GOAL HERE IN

3499
02:13:36,735 --> 02:13:38,603
DOING THIS WORK.

3500
02:13:38,603 --> 02:13:40,338
BUT THE PROBLEM WITH RETURNING

3501
02:13:40,338 --> 02:13:42,874
INDIVIDUAL RESEARCH RESULTS IT'S

3502
02:13:42,874 --> 02:13:44,075
REALLY COMPLICATED WITH THIS

3503
02:13:44,075 --> 02:13:45,043
WORK.

3504
02:13:45,043 --> 02:13:47,012
I THINK THE TOPIC OF INDIVIDUAL

3505
02:13:47,012 --> 02:13:49,080
RESEARCH RESULTS RETURN IS NOT

3506
02:13:49,080 --> 02:13:49,414
UNIQUE.

3507
02:13:49,414 --> 02:13:51,616
I THINK THIS IS NOT A

3508
02:13:51,616 --> 02:13:53,852
CATEGORICALLY UNIQUE PROBLEM BUT

3509
02:13:53,852 --> 02:13:55,220
POSES CHALLENGES DUE TO THE

3510
02:13:55,220 --> 02:13:57,088
AMOUNT AND TYPE OF DATA.

3511
02:13:57,088 --> 02:14:00,692
SO DR. BAKER HIGHLIGHTEDDED THIS

3512
02:14:00,692 --> 02:14:02,227
BUT YOU CAN SEE IN THIS BOX ON

3513
02:14:02,227 --> 02:14:03,962
THE SCREEN THERE'S A TON OF

3514
02:14:03,962 --> 02:14:06,598
TYPES OF DATA THAT IS COLLECTED

3515
02:14:06,598 --> 02:14:08,600
AND SO IT RAISES QUESTIONS

3516
02:14:08,600 --> 02:14:09,434
AROUND, YOU KNOW, HOW MUCH

3517
02:14:09,434 --> 02:14:13,271
YOU'RE GOING TO RETURN AND WHAT

3518
02:14:13,271 --> 02:14:14,239
YOU'RE GOING TO RETURN, HOW

3519
02:14:14,239 --> 02:14:18,643
YOU'RE GOING TO RETURN IT. 

3520
02:14:18,643 --> 02:14:21,046
EXISTING ETHICAL FRAMEWORKS WERE

3521
02:14:21,046 --> 02:14:21,579
INSUFFICIENT TO GUIDE

3522
02:14:21,579 --> 02:14:22,881
RESEARCHERS ON THIS.

3523
02:14:22,881 --> 02:14:25,083
JUST TO THINK ABOUT AT A HIGH

3524
02:14:25,083 --> 02:14:28,353
LEVEL I THINK THE ISSUES

3525
02:14:28,353 --> 02:14:31,289
ESSENTIALLY ARE RELATED TO

3526
02:14:31,289 --> 02:14:33,825
UNCERTAINTY ABOUT ANALYTICAL

3527
02:14:33,825 --> 02:14:35,994
VALIDITY, CLINICAL VALIDITY,

3528
02:14:35,994 --> 02:14:37,429
ACTIONABILITY, AND PERSONAL

3529
02:14:37,429 --> 02:14:38,797
UTILITY, AGAIN INCLUDING

3530
02:14:38,797 --> 02:14:42,801
POTENTIAL FOR BIAS.

3531
02:14:42,801 --> 02:14:44,002
QUICKLY ANALYTIC VALIDITY

3532
02:14:44,002 --> 02:14:45,203
MEANING IS THE ALGORITHM

3533
02:14:45,203 --> 02:14:46,638
ACTUALLY FUNCTIONING THE WAY

3534
02:14:46,638 --> 02:14:47,706
IT'S SUPPOSED TO?

3535
02:14:47,706 --> 02:14:49,908
IS IT PICKING UP AND COLLECTING

3536
02:14:49,908 --> 02:14:51,543
THE DATA AND PROVIDING THE DATA

3537
02:14:51,543 --> 02:14:53,178
THE WAY IT'S SUPPOSED TO?

3538
02:14:53,178 --> 02:14:55,580
ARE THE NUMBER OF STEPS THAT'S

3539
02:14:55,580 --> 02:14:56,348
REPORTING SOMEONE TOOK ACTUALLY

3540
02:14:56,348 --> 02:14:58,450
ACCURATE TO THE NUMBER OF STEPS

3541
02:14:58,450 --> 02:15:00,552
THEY TOOK?

3542
02:15:00,552 --> 02:15:04,055
AND IN CLINICAL VALIDITY, IS IT

3543
02:15:04,055 --> 02:15:05,924
ACTUALLY INDICATIVE OF A

3544
02:15:05,924 --> 02:15:07,359
CLINICAL SCENARIO, RIGHT?

3545
02:15:07,359 --> 02:15:08,660
SO THERE MAY BE THE ALGORITHM

3546
02:15:08,660 --> 02:15:11,062
MAY PICK UP ON SOMETHING RELATED

3547
02:15:11,062 --> 02:15:12,864
TO HEART RATE VARIABILITY, AND

3548
02:15:12,864 --> 02:15:14,899
COULD THAT BE, YOU KNOW, SOMEONE

3549
02:15:14,899 --> 02:15:17,068
EXERCISING OR COULD THAT BE

3550
02:15:17,068 --> 02:15:18,937
SOMEONE HAVING A CARDIAC EVENT

3551
02:15:18,937 --> 02:15:23,208
OR SOMEONE HAVING SEX, RIGHT?

3552
02:15:23,208 --> 02:15:26,478
WHAT THE VALIDITY OF THE KINDS

3553
02:15:26,478 --> 02:15:28,213
OF CLINICAL SCENARIO IS HAS TO

3554
02:15:28,213 --> 02:15:31,816
BE VALID IN ORDER TO RETURN THE

3555
02:15:31,816 --> 02:15:32,283
RESULTS.

3556
02:15:32,283 --> 02:15:32,951
ACTIONABILITY AND PERSONAL

3557
02:15:32,951 --> 02:15:33,685
UTILITY BASICALLY IS THERE

3558
02:15:33,685 --> 02:15:35,887
SOMETHING YOU CAN DO ABOUT THE

3559
02:15:35,887 --> 02:15:38,323
RESULTS AND/OR WOULD SOMEONE

3560
02:15:38,323 --> 02:15:39,524
FIND THEM PERSONALLY USEFUL?

3561
02:15:39,524 --> 02:15:41,493
ONE OF THE ISSUES IS THERE'S SO

3562
02:15:41,493 --> 02:15:44,763
MANY STREAMS OF DATA, YOU CAN

3563
02:15:44,763 --> 02:15:48,800
RETURN IT AT MANY TIME POINTS,

3564
02:15:48,800 --> 02:15:49,434
BASICALLY CONSTANTLY RETURNING

3565
02:15:49,434 --> 02:15:55,173
DATA, SOME DATA IS GOING TO

3566
02:15:55,173 --> 02:15:59,544
REVEAL POTENTIALLY SOCIALLY

3567
02:15:59,544 --> 02:16:00,078
STIGMATIZED, POLICEICALLY

3568
02:16:00,078 --> 02:16:00,745
SENSITIVE DATA, SOMEONE MIGHT

3569
02:16:00,745 --> 02:16:07,552
HAVE BEEN AT THE SCENE OF A

3570
02:16:07,552 --> 02:16:08,653
CRIME, ESSENTIALLY.

3571
02:16:08,653 --> 02:16:11,156
WE DON'T KNOW HOW RETURNING

3572
02:16:11,156 --> 02:16:13,658
DIGITAL PHENOTYPING RESULTS WILL

3573
02:16:13,658 --> 02:16:14,392
IMPACT BEHAVIOR.

3574
02:16:14,392 --> 02:16:15,860
YOU KNOW, IN THINKING ABOUT THIS

3575
02:16:15,860 --> 02:16:20,131
WORK WE CAME UP WITH EXAMPLES

3576
02:16:20,131 --> 02:16:21,666
WHERE WE THOUGHT THAT RETURNING

3577
02:16:21,666 --> 02:16:23,735
RESULTS WOULD IMPACT PEOPLE IN

3578
02:16:23,735 --> 02:16:25,603
ONE WAY, IT ACTUALLY ENDED UP

3579
02:16:25,603 --> 02:16:28,773
BEING THE OPPOSITE.

3580
02:16:28,773 --> 02:16:31,376
QUICK EXAMPLE, A SIMPLISTIC

3581
02:16:31,376 --> 02:16:35,013
EXAMPLE BUT RETURNING DATA FROM

3582
02:16:35,013 --> 02:16:37,849
A SLEEP WEARABLE TO PEOPLE IN

3583
02:16:37,849 --> 02:16:39,384
MEDICAL TRAINING, AND YOU WOULD

3584
02:16:39,384 --> 02:16:41,352
THINK WHEN YOU PROVIDED THEM

3585
02:16:41,352 --> 02:16:42,987
WITH DATA THAT SHOWED THEY WERE

3586
02:16:42,987 --> 02:16:47,125
SLEEPING LESS THAN THEIR PEERS,

3587
02:16:47,125 --> 02:16:50,195
I'M SORRY, SLEEPING, YEAH, LESS

3588
02:16:50,195 --> 02:16:51,396
THAN THEIR PEERS, THAT THEY

3589
02:16:51,396 --> 02:16:53,064
MIGHT, YOU KNOW, WANT TO CATCH

3590
02:16:53,064 --> 02:16:55,133
UP ON SLEEP ESSENTIALLY, RIGHT?

3591
02:16:55,133 --> 02:16:57,669
BUT ACTUALLY IT OFTEN HAS

3592
02:16:57,669 --> 02:16:59,404
OPPOSITE EFFECT, AND IT'S NOT

3593
02:16:59,404 --> 02:17:01,039
SLEEPING AND WORKING HARD IS

3594
02:17:01,039 --> 02:17:03,141
OVERLY PRIZED IN THAT

3595
02:17:03,141 --> 02:17:04,109
POPULATION.

3596
02:17:04,109 --> 02:17:05,844
AND SO IT ACTUALLY FOR SOME

3597
02:17:05,844 --> 02:17:07,946
PEOPLE MADE THEM SLEEP EVEN

3598
02:17:07,946 --> 02:17:08,379
LESS.

3599
02:17:08,379 --> 02:17:12,217
IT CAN BE VERY HARD TO KNOW

3600
02:17:12,217 --> 02:17:12,984
EXACTLY WHAT THE CONSEQUENCES

3601
02:17:12,984 --> 02:17:14,185
WILL BE.

3602
02:17:14,185 --> 02:17:15,920
AND THEN FINALLY THERE'S A LOT

3603
02:17:15,920 --> 02:17:17,689
OF THIRD PARTY DATA THAT'S

3604
02:17:17,689 --> 02:17:19,657
COLLECTED IN THIS DATA.

3605
02:17:19,657 --> 02:17:22,193
SO THERE'S POTENTIAL RISKS TO

3606
02:17:22,193 --> 02:17:23,194
THIRD PARTIES.

3607
02:17:23,194 --> 02:17:25,463
THESE ARE THE QUESTIONS THAT WE

3608
02:17:25,463 --> 02:17:25,697
POSED.

3609
02:17:25,697 --> 02:17:27,665
I'LL MOVE THROUGH THIS QUICKLY

3610
02:17:27,665 --> 02:17:28,533
FOR TIME.

3611
02:17:28,533 --> 02:17:30,502
BASICALLY ASKING IF PEOPLE HAVE

3612
02:17:30,502 --> 02:17:33,004
THOUGHT ABOUT THESE ISSUES IN

3613
02:17:33,004 --> 02:17:33,905
ADVANCE OF STARTING THEIR WORK.

3614
02:17:33,905 --> 02:17:35,640
AT THE END OF THE DAY WE DID

3615
02:17:35,640 --> 02:17:38,710
COME UP WITH A FRAMEWORK TO TRY

3616
02:17:38,710 --> 02:17:39,577
TO GUIDE PEOPLE.

3617
02:17:39,577 --> 02:17:41,212
I THINK IT RAISES MORE QUESTIONS

3618
02:17:41,212 --> 02:17:43,748
THAN ANSWERS BUT WE TRIED TO

3619
02:17:43,748 --> 02:17:46,384
BALANCE RISKS AND BENEFITS,

3620
02:17:46,384 --> 02:17:47,585
RESPECTING PERSONS BOTH RESEARCH

3621
02:17:47,585 --> 02:17:48,686
PARTICIPANTS AND THIRD PARTIES,

3622
02:17:48,686 --> 02:17:49,888
TRIED TO PROMOTE JUSTICE IN

3623
02:17:49,888 --> 02:17:50,822
RETURNING DATA.

3624
02:17:50,822 --> 02:17:54,025
WE DO THINK KIND OF AS A DEFAULT

3625
02:17:54,025 --> 02:17:55,460
PRESUMPTION THAT WE SHOULD BE

3626
02:17:55,460 --> 02:17:57,629
TRYING TO RETURN SOME OF THIS

3627
02:17:57,629 --> 02:17:59,497
DATA TO PROMOTE JUSTICE.

3628
02:17:59,497 --> 02:18:01,266
AND SO WE ENDED UP SAYING

3629
02:18:01,266 --> 02:18:03,001
ESSENTIALLY THAT IF SOMETHING IS

3630
02:18:03,001 --> 02:18:04,669
HIGH BENEFIT AND LOW RISK THAT

3631
02:18:04,669 --> 02:18:07,572
WE DO THINK YOU SHOULD BE

3632
02:18:07,572 --> 02:18:09,807
RETURNING RESULTS, IF IT IS LOW

3633
02:18:09,807 --> 02:18:11,676
RISK AND HIGH BENEFIT -- I'M

3634
02:18:11,676 --> 02:18:13,311
SORRY, LOW BENEFIT AND HIGH RISK

3635
02:18:13,311 --> 02:18:14,612
YOU SHOULD START WITH

3636
02:18:14,612 --> 02:18:15,947
PRESUMPTION OF NOT RETURNING

3637
02:18:15,947 --> 02:18:16,181
RESULTS.

3638
02:18:16,181 --> 02:18:18,016
AND EVERYTHING ELSE IS SOMEWHERE

3639
02:18:18,016 --> 02:18:19,884
IN THE MIDDLE.

3640
02:18:19,884 --> 02:18:20,752
SO NOTHING GROUNDBREAKING THERE

3641
02:18:20,752 --> 02:18:22,187
BUT TRYING TO JUST GIVE PEOPLE A

3642
02:18:22,187 --> 02:18:23,821
WAY TO THINK ABOUT THIS AND

3643
02:18:23,821 --> 02:18:25,089
REALLY ASKING PEOPLE TO CONSIDER

3644
02:18:25,089 --> 02:18:26,558
IF YOU'RE GOING TO DO THIS WHO

3645
02:18:26,558 --> 02:18:28,626
IS GOING TO RECEIVE THE RESULTS

3646
02:18:28,626 --> 02:18:36,834
AND WHO IS GOING TO PROVIDE

3647
02:18:36,834 --> 02:18:38,069
RESULTS, WHERE AND HOW

3648
02:18:38,069 --> 02:18:39,704
FREQUENTLY AND IN WHAT FORMAT

3649
02:18:39,704 --> 02:18:41,573
WILL IT BE RETURNED.

3650
02:18:41,573 --> 02:18:45,410
DUTY TO WARN OR REPORT, THIS IS

3651
02:18:45,410 --> 02:18:46,411
AT THE INTERSECTION, LIMITATIONS

3652
02:18:46,411 --> 02:18:48,613
OF OUR DUTIES OF PRIVACY AND

3653
02:18:48,613 --> 02:18:48,980
CONFIDENTIALITY.

3654
02:18:48,980 --> 02:18:52,650
AND THE RETURN OF INDIVIDUAL

3655
02:18:52,650 --> 02:18:54,152
RESEARCH RESULTS.

3656
02:18:54,152 --> 02:18:57,889
BECAUSE IF YOU'RE GOING TO USE

3657
02:18:57,889 --> 02:18:59,424
THESE RESULTS TO TRIGGER A DUTY

3658
02:18:59,424 --> 02:19:01,292
TO WARN OR DUTY TO REPORT YOU

3659
02:19:01,292 --> 02:19:03,795
HAVE TO THINK THE RESULTS ARE

3660
02:19:03,795 --> 02:19:04,596
ANALYTICALLY AND CLINICALLY

3661
02:19:04,596 --> 02:19:06,965
VALID, ABLE TO BE RETURNED AS A

3662
02:19:06,965 --> 02:19:08,166
RESEARCH RESULT, AND SO

3663
02:19:08,166 --> 02:19:10,034
ENCOURAGING PEOPLE TO THINK

3664
02:19:10,034 --> 02:19:11,669
ABOUT THIS IN ADVANCE, WHEN ARE

3665
02:19:11,669 --> 02:19:14,639
YOU GOING TO REACH THE POINT

3666
02:19:14,639 --> 02:19:20,678
WHERE YOU FEEL LIKE YOUR METHODS

3667
02:19:20,678 --> 02:19:22,981
ARE VALIDATED ENOUGH YOU HAVE TO

3668
02:19:22,981 --> 02:19:24,749
RETURN TO WARN OR REPORT.

3669
02:19:24,749 --> 02:19:26,818
ALSO WHAT ARE YOU GOING TO DO

3670
02:19:26,818 --> 02:19:29,554
WHEN YOU UNCOVER THINGS LIKE

3671
02:19:29,554 --> 02:19:31,322
CHILD PORNOGRAPHY OR OTHER

3672
02:19:31,322 --> 02:19:32,624
THINGS WHEN YOU'RE COLLECTING

3673
02:19:32,624 --> 02:19:34,492
THIS DATA WHICH IS INEVITABLE IF

3674
02:19:34,492 --> 02:19:35,793
YOU COLLECT ENOUGH DATA.

3675
02:19:35,793 --> 02:19:38,429
SO JUST A FEW TAKEHOME POINTS, I

3676
02:19:38,429 --> 02:19:40,498
THINK OUR ESTABLISHED GUIDELINES

3677
02:19:40,498 --> 02:19:42,700
AND REGULATIONS REALLY ARE

3678
02:19:42,700 --> 02:19:44,335
INSUFFICIENT TO GUIDE

3679
02:19:44,335 --> 02:19:45,903
RESEARCHERS ON ETHICAL

3680
02:19:45,903 --> 02:19:47,739
IMPLICATIONS OF PHENOTYPING AND

3681
02:19:47,739 --> 02:19:48,106
RESEARCH.

3682
02:19:48,106 --> 02:19:51,743
WE THINK AGAIN THE KEY TAKEAWAY,

3683
02:19:51,743 --> 02:19:53,011
YOU HAVE TO THINK ABOUT THESE

3684
02:19:53,011 --> 02:19:54,545
THINGS IN ADVANCE.

3685
02:19:54,545 --> 02:19:57,582
WE'RE ENCOURAGING RESEARCHERS TO

3686
02:19:57,582 --> 02:19:58,783
MAKE ETHICAL CHOICES EXPLICIT AS

3687
02:19:58,783 --> 02:20:01,753
PART OF DESIGN AND COLLABORATING

3688
02:20:01,753 --> 02:20:02,720
WITH THEIR IRBs, RELEVANT

3689
02:20:02,720 --> 02:20:03,521
STAKEHOLDERS, ET CETERA.

3690
02:20:03,521 --> 02:20:04,956
AND REALLY AT THE END OF THE DAY

3691
02:20:04,956 --> 02:20:07,258
THERE ARE A RANGE OF REASONABLE

3692
02:20:07,258 --> 02:20:08,459
CHOICES THAT RESEARCHERS CAN

3693
02:20:08,459 --> 02:20:09,327
MAKE RIGHT NOW.

3694
02:20:09,327 --> 02:20:11,829
FOR EXAMPLE ON RETURNING

3695
02:20:11,829 --> 02:20:12,897
INDIVIDUAL RESEARCH RESULTS, YOU

3696
02:20:12,897 --> 02:20:14,599
CAN MAKE CHOICES TO DO IT OR NOT

3697
02:20:14,599 --> 02:20:16,668
DO IT AND THAT COULD BE VERY

3698
02:20:16,668 --> 02:20:18,970
REASONABLE BUT I THINK PEOPLE

3699
02:20:18,970 --> 02:20:22,807
SHOULD BE REQUIRED TO THINK THIS

3700
02:20:22,807 --> 02:20:24,342
OUT AND GIVE RATIONALE BEHIND

3701
02:20:24,342 --> 02:20:27,312
THEIR DECISIONS TO THEIR IRBs

3702
02:20:27,312 --> 02:20:29,714
AND OTHER OVERSIGHT BODIES.

3703
02:20:29,714 --> 02:20:31,683
AGAIN, PROBABLY RAISING MORE

3704
02:20:31,683 --> 02:20:33,451
QUESTIONS THAN ANSWERS BUT

3705
02:20:33,451 --> 02:20:34,619
HOPEFULLY A WHIRLWIND TOUR

3706
02:20:34,619 --> 02:20:36,487
THROUGH THE ETHICS OF SOME OF

3707
02:20:36,487 --> 02:20:45,563
THE WORK WE DO AND AGAIN THANKS

3708
02:20:45,563 --> 02:20:46,564
FOR EVERYONE'S ATTENTION.

3709
02:20:46,564 --> 02:20:46,931
>> THANK YOU.

3710
02:20:46,931 --> 02:20:52,437
I WANT TO OPEN THE DISCUSSION

3711
02:20:52,437 --> 02:20:55,506
FOR EVERYBODY ONLINE TO FOCUS ON

3712
02:20:55,506 --> 02:20:57,275
TWO POINTS.

3713
02:20:57,275 --> 02:21:00,244
ONE IS WHAT'S UNIQUE HERE IN

3714
02:21:00,244 --> 02:21:02,180
TERMS OF NEUROSCIENCE OR BRAIN

3715
02:21:02,180 --> 02:21:04,382
RESEARCH AND SOME OF THE ISSUES

3716
02:21:04,382 --> 02:21:06,017
THAT HAVE BEEN RAISED, AND I

3717
02:21:06,017 --> 02:21:09,120
THINK THERE'S SO MANY

3718
02:21:09,120 --> 02:21:10,421
INTERESTING ISSUES.

3719
02:21:10,421 --> 02:21:12,123
YOU'VE RAISED ALL -- ALL OF THE

3720
02:21:12,123 --> 02:21:14,025
SPEAKERS HAVE RAISED A NUMBER OF

3721
02:21:14,025 --> 02:21:15,560
THEM, REALLY GREAT, BUT FOCUSING

3722
02:21:15,560 --> 02:21:17,862
IN WHAT CAN WE THINK ABOUT IN

3723
02:21:17,862 --> 02:21:20,064
TERMS OF BRAIN RESEARCH AND

3724
02:21:20,064 --> 02:21:22,033
ARTIFICIAL INTELLIGENCE AND THAT

3725
02:21:22,033 --> 02:21:22,467
INTERSECTION?

3726
02:21:22,467 --> 02:21:24,769
AND I WANT TO START WITH THE

3727
02:21:24,769 --> 02:21:28,139
FIRST QUESTION I THINK FOR DR.

3728
02:21:28,139 --> 02:21:28,406
SILVERMAN.

3729
02:21:28,406 --> 02:21:30,675
YOU MENTIONED THAT, YOU KNOW, WE

3730
02:21:30,675 --> 02:21:33,177
SHOULD KNOW A LITTLE BIT MORE

3731
02:21:33,177 --> 02:21:34,078
ABOUT WHAT REASONABLE PEOPLE

3732
02:21:34,078 --> 02:21:35,380
WANT TO KNOW AND WHAT THEY ARE

3733
02:21:35,380 --> 02:21:37,582
WORRIED ABOUT IN THE PROCESS OF

3734
02:21:37,582 --> 02:21:38,683
CONSENT.

3735
02:21:38,683 --> 02:21:40,218
AND THAT WE SHOULD ALSO BE

3736
02:21:40,218 --> 02:21:41,519
EXPLAINING TO THEM WHAT'S GOING

3737
02:21:41,519 --> 02:21:43,154
TO HAPPEN TO THEIR DATA, ET

3738
02:21:43,154 --> 02:21:50,161
CETERA, TALKED ABOUT IT AS

3739
02:21:50,161 --> 02:21:50,728
COMPLEX CONTINUOUS SOMETIMES

3740
02:21:50,728 --> 02:21:52,263
PASSIVE A.I. MACHINE LEARNING

3741
02:21:52,263 --> 02:21:53,231
SYSTEMS, ET CETERA.

3742
02:21:53,231 --> 02:21:55,733
I GUESS MY QUESTION IS, IS IT

3743
02:21:55,733 --> 02:21:58,736
REALISTIC TO EXPECT PEOPLE TO

3744
02:21:58,736 --> 02:22:01,339
UNDERSTAND HOW THE A.I. AND

3745
02:22:01,339 --> 02:22:03,941
MACHINE LEARNING SYSTEMS THAT

3746
02:22:03,941 --> 02:22:05,676
THEIR DATA ARE EITHER BEING

3747
02:22:05,676 --> 02:22:10,181
INTERPRETED WITH OR FED INTO OR

3748
02:22:10,181 --> 02:22:12,717
BOTH, WHAT THAT MEANS TO THEM I

3749
02:22:12,717 --> 02:22:14,552
GUESS, AND IS THAT SOMETHING

3750
02:22:14,552 --> 02:22:19,257
THAT WE CAN OR SHOULD BE TRYING

3751
02:22:19,257 --> 02:22:26,063
TO ADDRESS THROUGH CONSENT

3752
02:22:26,063 --> 02:22:26,330
PROCESSES?

3753
02:22:26,330 --> 02:22:27,298
>> GREAT QUESTION.

3754
02:22:27,298 --> 02:22:30,435
I'M SURE EVERYONE HAS THOUGHTS

3755
02:22:30,435 --> 02:22:32,737
SO I'LL SUMMARIZE MY THOUGHTS.

3756
02:22:32,737 --> 02:22:35,039
I THINK -- SO I COME FROM AS YOU

3757
02:22:35,039 --> 02:22:37,942
ALL KNOW THE WORLD OF RESEARCH

3758
02:22:37,942 --> 02:22:40,945
ETHICS, AND SO WHEN I MADE MY

3759
02:22:40,945 --> 02:22:43,347
SLIDES AND KIND OF IN PARTICULAR

3760
02:22:43,347 --> 02:22:44,348
ORDER, REASONABLE PERSON

3761
02:22:44,348 --> 02:22:46,517
STANDARD, YOU KNOW, THINKING

3762
02:22:46,517 --> 02:22:48,619
ABOUT QUOTE/UNQUOTE ELEMENTS OF

3763
02:22:48,619 --> 02:22:49,687
CONSENT IT'S FOR RESEARCH

3764
02:22:49,687 --> 02:22:50,555
PARTICIPATION THAT I'M THINKING

3765
02:22:50,555 --> 02:22:52,657
ABOUT BUT THE INTERESTING THING,

3766
02:22:52,657 --> 02:22:55,927
JUST TO TAKE A STEP BACK AND

3767
02:22:55,927 --> 02:22:58,429
SAY, YOU KNOW, ESSENTIALLY I

3768
02:22:58,429 --> 02:23:00,198
THINK THE QUESTION YOU'RE ASKING

3769
02:23:00,198 --> 02:23:00,998
CAN OR SHOULD PEOPLE CONSENT TO

3770
02:23:00,998 --> 02:23:04,035
THE USE OF A.I. AND CAN THEY

3771
02:23:04,035 --> 02:23:06,003
REALLY UNDERSTAND IT, AND, YOU

3772
02:23:06,003 --> 02:23:10,141
KNOW, I WOULD ARGUE THAT

3773
02:23:10,141 --> 02:23:12,910
CLINICALLY THEY PROBABLY CAN'T,

3774
02:23:12,910 --> 02:23:14,212
AND PROBABLY DON'T WANT TO.

3775
02:23:14,212 --> 02:23:16,514
AND JUST TO GIVE A QUICK

3776
02:23:16,514 --> 02:23:20,551
EXAMPLE, LIKE IF YOU GO TO THE

3777
02:23:20,551 --> 02:23:22,687
ICU OF ANY OF OUR MAJOR ACADEMIC

3778
02:23:22,687 --> 02:23:23,788
MEDICAL CENTERS, AND YOU HAVE

3779
02:23:23,788 --> 02:23:26,524
SOMEONE WHO IS CONNECTED TO A

3780
02:23:26,524 --> 02:23:29,894
VENTILATOR, THOSE VENTILATORS

3781
02:23:29,894 --> 02:23:30,895
ARE MAKING DECISIONS SOMEWHAT ON

3782
02:23:30,895 --> 02:23:32,230
THEIR OWN AT THIS POINT.

3783
02:23:32,230 --> 02:23:34,532
WE'RE NOT GOING AND CHANGING

3784
02:23:34,532 --> 02:23:37,034
EVERY TINY SETTING THE WAY THAT,

3785
02:23:37,034 --> 02:23:39,904
YOU KNOW, MAYBE WE DID BEFORE I

3786
02:23:39,904 --> 02:23:42,073
WAS IN MEDICAL TRAINING, RIGHT?

3787
02:23:42,073 --> 02:23:43,608
AND NOBODY'S TELLING YOU THAT.

3788
02:23:43,608 --> 02:23:46,244
THEY JUST WANT TO MAKE SURE, YOU

3789
02:23:46,244 --> 02:23:49,847
KNOW, YOU WANT TO MAKE SURE THE

3790
02:23:49,847 --> 02:23:50,748
VENTILATORS WORK, RIGHT?

3791
02:23:50,748 --> 02:23:53,784
WE USE A.I. AS PART OF OUR DAILY

3792
02:23:53,784 --> 02:23:53,985
WORK.

3793
02:23:53,985 --> 02:23:59,056
ANYTIME YOU PULL UP GOOGLE MAPS,

3794
02:23:59,056 --> 02:24:00,391
APPLE MAPS, CHECKING THE

3795
02:24:00,391 --> 02:24:01,926
TRAFFIC, TO GET THERE AS QUICKLY

3796
02:24:01,926 --> 02:24:03,794
AS YOU CAN, THEY ARE USING A.I.

3797
02:24:03,794 --> 02:24:05,630
ALGORITHMS TO DO THAT, RIGHT IN

3798
02:24:05,630 --> 02:24:07,398
THE DO YOU CARE HOW THAT'S

3799
02:24:07,398 --> 02:24:07,632
WORKING?

3800
02:24:07,632 --> 02:24:10,434
OR THAT IT MIGHT BE WRONG OR

3801
02:24:10,434 --> 02:24:14,372
BIASED? 

3802
02:24:14,372 --> 02:24:15,907
OR DO YOU JUST WANT TO KNOW THE

3803
02:24:15,907 --> 02:24:17,341
QUICKEST WAY TO YOUR DESTINATION

3804
02:24:17,341 --> 02:24:18,543
IN THE REASONABLE PERSON DOES

3805
02:24:18,543 --> 02:24:20,711
NOT CARE OR WANT TO KNOW THAT

3806
02:24:20,711 --> 02:24:21,345
INFORMATION.

3807
02:24:21,345 --> 02:24:23,047
WHEN WE THINK ABOUT WHAT THIS IS

3808
02:24:23,047 --> 02:24:28,052
EVENTUALLY GOING TO LOOK LIKE IN

3809
02:24:28,052 --> 02:24:29,687
CLINICAL CARE, I THINK WE MAY

3810
02:24:29,687 --> 02:24:30,021
NOT.

3811
02:24:30,021 --> 02:24:31,889
THE PLACE I DRAW THE

3812
02:24:31,889 --> 02:24:32,957
DISTINCTION, WHEN THERE'S A

3813
02:24:32,957 --> 02:24:34,959
PHYSICIAN INVOLVED OR NOT, A

3814
02:24:34,959 --> 02:24:37,695
CLINICIAN INVOLVED OR NOT, SO IF

3815
02:24:37,695 --> 02:24:40,298
YOUR CLINICIAN IS USING THESE

3816
02:24:40,298 --> 02:24:41,399
DIFFERENT TOOLS, USING JUDGMENT,

3817
02:24:41,399 --> 02:24:43,601
MOST PEOPLE WOULD JUST FEEL

3818
02:24:43,601 --> 02:24:45,770
COMFORTABLE CLINICIAN IS USING

3819
02:24:45,770 --> 02:24:46,537
THEIR JUDGMENT.

3820
02:24:46,537 --> 02:24:48,072
IF THERE'S A COMPUTER THAT'S

3821
02:24:48,072 --> 02:24:49,574
MAKING A DECISION WITHOUT MY

3822
02:24:49,574 --> 02:24:50,575
CLINICIAN ABOUT WHAT MEDICATION

3823
02:24:50,575 --> 02:24:53,010
I'M GOING TO RECEIVING ON

3824
02:24:53,010 --> 02:24:53,878
SOMETHING THAT, THAT FEELS

3825
02:24:53,878 --> 02:24:55,613
DIFFERENT AND PEOPLE WANT TO

3826
02:24:55,613 --> 02:24:55,846
KNOW.

3827
02:24:55,846 --> 02:24:57,281
ALL THAT SAID, THEN WHAT DOES

3828
02:24:57,281 --> 02:24:59,383
THAT MEAN FOR OUR CURRENT

3829
02:24:59,383 --> 02:25:00,351
RESEARCH STUDY?

3830
02:25:00,351 --> 02:25:02,086
WE HAVE TO HOLD OUR CURRENT

3831
02:25:02,086 --> 02:25:03,888
RESEARCH STUDIES TO A SLIGHTLY

3832
02:25:03,888 --> 02:25:04,922
HIGHER STANDARD SO I THINK WE

3833
02:25:04,922 --> 02:25:06,724
HAVE TO TRY TO AT LEAST LET

3834
02:25:06,724 --> 02:25:10,394
PEOPLE KNOW THAT A.I. IS BEING

3835
02:25:10,394 --> 02:25:11,696
USED.

3836
02:25:11,696 --> 02:25:12,797
BUT OF THE DEGREE TO HOW MUCH

3837
02:25:12,797 --> 02:25:17,168
THEY HAVE TO UNDERSTAND IS OPEN

3838
02:25:17,168 --> 02:25:17,835
FOR DEBATE.

3839
02:25:17,835 --> 02:25:18,803
>> OTHERS HAVE VIEWS.

3840
02:25:18,803 --> 02:25:23,441
THERE WAS A DISCUSSION IN THE

3841
02:25:23,441 --> 02:25:25,943
CHAT ABOUT EVEN IDENTIFIABILITY

3842
02:25:25,943 --> 02:25:28,446
IS NOT -- YOU CAN'T PROTECT THAT

3843
02:25:28,446 --> 02:25:29,647
COMPLETELY THROUGH A.I., FIRST

3844
02:25:29,647 --> 02:25:33,818
OF ALL, BUT, YOU KNOW, WE'RE

3845
02:25:33,818 --> 02:25:35,786
ALSO IN DEEP PHENOTYPING FOR

3846
02:25:35,786 --> 02:25:36,887
PSYCHIATRIC DISEASES THERE'S

3847
02:25:36,887 --> 02:25:38,656
SENSITIVITY IN SOME OF THOSE

3848
02:25:38,656 --> 02:25:40,925
DATA SO THAT'S A REALLY

3849
02:25:40,925 --> 02:25:44,228
COMPLICATED COMBINATION OF

3850
02:25:44,228 --> 02:25:46,297
FACTORS.

3851
02:25:46,297 --> 02:25:46,464
JIM?

3852
02:25:46,464 --> 02:25:48,165
>> THANK YOU.

3853
02:25:48,165 --> 02:25:50,668
I THINK MY QUESTION PLAYS ON

3854
02:25:50,668 --> 02:25:51,469
CHRISTINE'S QUESTION.

3855
02:25:51,469 --> 02:25:54,305
THAT'S WITH RESPECT TO PATIENT

3856
02:25:54,305 --> 02:25:54,639
PRIVACY.

3857
02:25:54,639 --> 02:25:56,941
WE HEARD EARLIER ABOUT A.I. AND

3858
02:25:56,941 --> 02:26:00,111
SOME OF THE ASPECTS OF FEDERATED

3859
02:26:00,111 --> 02:26:01,512
LEARNING, HOW ALGORITHMS

3860
02:26:01,512 --> 02:26:03,280
DEVELOP, LARGE DATASETS NEEDED.

3861
02:26:03,280 --> 02:26:05,916
IN TERMS OF FEDERATED DATASETS,

3862
02:26:05,916 --> 02:26:09,086
ARE THERE DIFFERENCES IN

3863
02:26:09,086 --> 02:26:15,626
THINKING ABOUT FEDERATED

3864
02:26:15,626 --> 02:26:16,627
DATASETS FOR NEUROPSYCHIATRIC

3865
02:26:16,627 --> 02:26:18,095
PATIENTS, IS THERE A DIFFERENT

3866
02:26:18,095 --> 02:26:21,999
STRUCTURE OF DATASETS TO HELP

3867
02:26:21,999 --> 02:26:27,138
PROTECT PRIVACY AS DATA IS

3868
02:26:27,138 --> 02:26:28,973
UTILIZED IN APPLICATIONS, ARE

3869
02:26:28,973 --> 02:26:31,175
THERE WAYS OF THINKING ABOUT

3870
02:26:31,175 --> 02:26:34,245
SPECIFICITY OF THOSE TYPES OF

3871
02:26:34,245 --> 02:26:34,679
STRUCTURES?

3872
02:26:34,679 --> 02:26:37,181
>> DR. BAKER, IF YOU WANT
TO

3873
02:26:37,181 --> 02:26:40,584
WEIGH IN, I CAN CERTAINLY

3874
02:26:40,584 --> 02:26:41,185
COMMENT BRIEFLY.

3875
02:26:41,185 --> 02:26:42,453
>> IT'S A GREAT QUESTION.

3876
02:26:42,453 --> 02:26:44,922
ONE OF THE THINGS THAT WE TRY TO

3877
02:26:44,922 --> 02:26:46,791
DO IN TERMS OF BRINGING SOME

3878
02:26:46,791 --> 02:26:49,560
APPROACHES INTO MORE OF A

3879
02:26:49,560 --> 02:26:52,496
CLINICAL USE CASE IS THINKING

3880
02:26:52,496 --> 02:26:54,465
ABOUT LIKE THE NORMS, WHAT ARE

3881
02:26:54,465 --> 02:26:56,133
THE NORMS FOR FACIAL EXPRESSION,

3882
02:26:56,133 --> 02:27:00,838
WHAT ARE NORMS FOR SLEEP OR

3883
02:27:00,838 --> 02:27:03,040
LOCATIONS, TO REALLY DERIVE

3884
02:27:03,040 --> 02:27:03,808
POPULATION-LEVEL NORMS YOU NEED

3885
02:27:03,808 --> 02:27:06,210
TO BE ABLE TO HAVE DATASETS THAT

3886
02:27:06,210 --> 02:27:07,845
SPAN MANY, MANY STUDIES, RIGHT?

3887
02:27:07,845 --> 02:27:12,349
WE'VE STRUGGLED WITH HOW DO YOU

3888
02:27:12,349 --> 02:27:13,451
COMBINE FACIAL EXPRESSION DATA

3889
02:27:13,451 --> 02:27:16,387
ACROSS HUNDREDS OR THOUSANDS OF

3890
02:27:16,387 --> 02:27:19,790
STUDIES, WHEN WE CAN'T SHARE RAW

3891
02:27:19,790 --> 02:27:22,193
DATA, SIMILARLY WITH GEO

3892
02:27:22,193 --> 02:27:22,460
LOCATION.

3893
02:27:22,460 --> 02:27:24,395
PARTLY WHY I VISUALIZE LOCATION

3894
02:27:24,395 --> 02:27:26,363
DATA THE WAY I DID WAS WE'RE

3895
02:27:26,363 --> 02:27:28,399
ALWAYS TRYING TO COME UP WITH

3896
02:27:28,399 --> 02:27:30,534
WAYS RESPECTING GRANULARITY OF

3897
02:27:30,534 --> 02:27:34,472
THE LOCATION DATA WITHOUT

3898
02:27:34,472 --> 02:27:39,510
SHARING THE ACTUAL X, Y

3899
02:27:39,510 --> 02:27:39,844
COORDINATES.

3900
02:27:39,844 --> 02:27:41,078
I DON'T HAVE A GREAT ANSWER BUT

3901
02:27:41,078 --> 02:27:45,316
THIS IS  ONE OF THE CHALLENGES,

3902
02:27:45,316 --> 02:27:47,585
ENCOURAGING PEOPLE TO SHARE

3903
02:27:47,585 --> 02:27:49,019
DE-IDENTIFIED FEATURES IN PUBLIC

3904
02:27:49,019 --> 02:27:51,522
DATASETS LIKE THE NIH DATA

3905
02:27:51,522 --> 02:27:53,824
ARCHIVE OR DO SOME KIND OF

3906
02:27:53,824 --> 02:27:55,159
FEDERATED LEARNING WHERE PEOPLE

3907
02:27:55,159 --> 02:27:57,561
CAN, YOU KNOW, KEEP THE

3908
02:27:57,561 --> 02:28:00,831
SENSITIVE DATA ON THEIR LOCAL

3909
02:28:00,831 --> 02:28:01,599
HIPAA-COMPLIANT SYSTEMS AND

3910
02:28:01,599 --> 02:28:04,668
STILL BE ABLE TO CREATE THOSE

3911
02:28:04,668 --> 02:28:07,838
POPULATION-LEVEL DISTRIBUTIONS. 

3912
02:28:07,838 --> 02:28:11,142
>> I WOULD JUST ADD
QUICKLY,

3913
02:28:11,142 --> 02:28:11,909
MAYBE A LITTLE CONTROVERSIAL,

3914
02:28:11,909 --> 02:28:13,611
ONE OF THE THINGS WE HAVE TO BE

3915
02:28:13,611 --> 02:28:16,213
CAREFUL OF IS I THINK SOMETIMES

3916
02:28:16,213 --> 02:28:20,451
WHEN WE TRY TO OVERPROTECT, YOU

3917
02:28:20,451 --> 02:28:22,853
KNOW, PSYCHIATRIC DATA OR

3918
02:28:22,853 --> 02:28:24,922
PSYCHIATRIC PATIENTS, WE

3919
02:28:24,922 --> 02:28:26,457
CONTRIBUTE TO STIGMATIZATION

3920
02:28:26,457 --> 02:28:26,724
ACTUALLY.

3921
02:28:26,724 --> 02:28:28,959
AND, YOU KNOW, IF WE OVERPROTECT

3922
02:28:28,959 --> 02:28:30,594
TO THE POINT THAT YOU CAN'T DO

3923
02:28:30,594 --> 02:28:32,596
THE WORK, THEN NO ONE IS GOING

3924
02:28:32,596 --> 02:28:34,865
TO EVER DO RESEARCH ON

3925
02:28:34,865 --> 02:28:36,667
PSYCHIATRIC CONDITIONS, RIGHT?

3926
02:28:36,667 --> 02:28:37,835
PEOPLE NEVER USED TO DO SUICIDE

3927
02:28:37,835 --> 02:28:40,471
RESEARCH BECAUSE IT WAS TOO

3928
02:28:40,471 --> 02:28:41,539
HARD.

3929
02:28:41,539 --> 02:28:43,874
RESEARCH DR. BAKER DOES ON

3930
02:28:43,874 --> 02:28:44,642
SCHIZOPHRENIC PATIENTS, PEOPLE

3931
02:28:44,642 --> 02:28:45,843
DIDN'T DO THAT WORK BECAUSE IT

3932
02:28:45,843 --> 02:28:55,386
WAS HOO HARD. -- TOO HARD, EVERY

3933
02:28:55,386 --> 02:28:57,121
EVERYONE WAS SO OVERPROTECTIVE.

3934
02:28:57,121 --> 02:28:58,522
WE HAVE A BAD HISTORY BUT WE

3935
02:28:58,522 --> 02:29:01,492
HAVE TO BE CAREFUL NOT TO BE TOO

3936
02:29:01,492 --> 02:29:02,793
EXCEPTIONALLISTIC WITH THIS TYPE

3937
02:29:02,793 --> 02:29:05,629
OF DATA ALSO.

3938
02:29:05,629 --> 02:29:08,499
>> TO FOLLOW ON THAT
QUICKLY,

3939
02:29:08,499 --> 02:29:11,235
WE'VE DONE A LOT OF INTERVIEWS

3940
02:29:11,235 --> 02:29:12,803
WITH PATIENTS IN OUR STUDIES.

3941
02:29:12,803 --> 02:29:14,772
VERY FEW OF THE PATIENTS

3942
02:29:14,772 --> 02:29:17,408
THEMSELVES ARE CONCERNED ABOUT

3943
02:29:17,408 --> 02:29:19,276
THE PRIVACY AND CONFIDENTIALITY.

3944
02:29:19,276 --> 02:29:21,879
MORE OFTEN THAN NOT INSTITUTIONS

3945
02:29:21,879 --> 02:29:22,780
ARE BOTHERED ABOUT BEING SUED

3946
02:29:22,780 --> 02:29:23,647
AND SO ON.

3947
02:29:23,647 --> 02:29:26,717
I DO THINK WE NEED TO BE CAREFUL

3948
02:29:26,717 --> 02:29:29,653
NOT TO SO-CALLED PATERNALLIZE

3949
02:29:29,653 --> 02:29:31,188
OUR PATIENTS AND THESE

3950
02:29:31,188 --> 02:29:31,455
DECISIONS.

3951
02:29:31,455 --> 02:29:32,056
MOST PATIENTS THAT PARTICIPATE

3952
02:29:32,056 --> 02:29:33,390
ARE MORE THAN HAPPY TO SHARE

3953
02:29:33,390 --> 02:29:34,892
THAT DATA WITH OTHERS IF IT

3954
02:29:34,892 --> 02:29:40,264
HELPS SOMEONE.

3955
02:29:40,264 --> 02:29:40,531
>> JOHN?

3956
02:29:40,531 --> 02:29:43,567
>> THANKS TO BOTH OF YOU
FOR

3957
02:29:43,567 --> 02:29:46,203
STIMULATING PRESENTATIONS.

3958
02:29:46,203 --> 02:29:49,907
A COUPLE THINGS PIQUE  MY

3959
02:29:49,907 --> 02:29:51,642
INTEREST, ONE IS THE ISSUE OF

3960
02:29:51,642 --> 02:29:52,443
INFORMED CONSENT.

3961
02:29:52,443 --> 02:29:56,480
I'M STILL LEARNING, THE MORE I'M

3962
02:29:56,480 --> 02:29:59,850
REALIZING HOW LOADED A TERM IT

3963
02:29:59,850 --> 02:30:00,084
IS.

3964
02:30:00,084 --> 02:30:02,052
DO YOU CONSENT TO A.I. BEING

3965
02:30:02,052 --> 02:30:03,254
INVOLVED IN LOOKING AT YOUR

3966
02:30:03,254 --> 02:30:04,788
DATA, WHAT DOES THAT MEAN?

3967
02:30:04,788 --> 02:30:05,789
IT COULD MEAN ANYTHING TO

3968
02:30:05,789 --> 02:30:06,223
ANYBODY.

3969
02:30:06,223 --> 02:30:08,192
A.I. TODAY IS NOT GOING TO BE

3970
02:30:08,192 --> 02:30:09,827
A.I. TOMORROW.

3971
02:30:09,827 --> 02:30:11,362
IT'S AN INTERESTING THOUGHT.

3972
02:30:11,362 --> 02:30:13,430
ONE THING THAT YOU BROUGHT UP,

3973
02:30:13,430 --> 02:30:15,966
THIS INTERESTING IDEA OF RETURN

3974
02:30:15,966 --> 02:30:17,067
OF INDIVIDUAL RESEARCH RESULTS,

3975
02:30:17,067 --> 02:30:19,470
IT KIND OF JOGGED MY MEMORY

3976
02:30:19,470 --> 02:30:20,905
ABOUT THIS QUESTION AND MAYBE

3977
02:30:20,905 --> 02:30:22,206
IT'S A NAIVE QUESTIONS, WHO

3978
02:30:22,206 --> 02:30:23,073
ACTUALLY OWNS THE DATA AT THE

3979
02:30:23,073 --> 02:30:24,508
END OF THE DAY, RIGHT?

3980
02:30:24,508 --> 02:30:26,477
THIS I THINK WE TALK ABOUT DATA

3981
02:30:26,477 --> 02:30:27,878
PROVENANCE AND THINGS LIKE THAT

3982
02:30:27,878 --> 02:30:32,182
BUT WHO ACTUALLY OWNS THE DATA

3983
02:30:32,182 --> 02:30:34,251
AND MIGHT BE ENTITLED TO BENEFIT

3984
02:30:34,251 --> 02:30:35,886
FROM IT IN WAYS THAT MAYBE

3985
02:30:35,886 --> 02:30:37,855
SOMEBODY SHOULD OR SHOULDN'T BE

3986
02:30:37,855 --> 02:30:39,390
TELLING THEM THAT THEY MIGHT,

3987
02:30:39,390 --> 02:30:42,893
AND THEN AS YOU INVOLVE THESE

3988
02:30:42,893 --> 02:30:46,063
LARGE MODELS, YOU KNOW, MY DATA

3989
02:30:46,063 --> 02:30:48,933
IS MIXING WITH OTHERS, IT'S A

3990
02:30:48,933 --> 02:30:51,769
FASCINATING SET OF BLURRY

3991
02:30:51,769 --> 02:30:52,102
BOUNDARIES.

3992
02:30:52,102 --> 02:30:55,172
LOVE TO HEAR YOUR THOUGHTS ON

3993
02:30:55,172 --> 02:30:56,607
THAT.

3994
02:30:56,607 --> 02:30:59,109
>> IT'S A REALLY GREAT
QUESTION.

3995
02:30:59,109 --> 02:31:00,444
I THINK DATA OWNERSHIP, IT'S NOT

3996
02:31:00,444 --> 02:31:03,280
JUST IN THE CONTEXT OF THIS

3997
02:31:03,280 --> 02:31:12,856
CONVERSATION, THIS COMES UP ON A

3998
02:31:12,856 --> 02:31:13,924
WEEK TO-WEEK BASIS.

3999
02:31:13,924 --> 02:31:15,793
I THINK SOMETIMES IT DEPENDS.

4000
02:31:15,793 --> 02:31:18,529
A CONCRETE ANSWER, SOMETIMES IT

4001
02:31:18,529 --> 02:31:21,699
DEPENDS ON WHO FUNDED THE WORK,

4002
02:31:21,699 --> 02:31:26,403
AND WHO COLLECTED THE DATA IN

4003
02:31:26,403 --> 02:31:29,273
TERMS OF WHICH THIRD-PARTY APP.

4004
02:31:29,273 --> 02:31:31,909
THE ACTUAL ANSWER CAN VARY

4005
02:31:31,909 --> 02:31:36,246
WIDELY, BUT I THINK FROM A

4006
02:31:36,246 --> 02:31:37,448
STANDPOINT OF INDIVIDUAL

4007
02:31:37,448 --> 02:31:37,948
RESEARCHERS THERE'S AN

4008
02:31:37,948 --> 02:31:40,751
INTERESTING PIECE IF YOU'RE A

4009
02:31:40,751 --> 02:31:43,287
HIPAA-COVERED ENTITY AND YOUR

4010
02:31:43,287 --> 02:31:45,456
WORK IS -- YOUR RESEARCH RECORDS

4011
02:31:45,456 --> 02:31:49,626
ARE PART OF YOUR DESIGNATED

4012
02:31:49,626 --> 02:31:51,095
RECORDS SET, RESEARCH

4013
02:31:51,095 --> 02:31:53,364
PARTICIPANTS ACTUALLY HAVE A

4014
02:31:53,364 --> 02:31:55,632
RIGHT TO ACCESS THEIR RECORDS,

4015
02:31:55,632 --> 02:31:56,066
RIGHT?

4016
02:31:56,066 --> 02:31:58,836
AND SO WE TALKED ABOUT THIS

4017
02:31:58,836 --> 02:32:00,571
ACTUALLY WITH PEOPLE WHO ARE

4018
02:32:00,571 --> 02:32:03,207
MORE EXPERT THAN I ON THE LAWS

4019
02:32:03,207 --> 02:32:07,444
AND REGULATIONS THERE BUT WE

4020
02:32:07,444 --> 02:32:08,645
TALKED ABOUT THIS FINE LINE

4021
02:32:08,645 --> 02:32:10,314
BETWEEN SAYING YOU'RE NOT

4022
02:32:10,314 --> 02:32:12,149
ALLOWED TO RETURN THOSE RESEARCH

4023
02:32:12,149 --> 02:32:15,986
RESULTS BECAUSE YOU MAY BE

4024
02:32:15,986 --> 02:32:19,390
VIOLATING SOME SORT OF OTHER

4025
02:32:19,390 --> 02:32:20,591
REGULATION OR ETHICAL NORM

4026
02:32:20,591 --> 02:32:21,558
AROUND, WELL, DOESN'T THE

4027
02:32:21,558 --> 02:32:22,826
PATIENT OWN THEIR OWN DATA AND

4028
02:32:22,826 --> 02:32:26,730
SHOULDN'T THEY HAVE THE RIGHT TO

4029
02:32:26,730 --> 02:32:29,900
SEE IT, SO THERE'S A TIGHTROPE

4030
02:32:29,900 --> 02:32:31,135
THERE THAT WE'RE WALKING WHERE,

4031
02:32:31,135 --> 02:32:34,171
YOU KNOW, WE'RE TRYING TO MAKE

4032
02:32:34,171 --> 02:32:35,706
DECISIONS TO ONLY SHARE DATA

4033
02:32:35,706 --> 02:32:39,643
WHEN WE KNOW THAT IT'S GOOD DATA

4034
02:32:39,643 --> 02:32:41,178
BASICALLY AND SO SOMEBODY ACTS

4035
02:32:41,178 --> 02:32:44,014
ON IT BUT ALSO RECOGNIZING THAT

4036
02:32:44,014 --> 02:32:45,949
WE HAVE TO BE CAREFUL, WHEN WE

4037
02:32:45,949 --> 02:32:53,891
SAY WE CAN'T.

4038
02:32:53,891 --> 02:32:54,224
>> GREAT.

4039
02:32:54,224 --> 02:32:54,458
SAMEER?

4040
02:32:54,458 --> 02:32:56,193
>> YEAH, THANKS.

4041
02:32:56,193 --> 02:32:58,495
TWO COMMENTS.

4042
02:32:58,495 --> 02:33:00,664
ONE QUICK ONE, TO ECHO JOHN SAID

4043
02:33:00,664 --> 02:33:04,401
A SECOND AGO ABOUT INFORMED

4044
02:33:04,401 --> 02:33:06,570
CONSENT, DR. SILVERMAN YOU'RE

4045
02:33:06,570 --> 02:33:07,905
TALKING ABOUT IN OUR LIVES WE

4046
02:33:07,905 --> 02:33:11,075
USE A.I. ALL THE TIME,

4047
02:33:11,075 --> 02:33:12,943
ACCEPTABILITY FOR MAKING

4048
02:33:12,943 --> 02:33:15,012
DECISIONS IS PROBABLY GOING TO

4049
02:33:15,012 --> 02:33:17,981
BE FAIRLY HIGH, TOTALLY AGREE.

4050
02:33:17,981 --> 02:33:20,050
JOHN, LIKE YOU SAID, THERE'S NO

4051
02:33:20,050 --> 02:33:21,018
SUCH THING AS INFORMED CONSENT

4052
02:33:21,018 --> 02:33:24,321
WHEN IT COMES TO THIS.

4053
02:33:24,321 --> 02:33:27,257
IT'S REALLY KIND OF POSSIBLE

4054
02:33:27,257 --> 02:33:32,062
COMPUTATIONALLY TO TRACE THE

4055
02:33:32,062 --> 02:33:33,163
BREAD CRUMBS FROM INPUTS.

4056
02:33:33,163 --> 02:33:43,707
I'M OKAY WITH THIS AND NOT THAT.

4057
02:33:47,010 --> 02:33:50,447
, THE SECOND COMMENT, I WANTED

4058
02:33:50,447 --> 02:33:52,483
TO FOCUS TOWARDS THE END YOU

4059
02:33:52,483 --> 02:33:57,121
TALKED ABOUT HOW, YOU KNOW, CASE

4060
02:33:57,121 --> 02:33:59,857
EXAMPLE I THINK WITH THE

4061
02:33:59,857 --> 02:34:01,191
PATIENTS WITH IMPLANTED DEVICE

4062
02:34:01,191 --> 02:34:03,494
ABOUT USING NOW A MUCH RICHER

4063
02:34:03,494 --> 02:34:05,162
STREAM OF INFORMATION AS YOU

4064
02:34:05,162 --> 02:34:10,501
SHOW IN TERMS OF BEHAVIORAL DATA

4065
02:34:10,501 --> 02:34:13,137
TELLING YOU REALLY PRECISELY HOW

4066
02:34:13,137 --> 02:34:15,005
ACTIVE YOU ARE, SOCIAL, MOBILE,

4067
02:34:15,005 --> 02:34:17,307
WHAT YOU'RE DOING.

4068
02:34:17,307 --> 02:34:20,711
YOU KNOW, THE CONCEPT OF USING

4069
02:34:20,711 --> 02:34:23,447
THOSE DATA IN A CLOSED LOOP

4070
02:34:23,447 --> 02:34:26,283
LET'S SAY MANNER TO TELL A

4071
02:34:26,283 --> 02:34:29,653
DEVICE HOW TO DELIVER

4072
02:34:29,653 --> 02:34:31,522
STIMULATION THAT'S GOING TO

4073
02:34:31,522 --> 02:34:34,491
CHANGE HOW SOMEONE BEHAVES OR

4074
02:34:34,491 --> 02:34:36,460
EMOTES, THIS IS ANOTHER QUESTION

4075
02:34:36,460 --> 02:34:41,298
ABOUT HOW TO USE A.I. OR JUST

4076
02:34:41,298 --> 02:34:43,033
KIND OF DEEP PHENOTYPE BEHAVIOR,

4077
02:34:43,033 --> 02:34:49,606
I THINK IT'S TRICKY. 

4078
02:34:49,606 --> 02:34:52,876
THERE'S THIS RUSH TOWARDS THIS

4079
02:34:52,876 --> 02:34:55,712
HIGH STANDARD OF CLOSED-LOOP

4080
02:34:55,712 --> 02:34:56,647
NEUROMODULATION WHERE AUTONOMY

4081
02:34:56,647 --> 02:34:59,049
IS BETTER, THAT'S ONE THING.

4082
02:34:59,049 --> 02:35:02,419
IF YOU'RE DEALING WITH TREMORS,

4083
02:35:02,419 --> 02:35:03,387
PARKINSON'S DISEASE, DECISION

4084
02:35:03,387 --> 02:35:06,356
MAKING, OCD OR WHAT NOT, THE

4085
02:35:06,356 --> 02:35:07,457
INFORMATION IS CRITICAL AND I

4086
02:35:07,457 --> 02:35:12,262
LOVE HOW YOU PUT ANYTIME T

4087
02:35:12,262 --> 02:35:13,497
ERMS -- PUT IT IN TERMS OF

4088
02:35:13,497 --> 02:35:14,765
ALLOWING THAT TO BE AVAILABLE TO

4089
02:35:14,765 --> 02:35:16,533
A TEAM, HOW MUCH THAT GETS

4090
02:35:16,533 --> 02:35:18,502
INFORMATION WITH AN ALGORITHM TO

4091
02:35:18,502 --> 02:35:20,571
MAKE DECISIONS ABOUT THERAPY

4092
02:35:20,571 --> 02:35:23,307
DELIVERY IS A SEPARATE QUESTION.

4093
02:35:23,307 --> 02:35:25,375
I'D LOVE TO HEAR YOUR THOUGHTS

4094
02:35:25,375 --> 02:35:26,810
ON THIS.

4095
02:35:26,810 --> 02:35:31,515
>> YES, THAT'S A GREAT --
BOTH

4096
02:35:31,515 --> 02:35:32,749
COMMENTS ARE ON POINT.

4097
02:35:32,749 --> 02:35:34,384
IN TERMS OF THE SECOND ONE I

4098
02:35:34,384 --> 02:35:39,423
THINK THIS IDEA OF A CLOSED LOOP

4099
02:35:39,423 --> 02:35:41,091
SYSTEM WHICH TAKES BEHAVIOR IN,

4100
02:35:41,091 --> 02:35:43,894
MAKES A CHANGE TO THE BRAIN,

4101
02:35:43,894 --> 02:35:47,397
IT'S OBVIOUSLY ONE OF THE HOLY

4102
02:35:47,397 --> 02:35:50,033
GRAILS OF, YOU KNOW, IMPLANTED

4103
02:35:50,033 --> 02:35:50,267
SYSTEMS.

4104
02:35:50,267 --> 02:35:54,605
I WAY I THINK ABOUT IT IS FOR AT

4105
02:35:54,605 --> 02:35:55,706
LEAST THE WAY CLINICAL MEDICINE

4106
02:35:55,706 --> 02:35:59,343
WORKS IS WE ARE DOING THAT.

4107
02:35:59,343 --> 02:36:04,248
WE TAKE INPUTS, TRY TO MAKE A

4108
02:36:04,248 --> 02:36:05,649
CHANGE, CONSTANTLY ITERATING, WE

4109
02:36:05,649 --> 02:36:07,184
DON'T KNOW HOW A PARTICULAR

4110
02:36:07,184 --> 02:36:08,952
MEDICINE IS GOING TO AFFECT

4111
02:36:08,952 --> 02:36:09,720
SOMEONE, MAYBE SIDE EFFECTS AND

4112
02:36:09,720 --> 02:36:10,821
SO ON.

4113
02:36:10,821 --> 02:36:12,889
SO I GUESS MY INTUITION WOULD BE

4114
02:36:12,889 --> 02:36:15,726
THAT YOU DON'T AT LEAST AT THE

4115
02:36:15,726 --> 02:36:18,362
BEGINNING WANT A SYSTEM THAT'S

4116
02:36:18,362 --> 02:36:18,895
FULLY AUTONOMOUS.

4117
02:36:18,895 --> 02:36:20,530
YOU WANT TO WORK WITH A HUMAN IN

4118
02:36:20,530 --> 02:36:23,600
THE LOOP TO CONTINUE TO TWEAK

4119
02:36:23,600 --> 02:36:24,801
THE INPUTS.

4120
02:36:24,801 --> 02:36:27,104
AND KIND OF EXPLORATORY

4121
02:36:27,104 --> 02:36:28,372
EXPLOITED WAY SO YOU SETTLE ON

4122
02:36:28,372 --> 02:36:30,507
SOMETHING THAT SEEMS TO BE

4123
02:36:30,507 --> 02:36:34,244
WORKING, MAYBE MOVE IT MORE INTO

4124
02:36:34,244 --> 02:36:36,513
AN AUTONOMOUS MODE BUT IN EVERY

4125
02:36:36,513 --> 02:36:39,349
SITUATION YOU WANT TO HAVE

4126
02:36:39,349 --> 02:36:40,350
PERIODIC EXPLORATION, IS THIS

4127
02:36:40,350 --> 02:36:41,885
STILL DOING THE THING WE

4128
02:36:41,885 --> 02:36:43,186
EXPECTED IT TO DO, WHERE THERE'S

4129
02:36:43,186 --> 02:36:47,257
A HUMAN IN THE LOOP SAYING --

4130
02:36:47,257 --> 02:36:48,225
WHETHER THAT'S THE PATIENT

4131
02:36:48,225 --> 02:36:50,927
SAYING NO, THIS NO LONGER FEELS

4132
02:36:50,927 --> 02:36:52,829
GOOD, OR THE CLINICIAN, THIS IS

4133
02:36:52,829 --> 02:36:54,364
GOING TO CAUSE SIDE EFFECTS.

4134
02:36:54,364 --> 02:36:55,999
I'M NOT SURE WE'RE EVER GOING TO

4135
02:36:55,999 --> 02:36:58,602
GET TO A POINT SIMILAR WITH

4136
02:36:58,602 --> 02:37:00,904
SELF-DRIVING CARS, LIKE A FULLY

4137
02:37:00,904 --> 02:37:02,339
AUTONOMOUS SYSTEM.

4138
02:37:02,339 --> 02:37:03,640
I THINK THERE'S ALWAYS GOING TO

4139
02:37:03,640 --> 02:37:05,442
BE EDGE OR CORNER CASES WE

4140
02:37:05,442 --> 02:37:07,044
DIDN'T ANTICIPATE, WHERE THE

4141
02:37:07,044 --> 02:37:08,779
BRAIN OR THE BEHAVIOR OR THE

4142
02:37:08,779 --> 02:37:10,881
ENVIRONMENT IS CHANGING IN SUCH

4143
02:37:10,881 --> 02:37:12,382
A WAY WE'RE ACTUALLY OUTSIDE OF

4144
02:37:12,382 --> 02:37:14,584
WHAT OUR MODEL IS DESIGNED FOR

4145
02:37:14,584 --> 02:37:17,321
AND HAVE TO RECALIBRATE.

4146
02:37:17,321 --> 02:37:19,056
THESE ARE KEY THOUGHTS AS WE TRY

4147
02:37:19,056 --> 02:37:26,663
TO MOVE MORE TOWARDS THE CLOSED

4148
02:37:26,663 --> 02:37:28,498
LOOP SYSTEMS.

4149
02:37:28,498 --> 02:37:29,599
>> THANKS A LOT.

4150
02:37:29,599 --> 02:37:31,234
WE CAN ALSO QUESTION WHETHER

4151
02:37:31,234 --> 02:37:37,174
THAT'S A HOLY GRAIL, YOU KNOW,

4152
02:37:37,174 --> 02:37:40,577
OR NOT.

4153
02:37:40,577 --> 02:37:41,111
>> ABSOLUTELY.

4154
02:37:41,111 --> 02:37:41,445
>> AMY?

4155
02:37:41,445 --> 02:37:44,314
YEAH, THANK YOU BOTH SO MUCH FOR

4156
02:37:44,314 --> 02:37:44,848
EXCELLENT PRESENTATIONS.

4157
02:37:44,848 --> 02:37:46,383
FIRST I WANT TO START BY MAKING

4158
02:37:46,383 --> 02:37:48,585
A COMMENT THAT I APPRECIATE

4159
02:37:48,585 --> 02:37:49,353
CHRISTINE'S FRAMING AND

4160
02:37:49,353 --> 02:37:50,554
DISCUSSION AROUND LIKE WHAT IS

4161
02:37:50,554 --> 02:37:52,622
NEW HERE AND HOW CAN WE LEARN

4162
02:37:52,622 --> 02:37:54,725
FROM OTHER AREAS AND I'M

4163
02:37:54,725 --> 02:37:55,792
THINKING PARTICULARLY HOW MANY

4164
02:37:55,792 --> 02:37:57,127
OF THESE ISSUES THAT WE'RE

4165
02:37:57,127 --> 02:38:01,598
TALKING ABOUT NOW HAVE BEEN WELL

4166
02:38:01,598 --> 02:38:04,134
RESEARCHED, DEBATED, FIELD OF

4167
02:38:04,134 --> 02:38:05,202
GENETICS AND GENOMICS.

4168
02:38:05,202 --> 02:38:10,907
I APPRECIATED DR. SILVERMAN HOW

4169
02:38:10,907 --> 02:38:13,009
AND WHERE TO FRAMEWORKS FALL

4170
02:38:13,009 --> 02:38:15,379
SHORT AND WHY, WHAT'S UNIQUE, IN

4171
02:38:15,379 --> 02:38:17,814
THE CHAT SOME CONVERSATIONS

4172
02:38:17,814 --> 02:38:19,783
AROUND ISSUES OF CONSENT AND

4173
02:38:19,783 --> 02:38:22,152
CHALLENGES OF FUTURE UNCERTAINTY

4174
02:38:22,152 --> 02:38:23,620
IS SOMETHING THAT HAS BEEN WELL

4175
02:38:23,620 --> 02:38:25,689
STUDIED AND DO WE NEED RECONSENT

4176
02:38:25,689 --> 02:38:27,657
AND WHAT DO WE DO WITH MINORS

4177
02:38:27,657 --> 02:38:30,026
AND THOSE THINGS HAVE BEEN WELL

4178
02:38:30,026 --> 02:38:31,695
STUDIED IN THE FIELD OF

4179
02:38:31,695 --> 02:38:33,997
BIOETHICS AND OTHER AREAS.

4180
02:38:33,997 --> 02:38:34,998
SO JUST GENERAL ENCOURAGEMENT

4181
02:38:34,998 --> 02:38:38,068
FOR US TO LOOK AT THAT AND

4182
02:38:38,068 --> 02:38:39,836
REALLY PARSE OUT WHAT'S NEW OR

4183
02:38:39,836 --> 02:38:42,305
DIFFERENT HERE IN TERMS OF WHAT

4184
02:38:42,305 --> 02:38:46,276
NEEDS TO CHANGE.

4185
02:38:46,276 --> 02:38:48,345
TWO OTHER QUESTIONS/COMMENTS,

4186
02:38:48,345 --> 02:38:51,415
WHEN TALKING ABOUT PRIVACY, YOU

4187
02:38:51,415 --> 02:38:54,017
KNOW, SOME RESEARCH AROUND BRAIN

4188
02:38:54,017 --> 02:38:56,086
PRIVACY WE'VE FUND TO MY

4189
02:38:56,086 --> 02:38:57,754
SURPRISE PARTICIPANTS THAT WE'VE

4190
02:38:57,754 --> 02:38:59,389
SURVEYED ABOUT THIS ARE LESS

4191
02:38:59,389 --> 02:39:00,490
CONCERNED ABOUT THEIR BRAIN DATA

4192
02:39:00,490 --> 02:39:03,660
THAN THEY ARE THINGS LIKE TEXT

4193
02:39:03,660 --> 02:39:05,729
MESSAGE, THE CONTENT OF TEXT

4194
02:39:05,729 --> 02:39:07,497
MESSAGES AND THINGS LIKE THAT.

4195
02:39:07,497 --> 02:39:11,201
DR. BAKER MENTIONED THAT IT'S

4196
02:39:11,201 --> 02:39:13,303
VERY DIFFICULT TO OPT OUT OF THE

4197
02:39:13,303 --> 02:39:14,504
COLLECTION OR USE OF CERTAIN

4198
02:39:14,504 --> 02:39:18,642
TYPES OF DATA WHEN YOU HAVE THIS

4199
02:39:18,642 --> 02:39:21,812
HOLISTIC APPROACH TO A.I.-BASED

4200
02:39:21,812 --> 02:39:23,346
DATA COLLECTION AND I'M

4201
02:39:23,346 --> 02:39:25,449
WONDERING IF YOU HAVE THOUGHTS

4202
02:39:25,449 --> 02:39:27,417
IS AN OPT-OUT POSSIBLE IF

4203
02:39:27,417 --> 02:39:28,385
SOMEBODY SAYS I'M COMFORTABLE

4204
02:39:28,385 --> 02:39:31,354
BUT DON'T WANT YOU TRACKING X,

4205
02:39:31,354 --> 02:39:32,656
RIGHT?

4206
02:39:32,656 --> 02:39:34,524
EITHER MY TEXT MESSAGES OR MY --

4207
02:39:34,524 --> 02:39:36,159
WHATEVER IT IS.

4208
02:39:36,159 --> 02:39:38,028
IS THERE A WAY TECHNICALLY AND

4209
02:39:38,028 --> 02:39:39,329
ETHICALLY TO HAVE THEM OPT OUT

4210
02:39:39,329 --> 02:39:39,963
OF THAT?

4211
02:39:39,963 --> 02:39:42,833
AND THEN I GUESS MY SECOND

4212
02:39:42,833 --> 02:39:45,669
QUESTION REALLY RELATES TO SORT

4213
02:39:45,669 --> 02:39:47,437
OF THIS ACTIONABILITY ISSUE THAT

4214
02:39:47,437 --> 02:39:48,405
SAMEER BROUGHT UP.

4215
02:39:48,405 --> 02:39:49,940
ANOTHER THING WE'VE FOUND IN OUR

4216
02:39:49,940 --> 02:39:51,141
RESEARCH THAT OTHERS HAVE FOUND

4217
02:39:51,141 --> 02:39:54,411
AS WELL IS THAT WHEN WE TALK TO

4218
02:39:54,411 --> 02:39:58,582
PEOPLE WHO HAVE MENTAL HEALTH

4219
02:39:58,582 --> 02:40:00,784
DISORDER AND WE ASK ABOUT USING

4220
02:40:00,784 --> 02:40:02,319
AND SHARING BRAIN DATA THEY HAVE

4221
02:40:02,319 --> 02:40:05,822
SAID THAT'S THE WHOLE POINT OF

4222
02:40:05,822 --> 02:40:06,790
PSYCHIATRY, TRYING TO UNDERSTAND

4223
02:40:06,790 --> 02:40:08,325
WHAT'S GOING ON IN MY BRAIN SO

4224
02:40:08,325 --> 02:40:12,262
THE BETTER YOU CAN DO THAT, THE

4225
02:40:12,262 --> 02:40:14,798
BETTER IT IS TO COME UP WITH A

4226
02:40:14,798 --> 02:40:16,867
TREATMENT FOR ME.

4227
02:40:16,867 --> 02:40:18,168
THIS IDEA OF ACTIONABILITY, MY

4228
02:40:18,168 --> 02:40:20,270
QUESTION IS FOR DR. SILVERMAN,

4229
02:40:20,270 --> 02:40:22,572
WHETHER YOUR ANALYSIS AND

4230
02:40:22,572 --> 02:40:24,441
RECOMMENDATIONS ABOUT RETURN OF

4231
02:40:24,441 --> 02:40:25,642
RESULTS WOULD CHANGE DEPENDING

4232
02:40:25,642 --> 02:40:27,177
ON THE ACTIONABILITY AND DOES

4233
02:40:27,177 --> 02:40:29,045
THAT FIT INTO YOUR RISK/BENEFIT

4234
02:40:29,045 --> 02:40:35,485
ANALYSIS OR IS THAT CONSIDERED

4235
02:40:35,485 --> 02:40:35,919
DIFFERENTLY?

4236
02:40:35,919 --> 02:40:37,687
>> ALL GREAT QUESTIONS AND

4237
02:40:37,687 --> 02:40:39,322
POINTS, CERTAINLY APPRECIATE IT.

4238
02:40:39,322 --> 02:40:43,159
I THINK THAT WHEN WE THINK ABOUT

4239
02:40:43,159 --> 02:40:45,262
ACTIONABILITY IT HAD TO DO WITH

4240
02:40:45,262 --> 02:40:46,563
ESSENTIALLY I GUESS I WOULD SAY

4241
02:40:46,563 --> 02:40:48,765
TWO THINGS.

4242
02:40:48,765 --> 02:40:51,935
ONE IS THE ACUITY OF WHAT YOU'RE

4243
02:40:51,935 --> 02:40:57,073
POTENTIALLY PICKING UP ON, IS

4244
02:40:57,073 --> 02:40:59,609
THE ALGORITHM TELLING ME THAT

4245
02:40:59,609 --> 02:41:00,777
THIS PERSON WITH DEPRESSION

4246
02:41:00,777 --> 02:41:02,479
SPENT FIVE HOURS AT A LIQUOR

4247
02:41:02,479 --> 02:41:05,515
STORE AND NOW ON A BRIDGE FOR

4248
02:41:05,515 --> 02:41:07,050
TWO HOURS, JUST SEND A HUNDRED

4249
02:41:07,050 --> 02:41:07,918
TEXT MESSAGES IN 25 MINUTES,

4250
02:41:07,918 --> 02:41:12,389
THAT WOULD BE A SITUATION OF

4251
02:41:12,389 --> 02:41:15,358
PRESUMED VERY HIGH

4252
02:41:15,358 --> 02:41:16,359
URGENCY/ACTIONABILITY, ASSUMING

4253
02:41:16,359 --> 02:41:17,661
THERE'S ANALYTICAL AND CLINICAL

4254
02:41:17,661 --> 02:41:20,830
VALIDITY AND SO WE WOULD HAVE I

4255
02:41:20,830 --> 02:41:23,867
THINK A FAIRLY LOW BAR TO

4256
02:41:23,867 --> 02:41:25,302
WANTING TO, YOU KNOW, RETURN

4257
02:41:25,302 --> 02:41:26,603
THOSE RESULTS, WITH THE CAVEAT

4258
02:41:26,603 --> 02:41:28,905
THAT I MEAN ONE MIGHT ASK THE

4259
02:41:28,905 --> 02:41:30,040
QUESTION OR ARGUE WHY WOULDN'T

4260
02:41:30,040 --> 02:41:32,008
YOU JUST ALWAYS RETURN THOSE

4261
02:41:32,008 --> 02:41:33,743
RESULTS?  WHY IS THERE EVEN

4262
02:41:33,743 --> 02:41:33,977
CONCERN?

4263
02:41:33,977 --> 02:41:35,378
OF COURSE THE PROBLEM IS THAT

4264
02:41:35,378 --> 02:41:37,581
THERE COULD BE FALSE POSITIVES,

4265
02:41:37,581 --> 02:41:40,083
AND THE SYSTEM IN WHICH WE

4266
02:41:40,083 --> 02:41:41,384
THE -- THE WORLD IN WHICH WE

4267
02:41:41,384 --> 02:41:44,387
LIVE IN, IF YOU CALL 911 ON

4268
02:41:44,387 --> 02:41:45,889
SOMEONE THE RESULT IS NOT ALWAYS

4269
02:41:45,889 --> 02:41:46,656
GOOD, RIGHT?

4270
02:41:46,656 --> 02:41:48,959
AND IF YOU GET HOSPITALIZED

4271
02:41:48,959 --> 02:41:50,293
INVOLUNTARY AGAINST YOUR WILL

4272
02:41:50,293 --> 02:41:52,128
THERE CAN BE TRAUMA AND

4273
02:41:52,128 --> 02:41:52,462
SEQUELAE.

4274
02:41:52,462 --> 02:41:55,098
SO IT'S NOT A PERFECT SYSTEM

4275
02:41:55,098 --> 02:41:57,167
WHERE WE HAVE -- I THINK THAT

4276
02:41:57,167 --> 02:41:58,134
PART OF THE ISSUE, THIS IS TO

4277
02:41:58,134 --> 02:41:59,803
ANSWER THE OTHER PART OF THE

4278
02:41:59,803 --> 02:42:02,539
QUESTION, CAN WE DO SOMETHING?

4279
02:42:02,539 --> 02:42:03,640
ACTIONABILITY IN MY VIEW ALSO

4280
02:42:03,640 --> 02:42:05,175
TIES INTO CAN WE DO SOMETHING

4281
02:42:05,175 --> 02:42:06,676
ABOUT IT, RIGHT?

4282
02:42:06,676 --> 02:42:09,079
THERE MAY BE PEOPLE -- I THINK

4283
02:42:09,079 --> 02:42:10,513
WE'RE -- IN THAT SITUATION

4284
02:42:10,513 --> 02:42:11,448
THERE'S PROBABLY SOMETHING WE

4285
02:42:11,448 --> 02:42:12,482
COULD DO.

4286
02:42:12,482 --> 02:42:16,086
I THINK IN SITUATIONS WHERE

4287
02:42:16,086 --> 02:42:16,853
SOMEONE HAS BIPOLAR DISORDER

4288
02:42:16,853 --> 02:42:19,422
STARTING TO RAMP UP AND GET

4289
02:42:19,422 --> 02:42:20,690
MANIC THERE'S PROBABLY THINGS WE

4290
02:42:20,690 --> 02:42:23,526
CAN DO BUT SOMEONE WHO HAS, YOU

4291
02:42:23,526 --> 02:42:25,261
KNOW, CHRONIC DEPRESSION THAT'S

4292
02:42:25,261 --> 02:42:26,563
HAVING VERY FREQUENT DEPRESSIVE

4293
02:42:26,563 --> 02:42:28,131
EPISODES, THERE MAY BE LESS WE

4294
02:42:28,131 --> 02:42:29,466
CAN DO AT TIMES, RIGHT?

4295
02:42:29,466 --> 02:42:31,067
JUST BASED ON THE STATE OF OUR

4296
02:42:31,067 --> 02:42:31,334
TREATMENT.

4297
02:42:31,334 --> 02:42:32,469
THAT'S WHEN I THINK ABOUT

4298
02:42:32,469 --> 02:42:37,307
ACTIONABILITY IN TERMS OF THE

4299
02:42:37,307 --> 02:42:43,446
DIFFERENT TIERS OF ACTIONABILITY

4300
02:42:43,446 --> 02:42:45,615
AND POTENTIAL OF FALSE POSITIVES

4301
02:42:45,615 --> 02:42:47,250
AND LIMITS OF WHAT WE CAN DO TO

4302
02:42:47,250 --> 02:42:48,918
HELP PEOPLE, IF THAT ANSWERS

4303
02:42:48,918 --> 02:42:50,020
YOUR QUESTION.

4304
02:42:50,020 --> 02:42:51,688
>> YEAH, THAT'S SUPER
HELPFUL.

4305
02:42:51,688 --> 02:42:52,989
THANK YOU.

4306
02:42:52,989 --> 02:42:53,790
>> I CAN JUST COMMENT
MAYBE ON

4307
02:42:53,790 --> 02:42:55,825
THE FIRST PART OF THE QUESTION

4308
02:42:55,825 --> 02:42:59,896
IN TERMS OF HOW DO WE LET PEOPLE

4309
02:42:59,896 --> 02:43:01,665
SELECTIVELY OPT OUT, THE SHORT

4310
02:43:01,665 --> 02:43:03,166
ANSWER IS, YES, WE HAVE ALL

4311
02:43:03,166 --> 02:43:06,670
SORTS OF WAYS WE CAN HELP

4312
02:43:06,670 --> 02:43:07,237
PARTICIPANTS OR PATIENTS

4313
02:43:07,237 --> 02:43:09,205
SELECTIVELY OPT OUT OF CERTAIN

4314
02:43:09,205 --> 02:43:11,174
DATA TYPES, FOR INSTANCE WE'RE

4315
02:43:11,174 --> 02:43:18,181
DOING A LARGE GLOBAL STUDY,

4316
02:43:18,181 --> 02:43:20,283
INDIVIDUALS WITH PRODROMAL

4317
02:43:20,283 --> 02:43:21,117
SCHIZOPHRENIA, OPTION OF

4318
02:43:21,117 --> 02:43:23,987
COLLECTING NO SENSOR DATA OR

4319
02:43:23,987 --> 02:43:25,055
COLLECTING EVERYTHING BUT

4320
02:43:25,055 --> 02:43:26,723
LOCATION DATA, IT'S WHAT WE

4321
02:43:26,723 --> 02:43:29,659
FOUND THAT VERY FEW PATIENTS OR

4322
02:43:29,659 --> 02:43:30,560
FEW PARTICIPANTS SELECTIVELY OPT

4323
02:43:30,560 --> 02:43:33,596
OUT OF LOCATION BUT THERE ARE

4324
02:43:33,596 --> 02:43:34,497
SOME WHO DO.

4325
02:43:34,497 --> 02:43:37,333
ALSO WE TRY TO TEACH THEM HOW

4326
02:43:37,333 --> 02:43:38,868
TO, YOU KNOW, YOU CAN GO INTO

4327
02:43:38,868 --> 02:43:41,271
YOUR PHONE AND TURN OFF LOCATION

4328
02:43:41,271 --> 02:43:42,205
ACCESS FOR GIVEN APPLICATION AND

4329
02:43:42,205 --> 02:43:45,208
SO ON SO THERE ARE WAYS TO DO

4330
02:43:45,208 --> 02:43:45,408
THAT.

4331
02:43:45,408 --> 02:43:47,077
THE OTHER POINT I'LL MAKE IS

4332
02:43:47,077 --> 02:43:48,812
THAT AS WE THINK ABOUT REALLY

4333
02:43:48,812 --> 02:43:50,914
SOME OF THESE IDEAS BRINGING

4334
02:43:50,914 --> 02:43:52,882
INTO A CLINICAL SETTING ONE IDEA

4335
02:43:52,882 --> 02:43:56,720
WAS THAT PERHAPS YOU CAN HAVE A

4336
02:43:56,720 --> 02:43:59,222
CLINICIAN AND PATIENT ONLY

4337
02:43:59,222 --> 02:44:00,023
COLLABORATIVELY TRACK A SPECIFIC

4338
02:44:00,023 --> 02:44:01,591
LOCATION AND NOT ALL LOCATIONS,

4339
02:44:01,591 --> 02:44:03,059
LIKE I WANT TO TRACK WHEN YOU GO

4340
02:44:03,059 --> 02:44:04,928
TO THE BAR BUT DON'T WANT TO

4341
02:44:04,928 --> 02:44:06,563
KNOW WHEN YOU GO TO SCHOOL OR

4342
02:44:06,563 --> 02:44:08,865
WHEN YOU'RE AT YOUR GIRLFRIEND'S

4343
02:44:08,865 --> 02:44:10,300
HOUSE, ET CETERA, WORKING ON

4344
02:44:10,300 --> 02:44:12,569
YOUR ALCOHOL USE DISORDER SO

4345
02:44:12,569 --> 02:44:14,003
OBVIOUSLY THE APP MIGHT COLLECT

4346
02:44:14,003 --> 02:44:14,771
MORE LOCATION DATA BUT IN TERMS

4347
02:44:14,771 --> 02:44:16,840
OF WHAT MAKES IT BACK TO THE

4348
02:44:16,840 --> 02:44:19,909
HUMAN REVIEWING IT WE CAN USE

4349
02:44:19,909 --> 02:44:22,979
VARIOUS FILTERING METHODS TO

4350
02:44:22,979 --> 02:44:23,747
SELECTIVELY CAPTURE CERTAIN

4351
02:44:23,747 --> 02:44:30,086
DATA, CERTAIN FEATURES FROM THAT

4352
02:44:30,086 --> 02:44:31,955
DATA.

4353
02:44:31,955 --> 02:44:32,489
>> GREAT.

4354
02:44:32,489 --> 02:44:33,289
KAREN?

4355
02:44:33,289 --> 02:44:35,492
>> THANK YOU FOR THOSE
GREAT

4356
02:44:35,492 --> 02:44:35,792
TALKS.

4357
02:44:35,792 --> 02:44:38,328
I HAD A QUESTION REALLY RELATED

4358
02:44:38,328 --> 02:44:41,397
TO WHAT AMY WAS SAYING THAT, YOU

4359
02:44:41,397 --> 02:44:43,900
KNOW, WHAT IS THE DIFFERENCE

4360
02:44:43,900 --> 02:44:45,335
BETWEEN THE STUDIES AND OTHERS,

4361
02:44:45,335 --> 02:44:48,271
I GUESS CHRISTINE ALSO POSED TO

4362
02:44:48,271 --> 02:44:51,808
US, BUT ALSO INVITING US TO

4363
02:44:51,808 --> 02:44:57,280
THINK A LITTLE MORE EMPIRICALLY

4364
02:44:57,280 --> 02:44:58,715
ABOUT OUR WHAT-IFs, WONDERFUL

4365
02:44:58,715 --> 02:45:01,518
THAT YOUR AUTHOR ON THE CHECK

4366
02:45:01,518 --> 02:45:03,286
LIST PAPER, YOU ALREADY TALKED

4367
02:45:03,286 --> 02:45:05,588
ABOUT THIS RECENT THING, YOU

4368
02:45:05,588 --> 02:45:07,757
HAVE AN IDEA HOW YOU WORK WITH

4369
02:45:07,757 --> 02:45:08,758
PATIENTS ONE ON ONE,

4370
02:45:08,758 --> 02:45:10,927
PARTICIPANTS ONE ON ONE,

4371
02:45:10,927 --> 02:45:13,029
SELECTING FROM THEIR PHONE.

4372
02:45:13,029 --> 02:45:15,465
DO YOU HAVE -- CURIOUS, ONE,

4373
02:45:15,465 --> 02:45:17,200
WHAT KIND OF RISKS YOU CURRENTLY

4374
02:45:17,200 --> 02:45:18,902
STATE IN YOUR CONSENT FORM AS

4375
02:45:18,902 --> 02:45:21,137
RELATES TO SOME OF THE MINIMAL

4376
02:45:21,137 --> 02:45:24,107
RISK INFORMATION, YOU KNOW, USE

4377
02:45:24,107 --> 02:45:26,509
OF THESE MULTIPLE SENSORS,

4378
02:45:26,509 --> 02:45:27,377
PARTICULARLY PASSIVE DATA

4379
02:45:27,377 --> 02:45:28,578
COLLECTION, AND, TWO, I'M

4380
02:45:28,578 --> 02:45:31,214
CURIOUS IF YOU HAVE ANY ADVICE

4381
02:45:31,214 --> 02:45:32,415
ON BEST PRACTICES FOR

4382
02:45:32,415 --> 02:45:33,983
SYSTEMATICALLY HOW WOULD YOU

4383
02:45:33,983 --> 02:45:35,185
TELL OTHERS DOING THIS RESEARCH

4384
02:45:35,185 --> 02:45:36,319
BASED ON YOUR LEARNINGS HOW

4385
02:45:36,319 --> 02:45:40,456
YOU'VE BEEN ABLE TO ASSESS

4386
02:45:40,456 --> 02:45:42,192
VALUES OF YOUR PARTICIPANTS IN

4387
02:45:42,192 --> 02:45:43,626
THE SENSE WHETHER THEY BELIEVE

4388
02:45:43,626 --> 02:45:45,895
SOMETHING IS DIFFERENT ABOUT THE

4389
02:45:45,895 --> 02:45:48,765
COMBINATION OF BRAIN DATA AND,

4390
02:45:48,765 --> 02:45:54,437
YOU KNOW, SOMETHING THAT'S

4391
02:45:54,437 --> 02:45:55,405
AUTOMATED IN A.I.?

4392
02:45:55,405 --> 02:45:57,273
>> I CAN ANSWER THE FIRST
PART.

4393
02:45:57,273 --> 02:46:00,143
SECOND IS IT'S A BIGGER LONGER

4394
02:46:00,143 --> 02:46:00,677
DISCUSSION I GUESS.

4395
02:46:00,677 --> 02:46:02,645
IN TERMS OF WHAT WE CURRENTLY

4396
02:46:02,645 --> 02:46:06,382
PUT IN MOST CONSENT FORMS IT'S

4397
02:46:06,382 --> 02:46:08,218
PRETTY VANILLA IN THE SENSE OF,

4398
02:46:08,218 --> 02:46:11,855
YOU KNOW, THE RISK IS BREACH OF

4399
02:46:11,855 --> 02:46:14,791
CONFIDENTIALITY IS THE MAIN RISK

4400
02:46:14,791 --> 02:46:15,992
THAT PEOPLE MIGHT EXPERIENCE

4401
02:46:15,992 --> 02:46:18,294
WITH HAVING THESE DATA BE

4402
02:46:18,294 --> 02:46:18,561
COLLECTED.

4403
02:46:18,561 --> 02:46:22,432
AS PART OF MOST OF OUR ONGOING

4404
02:46:22,432 --> 02:46:24,634
STUDIES WE DON'T RETURN

4405
02:46:24,634 --> 02:46:26,169
INDIVIDUALIZED RESEARCH RESULTS

4406
02:46:26,169 --> 02:46:27,470
TO PARTICIPANTS.

4407
02:46:27,470 --> 02:46:28,671
WE'RE STARTING TO EXPLORE HOW TO

4408
02:46:28,671 --> 02:46:31,741
DO THAT, BUT IN MOST OF THOSE

4409
02:46:31,741 --> 02:46:33,610
SITUATIONS IT'S MEDIATED BY

4410
02:46:33,610 --> 02:46:34,577
CLINICIAN, SO SOMEONE EITHER

4411
02:46:34,577 --> 02:46:36,679
THEIR DOCTOR OR ONE OF THE

4412
02:46:36,679 --> 02:46:41,384
RESEARCH STAFF WOULD GO OVER THE

4413
02:46:41,384 --> 02:46:47,857
RESULTS WITH THEM TO MITIGATE

4414
02:46:47,857 --> 02:46:50,827
POTENTIAL UNINTENDED QUESTIONS.

4415
02:46:50,827 --> 02:46:52,762
YOU MAY FIND ASSESSMENTS

4416
02:46:52,762 --> 02:46:54,430
EMOTIONALLY DISTRESSING WHEN WE

4417
02:46:54,430 --> 02:46:57,467
ASK ABOUT CERTAIN SYMPTOMS AND

4418
02:46:57,467 --> 02:46:58,468
POTENTIAL BREACH OF

4419
02:46:58,468 --> 02:47:00,003
CONFIDENTIALITY SO I WOULDN'T

4420
02:47:00,003 --> 02:47:07,110
SAY WE'RE SUPER EXHAUSTIVE IN

4421
02:47:07,110 --> 02:47:08,311
HOW WE FRAME THOSE RISKS.

4422
02:47:08,311 --> 02:47:11,281
>> I THINK IT'S
INTERESTING YOU

4423
02:47:11,281 --> 02:47:12,015
MENTIONED EXAMPLE OF THE

4424
02:47:12,015 --> 02:47:13,449
LOCATION PIECE, THAT YOU TALKED

4425
02:47:13,449 --> 02:47:14,884
ABOUT DOING THIS.

4426
02:47:14,884 --> 02:47:16,519
WAS THAT PROMPTED BY YOUR TEAM

4427
02:47:16,519 --> 02:47:18,922
SIGHING, HEY, WE OUGHT TO START

4428
02:47:18,922 --> 02:47:20,356
DOING THIS WITH PARTICIPANTS OR

4429
02:47:20,356 --> 02:47:22,759
WAS IT ONE PARTICULAR

4430
02:47:22,759 --> 02:47:24,394
PARTICIPANT, I DON'T LIKE THIS

4431
02:47:24,394 --> 02:47:26,362
GPS PIECE AND YOU NEGOTIATING

4432
02:47:26,362 --> 02:47:29,332
SOMETHING WITH THEM?

4433
02:47:29,332 --> 02:47:30,867
>> WELL, PART OF IT WAS

4434
02:47:30,867 --> 02:47:34,637
MOTIVATED BY THE FACT THAT NIH

4435
02:47:34,637 --> 02:47:36,572
ACTUALLY REQUIRES, PART OF THE

4436
02:47:36,572 --> 02:47:42,478
GLOBAL STUDY I MENTIONED THE NIH

4437
02:47:42,478 --> 02:47:46,649
IS INCREASING IF YOU'RE FUNDED

4438
02:47:46,649 --> 02:47:48,618
BY NIMH DOLLARS, IT'S GOVERNMENT

4439
02:47:48,618 --> 02:47:50,820
PROPERTY IN THE NIH ARCHIVES, WE

4440
02:47:50,820 --> 02:47:52,822
WORD THAT WOULD HAVE A NEGATIVE

4441
02:47:52,822 --> 02:47:54,090
EFFECT ON PARTICIPANTS BEING

4442
02:47:54,090 --> 02:47:56,492
WILLING TO PARTICIPATE IN THESE

4443
02:47:56,492 --> 02:47:58,795
TRIALS, WHEREAS WE SAW IF THE

4444
02:47:58,795 --> 02:48:00,763
DATA COULD POTENTIALLY BE REALLY

4445
02:48:00,763 --> 02:48:03,066
USEFUL BUT REQUIRING THAT THE

4446
02:48:03,066 --> 02:48:06,369
ACTUAL RAW GEOSPATIAL DATA GO

4447
02:48:06,369 --> 02:48:07,904
INTO A PERPETUAL DATABASE STRUCK

4448
02:48:07,904 --> 02:48:10,206
US AS MAYBE A LITTLE BIT OF

4449
02:48:10,206 --> 02:48:12,475
GOVERNMENT OVERREACH, I DON'T

4450
02:48:12,475 --> 02:48:13,910
KNOW, SO WE DECIDED TO PUT THAT

4451
02:48:13,910 --> 02:48:16,446
BACK IN THE HANDS OF

4452
02:48:16,446 --> 02:48:18,314
PARTICIPANTS TO SAY THAT'S NOT

4453
02:48:18,314 --> 02:48:19,215
SOMETHING I'M COMFORTABLE WITH,

4454
02:48:19,215 --> 02:48:25,521
I'LL JUST LEAVE THAT OUT, AS A

4455
02:48:25,521 --> 02:48:28,057
WAY OF MITIGATING THAT CONCERN.

4456
02:48:28,057 --> 02:48:29,726
WE WERE PLEASANTLY SURPRISED

4457
02:48:29,726 --> 02:48:30,727
PARTICIPANTS DIDN'T SEEM THAT

4458
02:48:30,727 --> 02:48:32,929
CONCERNED ABOUT SHARING IT IN

4459
02:48:32,929 --> 02:48:33,930
GENERAL.

4460
02:48:33,930 --> 02:48:37,200
THAT SAID WE THINK THAT GIVING

4461
02:48:37,200 --> 02:48:38,267
PEOPLE THE OPTIONALITY IS THE

4462
02:48:38,267 --> 02:48:41,037
RIGHT WAY TO HANDLE THIS.

4463
02:48:41,037 --> 02:48:42,105
THE OTHER WAY WE ENVISION

4464
02:48:42,105 --> 02:48:43,873
HANDLING THIS IN THE FUTURE IS

4465
02:48:43,873 --> 02:48:49,746
THAT, YOU KNOW, APPLE AND GOOGLE

4466
02:48:49,746 --> 02:48:50,880
AND OTHER OPERATING SYSTEM

4467
02:48:50,880 --> 02:48:53,282
DEVICE MANUFACTURERS ARE

4468
02:48:53,282 --> 02:48:54,717
STARTING TO EMBED SOFTWARE TOOLS

4469
02:48:54,717 --> 02:48:58,755
INTO THE OPERATING SYSTEMS THAT

4470
02:48:58,755 --> 02:48:59,756
DO SOME INITIAL FEATURE

4471
02:48:59,756 --> 02:49:01,724
EXTRACTION SO THE RESEARCHERS

4472
02:49:01,724 --> 02:49:02,692
THEMSELVES DON'T EVER GET ACCESS

4473
02:49:02,692 --> 02:49:04,460
TO THE RAW DATA.

4474
02:49:04,460 --> 02:49:09,465
THEY ONLY GET ACCESS TO FEATURES

4475
02:49:09,465 --> 02:49:10,666
WHICH WOULD DE-IDENTIFY FOR

4476
02:49:10,666 --> 02:49:11,000
PARTICIPANTS.

4477
02:49:11,000 --> 02:49:12,869
THAT COMES WITH PROS AND CONS.

4478
02:49:12,869 --> 02:49:15,038
YOU DON'T NECESSARILY KNOW HOW

4479
02:49:15,038 --> 02:49:16,039
TO REGENERATE THOSE FEATURES,

4480
02:49:16,039 --> 02:49:17,673
YOU DO IT THAT WAY.

4481
02:49:17,673 --> 02:49:19,876
BUT WE'RE TRYING TO FIND WAYS SO

4482
02:49:19,876 --> 02:49:24,580
MOST CALCULATIONS CAN BE DONE ON

4483
02:49:24,580 --> 02:49:25,581
DEVICE, AT THE PARTICIPANT'S

4484
02:49:25,581 --> 02:49:28,851
PHONE SO WE CAN CONTINUE TO

4485
02:49:28,851 --> 02:49:31,921
ACQUIRE BEHAVIORAL DATA WITHOUT

4486
02:49:31,921 --> 02:49:32,922
INCREASING RISKS FOR

4487
02:49:32,922 --> 02:49:33,256
PARTICIPANTS.

4488
02:49:33,256 --> 02:49:33,556
>> GREAT.

4489
02:49:33,556 --> 02:49:35,858
THANK YOU SO MUCH FOR ALL OF THE

4490
02:49:35,858 --> 02:49:36,192
SPEAKERS.

4491
02:49:36,192 --> 02:49:38,361
I WISH WE COULD CONTINUE THIS

4492
02:49:38,361 --> 02:49:40,329
CONVERSATION BUT WE HAVE OTHER

4493
02:49:40,329 --> 02:49:43,066
THINGS ON THE AGENDA AND PEOPLE

4494
02:49:43,066 --> 02:49:44,367
NEED A BREAK AND PROBABLY SOME

4495
02:49:44,367 --> 02:49:47,236
PEOPLE HAVE TO LEAVE.

4496
02:49:47,236 --> 02:49:53,342
SO THOSE WHO CAN STAY, INCLUDING

4497
02:49:53,342 --> 02:49:54,811
SPEAKERS WE'LL HAVE MORE TIME

4498
02:49:54,811 --> 02:49:55,878
LATER TO DISCUSS WHAT THE

4499
02:49:55,878 --> 02:49:57,080
WORKING GROUP MIGHT BE ABLE TO

4500
02:49:57,080 --> 02:49:57,246
DO.

4501
02:49:57,246 --> 02:49:58,948
IF YOU CAN'T STAY AND HAVE

4502
02:49:58,948 --> 02:50:01,918
IDEAS, WE'D LOVE TO HEAR THEM.

4503
02:50:01,918 --> 02:50:04,220
THANK YOU SO MUCH FOR BEING HERE

4504
02:50:04,220 --> 02:50:05,521
AND EVERYBODY GETS A 10-MINUTE

4505
02:50:05,521 --> 02:50:07,857
BREAK BEFORE YOU COME BACK AND

4506
02:50:07,857 --> 02:50:10,000
DEAL WITH CASE NUMBER TWO.

4507
02:50:13,247 --> 02:50:15,216
>> I'D LIKE TO WELCOME
EVERYONE

4508
02:50:15,216 --> 02:50:17,518
BACK TO TRAIN LARGE MODELS ON

4509
02:50:17,518 --> 02:50:19,487
BRAIN DATA.

4510
02:50:19,487 --> 02:50:22,323
I'M GRACE HWANG, PROGRAM

4511
02:50:22,323 --> 02:50:24,425
DIRECTOR FOR NINDS, THE BRAIN

4512
02:50:24,425 --> 02:50:28,262
INITIATIVE, ALONG WITH NITA

4513
02:50:28,262 --> 02:50:29,797
FARAHANY WE'LL CO-MODERATE THIS

4514
02:50:29,797 --> 02:50:36,337
SESSION. 

4515
02:50:36,337 --> 02:50:39,940
LET'S INTRODUCE DR. KARIM JERBI,

4516
02:50:39,940 --> 02:50:45,980
PROFESSOR OF PSYCHOLOGY FROM

4517
02:50:45,980 --> 02:50:52,987
UNIVERSITY OF MONTREAL, CANADA

4518
02:50:52,987 --> 02:50:55,055
RESEARCH CHAIR, HEADS UNIFYING

4519
02:50:55,055 --> 02:50:57,591
NEURAL SCIENCE IN QUEBEC, ALSO

4520
02:50:57,591 --> 02:50:58,993
ASSOCIATE PROFESSOR AT MONTREAL

4521
02:50:58,993 --> 02:51:00,861
BASED A.I. RESEARCH INSTITUTE,

4522
02:51:00,861 --> 02:51:03,430
RESEARCH LIES AT THE CROSSROADS

4523
02:51:03,430 --> 02:51:09,503
BETWEEN COGNITIVE, COMPUTATIONAL

4524
02:51:09,503 --> 02:51:10,371
AND CLINICAL NEUROSCIENCE.

4525
02:51:10,371 --> 02:51:16,277
HE WILL WALK US THROUGH HIS

4526
02:51:16,277 --> 02:51:16,944
PAPER ON NEURO-GPT.

4527
02:51:16,944 --> 02:51:18,145
>> THANK YOU FOR THE NICE

4528
02:51:18,145 --> 02:51:20,214
INTRODUCTION, GRACE.

4529
02:51:20,214 --> 02:51:24,285
I'M GOING TO SHARE MY SCREEN. 

4530
02:51:24,285 --> 02:51:32,126
I HOPE THIS WORKS OUT.

4531
02:51:32,126 --> 02:51:36,630
LET'S SEE. 

4532
02:51:36,630 --> 02:51:36,830
OKAY.

4533
02:51:36,830 --> 02:51:39,200
WELL, THANK YOU SO MUCH FOR

4534
02:51:39,200 --> 02:51:40,367
HAVING ME.

4535
02:51:40,367 --> 02:51:42,136
I'D LIKE TO START BY LAND

4536
02:51:42,136 --> 02:51:42,736
ACKNOWLEDGMENT, ACKNOWLEDGING

4537
02:51:42,736 --> 02:51:47,374
THE LAND ON WHICH WE GATHER IS

4538
02:51:47,374 --> 02:51:50,544
THE TRADITIONAL AN UNCEDED

4539
02:51:50,544 --> 02:51:54,381
TERRITORY OF THE KANIEN'

4540
02:51:54,381 --> 02:51:57,551
KEHA:KA, THE MOHAWK.

4541
02:51:57,551 --> 02:51:59,887
I'M ALSO THANKING EVERYBODY THAT

4542
02:51:59,887 --> 02:52:01,855
MADE THIS POSSIBLE, LOOKING

4543
02:52:01,855 --> 02:52:05,159
FORWARD TO THE DISCUSSION WITH

4544
02:52:05,159 --> 02:52:07,695
DEBRA MATTHEWS, THIS DISCUSSION

4545
02:52:07,695 --> 02:52:13,767
CO-MODERATED BY NITA FARAHANY

4546
02:52:13,767 --> 02:52:19,540
AND GRACE HWANG AND THANKS TO

4547
02:52:19,540 --> 02:52:21,709
NINA HSU, THANK YOU FOR THIS

4548
02:52:21,709 --> 02:52:22,476
OPPORTUNITY.

4549
02:52:22,476 --> 02:52:27,314
I WOULD LIKE TO START BY GIVING

4550
02:52:27,314 --> 02:52:29,817
CONTEXT, SO AS GRACE MENTIONED

4551
02:52:29,817 --> 02:52:33,087
EARLIER THE CURRENT DIRECTOR OF

4552
02:52:33,087 --> 02:52:41,061
THE UNIQUE NEURO-A.I. CENTRE IN

4553
02:52:41,061 --> 02:52:41,829
MONTREAL, BRIDGING ARTIFICIAL

4554
02:52:41,829 --> 02:52:43,497
INTELLIGENCE, IN LINE WITH

4555
02:52:43,497 --> 02:52:46,333
TOPICS WE'VE HEARD AND DISCUSSED

4556
02:52:46,333 --> 02:52:49,069
TODAY, THE IDEA IS BRINGING

4557
02:52:49,069 --> 02:52:51,472
THESE TWO FIELDS TOGETHER,

4558
02:52:51,472 --> 02:52:55,209
HOPEFULLY FOR THE BEST OF

4559
02:52:55,209 --> 02:52:56,310
HUMANITY, PITFALLS OF THINGS WE

4560
02:52:56,310 --> 02:52:57,478
NEED TO THINK ABOUT.

4561
02:52:57,478 --> 02:52:59,113
IDEA TO USE KNOWLEDGE FROM

4562
02:52:59,113 --> 02:53:01,115
NEUROSCIENCE TO ENHANCE A.I.,

4563
02:53:01,115 --> 02:53:03,984
BUT ALSO TO USE A.I. TO IMPROVE

4564
02:53:03,984 --> 02:53:05,386
OUR UNDERSTANDING OF

4565
02:53:05,386 --> 02:53:06,987
NEUROBIOLOGY OF OUR SYSTEMS AND

4566
02:53:06,987 --> 02:53:09,023
MAYBE HOPEFULLY AT THE END OF

4567
02:53:09,023 --> 02:53:12,626
THE DAY IMPROVE CLINICAL CARE.

4568
02:53:12,626 --> 02:53:15,229
AND WITHIN THE CENTER WE CREATED

4569
02:53:15,229 --> 02:53:18,098
IN QUEBEC, THE IDEA IS TO MOVE

4570
02:53:18,098 --> 02:53:18,932
BEYOND NEUROSCIENCE ON ONE HAND

4571
02:53:18,932 --> 02:53:21,935
AND A.I. ON THE OTHER TOWARDS A

4572
02:53:21,935 --> 02:53:24,538
FORM OF INTEGRATION WHERE BOTH

4573
02:53:24,538 --> 02:53:27,841
ARE JOINTLY EXPLORED FOR THEIR

4574
02:53:27,841 --> 02:53:29,143
MUTUAL BENEFITS THROUGH THE

4575
02:53:29,143 --> 02:53:35,382
FIELD WE REFER TO AS BEING

4576
02:53:35,382 --> 02:53:37,117
NEURO-A.I.

4577
02:53:37,117 --> 02:53:41,855
WE HAVE 70 RESEARCHERS, EIGHT

4578
02:53:41,855 --> 02:53:42,923
ACADEMIC INSTITUTIONS ACROSS

4579
02:53:42,923 --> 02:53:43,957
QUEBEC, 100 RESEARCHERS IN

4580
02:53:43,957 --> 02:53:45,359
NEUROSCIENCE OR A.I. OR

4581
02:53:45,359 --> 02:53:48,862
INTERSECTION BETWEEN THE TWO

4582
02:53:48,862 --> 02:53:50,097
FIELDS.

4583
02:53:50,097 --> 02:53:51,732
THIS STARTED IN 2019, THIS IS

4584
02:53:51,732 --> 02:53:54,335
ALSO PART OF A LARGER MOVEMENT

4585
02:53:54,335 --> 02:53:58,739
IN CANADA WHERE THE CANADIAN

4586
02:53:58,739 --> 02:54:00,574
BRAIN RESEARCH STRATEGY IS

4587
02:54:00,574 --> 02:54:02,042
BUILDING THE SIX TRANSFORMATIVE

4588
02:54:02,042 --> 02:54:06,580
INITIATIVES AND NEUROA.I. IS ONE

4589
02:54:06,580 --> 02:54:10,417
OF THE INITIATIVES DEEMS TO BE

4590
02:54:10,417 --> 02:54:12,619
TRANSFORMATIVE BECAUSE A STRONG

4591
02:54:12,619 --> 02:54:13,921
IMPACT ON CLINICAL APPLICATION

4592
02:54:13,921 --> 02:54:15,889
AND IMPROVING QUALITY OF LIFE OF

4593
02:54:15,889 --> 02:54:16,457
EVERYBODY.

4594
02:54:16,457 --> 02:54:19,727
MOST PEOPLE HERE ARE AWARE OF

4595
02:54:19,727 --> 02:54:21,362
THE INTERFACE BETWEEN BIOLOGY

4596
02:54:21,362 --> 02:54:23,430
AND MACHINE WHEN YOU'RE TALKING

4597
02:54:23,430 --> 02:54:25,265
ABOUT INTELLIGENCE FROM A

4598
02:54:25,265 --> 02:54:26,133
BIOLOGICAL PERSPECTIVE OR

4599
02:54:26,133 --> 02:54:27,368
MACHINE PERSPECTIVE.

4600
02:54:27,368 --> 02:54:30,337
AS YOU KNOW, THE INITIAL

4601
02:54:30,337 --> 02:54:31,004
INSPIRATION FOR ACTUALLY

4602
02:54:31,004 --> 02:54:32,973
GENERATING WHAT WE REFER TO

4603
02:54:32,973 --> 02:54:34,708
NOWADAYS AS ARTIFICIAL

4604
02:54:34,708 --> 02:54:37,244
INTELLIGENCE COMES FROM THE

4605
02:54:37,244 --> 02:54:38,979
DESIRE TO BUILD SOMETHING THAT

4606
02:54:38,979 --> 02:54:42,716
IS BRAIN INSPIRED A.I., AND IN

4607
02:54:42,716 --> 02:54:45,552
MANY CASES TODAY WE NEED LESS OF

4608
02:54:45,552 --> 02:54:46,553
THIS, SOME PEOPLE BELIEVE A.I.

4609
02:54:46,553 --> 02:54:48,422
HAS BECOME A FIELD OF ITS OWN

4610
02:54:48,422 --> 02:54:50,290
AND NEUROSCIENCE OF BIOLOGY THAT

4611
02:54:50,290 --> 02:54:51,925
INSPIRED THE BEGINNINGS IS MAYBE

4612
02:54:51,925 --> 02:54:53,360
NO LONGER NECESSARY FOR SOME

4613
02:54:53,360 --> 02:54:55,763
TASKS THAT WE USE A.I. FOR.

4614
02:54:55,763 --> 02:54:57,598
MANY OF US ALSO BELIEVE THERE

4615
02:54:57,598 --> 02:54:59,800
ARE MANY APPLICATIONS FOR A.I.

4616
02:54:59,800 --> 02:55:02,536
THAT WOULD BENEFIT FROM BETTER

4617
02:55:02,536 --> 02:55:05,172
INTEGRATION OF BIOLOGY AND

4618
02:55:05,172 --> 02:55:06,707
NEUROSCIENCE KNOWLEDGE SO MORE

4619
02:55:06,707 --> 02:55:08,342
INSPIRATION FROM NEUROSCIENCE

4620
02:55:08,342 --> 02:55:09,009
AND COGNITIVE NEUROSCIENCE IN

4621
02:55:09,009 --> 02:55:10,110
PARTICULAR WOULD BE USEFUL.

4622
02:55:10,110 --> 02:55:12,513
THE OTHER DIRECTION IS USING

4623
02:55:12,513 --> 02:55:15,783
A.I. TO INFORM WHAT WE DO IN

4624
02:55:15,783 --> 02:55:17,351
NEUROSCIENCE BRAIN RESEARCH,

4625
02:55:17,351 --> 02:55:18,085
ALSO CLINICAL APPLICATIONS, AND

4626
02:55:18,085 --> 02:55:22,256
THIS IS WHAT WE REFER TO AS

4627
02:55:22,256 --> 02:55:25,192
A.I.-POWERED BRAIN RESEARCH.

4628
02:55:25,192 --> 02:55:26,727
FOR THIS DIRECTION MACHINE

4629
02:55:26,727 --> 02:55:27,594
LEARNING HAS BEEN INCREASINGLY

4630
02:55:27,594 --> 02:55:29,163
USED AS A TOOL TO ADVANCE OUR

4631
02:55:29,163 --> 02:55:30,931
UNDERSTANDING OF BRAIN FUNCTION

4632
02:55:30,931 --> 02:55:32,299
AND DYSFUNCTION, AND THERE ARE

4633
02:55:32,299 --> 02:55:34,301
MANY WAYS THIS CAN HAPPEN.

4634
02:55:34,301 --> 02:55:37,771
AND RELATED ALSO TO TODAY'S

4635
02:55:37,771 --> 02:55:39,873
TOPIC, TWO WAYS THAT I'D LIKE TO

4636
02:55:39,873 --> 02:55:42,910
THINK ABOUT THIS IS SEEING THAT

4637
02:55:42,910 --> 02:55:46,213
A.I. CAN BE USED AS A TOOL TO

4638
02:55:46,213 --> 02:55:49,583
REVERSE ENGINEER THE BRAIN,

4639
02:55:49,583 --> 02:55:50,818
BUILDING MODELS, OF THE BRAIN

4640
02:55:50,818 --> 02:55:51,685
AND BRAIN FUNCTION.

4641
02:55:51,685 --> 02:55:53,420
IT'S A WAY FOR US TO UNDERSTAND

4642
02:55:53,420 --> 02:55:54,621
HOW THE BRAIN FUNCTIONS OR

4643
02:55:54,621 --> 02:55:57,257
DYSFUNCTIONS IN THE CASE OF

4644
02:55:57,257 --> 02:55:58,025
CLINICAL APPLICATIONS BUT ALSO

4645
02:55:58,025 --> 02:55:59,560
AS A DATA MINING TOOL.

4646
02:55:59,560 --> 02:56:01,862
AND DATA MINING TOOL IN THIS

4647
02:56:01,862 --> 02:56:07,668
CASE IS BASICALLY USING PATTERN

4648
02:56:07,668 --> 02:56:10,404
RECOGNITION OR WAYS OF EXPLORING

4649
02:56:10,404 --> 02:56:13,774
THE DATA THAT ARE VERY MUCH DATA

4650
02:56:13,774 --> 02:56:15,742
DRIVEN THAT COULD COMBINE ALSO

4651
02:56:15,742 --> 02:56:18,812
HYPOTHESIS DRIVEN BUT IN MANY

4652
02:56:18,812 --> 02:56:19,713
CASES STRONGLY DATA DRIVEN, VERY

4653
02:56:19,713 --> 02:56:23,650
MUCH RELATED TO OUR TOPIC TODAY

4654
02:56:23,650 --> 02:56:26,487
EXPLORING LARGE MODELS,

4655
02:56:26,487 --> 02:56:27,788
FOUNDATION MODELS FOR

4656
02:56:27,788 --> 02:56:28,121
NEUROSCIENCE.

4657
02:56:28,121 --> 02:56:29,456
SO THAT WAS MY QUICK

4658
02:56:29,456 --> 02:56:31,425
INTRODUCTION TO THE CONTEXT OF

4659
02:56:31,425 --> 02:56:32,526
THIS TYPE OF RESEARCH.

4660
02:56:32,526 --> 02:56:41,068
NOW I'D LIKE TO MOVE ON TO MORE

4661
02:56:41,068 --> 02:56:42,603
SPECIFICALLY RICHARD LEAHY AND I

4662
02:56:42,603 --> 02:56:47,841
WERE INVITED TO GIVE AN OVERVIEW

4663
02:56:47,841 --> 02:56:50,577
OF THE NEURO-GPT PAPER, A

4664
02:56:50,577 --> 02:56:52,980
PRE-PRINT NOT LONG AGO, WORK LED

4665
02:56:52,980 --> 02:57:02,189
BY A PhD student at Southern

4666
02:57:02,189 --> 02:57:03,457
California in electrical

4667
02:57:03,457 --> 02:57:05,459
engineering and a project in my

4668
02:57:05,459 --> 02:57:08,228
sabbatical a year ago as USC

4669
02:57:08,228 --> 02:57:09,730
collaborating with this team

4670
02:57:09,730 --> 02:57:12,900
that is also headedded by

4671
02:57:12,900 --> 02:57:13,901
Richard Leahy at USC.

4672
02:57:13,901 --> 02:57:15,969
YOU CAN SEE THE LIST OF OTHER

4673
02:57:15,969 --> 02:57:16,870
COLLABORATORS HERE.

4674
02:57:16,870 --> 02:57:19,706
I WANTED TO ACKNOWLEDGE THEM.

4675
02:57:19,706 --> 02:57:23,310
NOW BY WAY OF INTRODUCTION

4676
02:57:23,310 --> 02:57:26,480
YOU'VE SEEN THE GREAT OVERVIEW

4677
02:57:26,480 --> 02:57:28,782
OF THE FIELD OF FOUNDATION

4678
02:57:28,782 --> 02:57:30,417
MODELS FOR NEUROSCIENCE BY

4679
02:57:30,417 --> 02:57:31,952
PATRICK EARLIER TODAY, SO I WILL

4680
02:57:31,952 --> 02:57:33,487
NOT DELVE INTO MANY DETAILS BUT

4681
02:57:33,487 --> 02:57:37,624
WOULD LIKE TO GIVE A REFRESHER

4682
02:57:37,624 --> 02:57:40,193
OVERVIEW FOR THOSE NOT FAMILIAR

4683
02:57:40,193 --> 02:57:42,029
WITH SELF-SUPERVISED LEARNING IN

4684
02:57:42,029 --> 02:57:43,263
THE CONTEXT OF NEUROSCIENTIFIC

4685
02:57:43,263 --> 02:57:45,732
APPLICATIONS, THE IDEA THAT WE

4686
02:57:45,732 --> 02:57:48,902
MOVE FROM PRE-TRAINING A MODEL

4687
02:57:48,902 --> 02:57:50,237
TOWARDS APPLICATIONS WHICH WE

4688
02:57:50,237 --> 02:57:52,306
REFER TO AS DOWNSTREAM TASKS.

4689
02:57:52,306 --> 02:57:54,174
TO GIVE AN EXAMPLE

4690
02:57:54,174 --> 02:57:55,576
SELF-SUPERVISED PRE-TRAINING CAN

4691
02:57:55,576 --> 02:58:01,848
BE USE OF LARGE MEDICAL IMAGE

4692
02:58:01,848 --> 02:58:03,150
DATASETS, WITHOUT ANNOTATIONS

4693
02:58:03,150 --> 02:58:06,320
HENCE SELF-SUPERVISED, NOT USING

4694
02:58:06,320 --> 02:58:06,553
LABELS.

4695
02:58:06,553 --> 02:58:07,688
SELF-SUPERVISED LEARNING WILL BE

4696
02:58:07,688 --> 02:58:10,290
ABLE TO REBUILD FOR EXAMPLE

4697
02:58:10,290 --> 02:58:13,594
RECONSTRUCT DATA BY FILLING IN,

4698
02:58:13,594 --> 02:58:15,128
FOR EXAMPLE, MISSING GAPS IN THE

4699
02:58:15,128 --> 02:58:15,329
DATA.

4700
02:58:15,329 --> 02:58:18,632
WE CAN REMOVE PARTS OF THE DATA,

4701
02:58:18,632 --> 02:58:21,001
DO MASKING, TRY TO RECONSTRUCT

4702
02:58:21,001 --> 02:58:21,368
THEM.

4703
02:58:21,368 --> 02:58:22,002
THERE'S NO CLASSIFICATION TASK

4704
02:58:22,002 --> 02:58:24,204
PER SE IN THIS CASE HERE.

4705
02:58:24,204 --> 02:58:25,505
IT'S A SELF-SUPERVISED COMPONENT

4706
02:58:25,505 --> 02:58:27,908
WHERE WE'RE BUILDING A MODEL

4707
02:58:27,908 --> 02:58:30,110
THAT IS LEARNING

4708
02:58:30,110 --> 02:58:30,477
REPRESENTATIONS.

4709
02:58:30,477 --> 02:58:34,014
NOW, THAT IS FOLLOWED BY THE

4710
02:58:34,014 --> 02:58:35,382
SUPERVISED FINE TUNING ON

4711
02:58:35,382 --> 02:58:36,183
DOWNSTREAM TASKS.

4712
02:58:36,183 --> 02:58:39,519
THIS IS WHERE YOU TAKE THE MODEL

4713
02:58:39,519 --> 02:58:41,488
THAT HAS BEEN PRE-TRAINED ON

4714
02:58:41,488 --> 02:58:44,324
VERY LARGE DATASETS, AND NOW YOU

4715
02:58:44,324 --> 02:58:47,628
CAN FINE TUNE IT ON A

4716
02:58:47,628 --> 02:58:49,029
CLASSIFICATION TASK ON SMALLER

4717
02:58:49,029 --> 02:58:51,098
DATASET, FOR EXAMPLE, SMALL

4718
02:58:51,098 --> 02:58:53,533
MEDICAL IMAGE DATASET, AND NOW

4719
02:58:53,533 --> 02:58:56,470
YOU USE EITHER THE WHOLE MODEL

4720
02:58:56,470 --> 02:58:58,138
OR PARTS OF THE LEARNING IN A

4721
02:58:58,138 --> 02:59:00,407
SUPERVISED LEARNING CONTEXT TO

4722
02:59:00,407 --> 02:59:01,441
MAKE CLASSIFICATIONS.

4723
02:59:01,441 --> 02:59:04,344
THE IDEA BEING HERE THAT IF

4724
02:59:04,344 --> 02:59:06,880
YOU'RE TRAINING THE MODEL AND

4725
02:59:06,880 --> 02:59:10,283
DOING A PRE-TRAINING, ALSO

4726
02:59:10,283 --> 02:59:11,918
REFERRED TO AS PRETEXT, ON A

4727
02:59:11,918 --> 02:59:13,854
VERY LARGE DATASET THAT YOU

4728
02:59:13,854 --> 02:59:14,755
MIGHT LEARN REPRESENTATIONS,

4729
02:59:14,755 --> 02:59:15,956
MIGHT LEARN SPECIFIC THINGS

4730
02:59:15,956 --> 02:59:17,491
ABOUT THIS TYPE OF DATA, THAT

4731
02:59:17,491 --> 02:59:19,893
YOU WOULD NOT BE ABLE TO EXTRACT

4732
02:59:19,893 --> 02:59:21,862
FROM A SMALLER SUBSET OF DATA.

4733
02:59:21,862 --> 02:59:24,297
AND SO YOU START OFF INSTEAD OF

4734
02:59:24,297 --> 02:59:33,040
STARTING FROM SCRATCH

4735
02:59:33,040 --> 02:59:34,775
BUILDELLING A MODEL, APPLY AS A

4736
02:59:34,775 --> 02:59:38,045
CLASSIFIER ON A SPECIFIC TASK,

4737
02:59:38,045 --> 02:59:39,513
DOWNSTREAM TASKS.

4738
02:59:39,513 --> 02:59:42,115
IN THE CONTEXT OF THIS WORK,

4739
02:59:42,115 --> 02:59:44,284
OBVIOUSLY WE'RE GOING TO BE

4740
02:59:44,284 --> 02:59:45,152
TALKING ABOUT PRE-TRAINING

4741
02:59:45,152 --> 02:59:46,253
DATASETS AND THEN THE

4742
02:59:46,253 --> 02:59:47,988
PRE-TRAINING MODEL AND MOVE TO

4743
02:59:47,988 --> 02:59:49,856
THE DOWNSTREAM TASKS AS I JUST

4744
02:59:49,856 --> 02:59:50,524
EXPLAINED.

4745
02:59:50,524 --> 02:59:53,460
NOW HERE THE PRE-TRAINING DATA

4746
02:59:53,460 --> 02:59:55,562
AND THIS CONTEXT WAS TEMPLE

4747
02:59:55,562 --> 02:59:58,532
UNIVERSITY HOSPITAL EEG CORPUS,

4748
02:59:58,532 --> 03:00:00,467
THIS HAS OVER 25,000 RECORDINGS

4749
03:00:00,467 --> 03:00:04,971
FROM 25,000 SUBJECTS, EEG

4750
03:00:04,971 --> 03:00:06,139
RECORDINGS.

4751
03:00:06,139 --> 03:00:08,575
VERY DIVERSE RECORDINGS, OVER 40

4752
03:00:08,575 --> 03:00:11,445
CHANNEL CONFIGURATIONS OF EEG,

4753
03:00:11,445 --> 03:00:13,380
VARYING DURATIONS, AND DIFFERENT

4754
03:00:13,380 --> 03:00:14,381
PARTICIPANTS IN DIFFERENT

4755
03:00:14,381 --> 03:00:16,783
CLINICAL CONDITIONS.

4756
03:00:16,783 --> 03:00:17,451
DIFFERENT SAMPLING FREQUENCIES.

4757
03:00:17,451 --> 03:00:24,024
AND FOR THE SAKE OF THIS

4758
03:00:24,024 --> 03:00:28,161
PROJECT, WHEN WE AND PHILIP AND

4759
03:00:28,161 --> 03:00:36,470
WUJAY PRE-PROCESSED OVER 20,000

4760
03:00:36,470 --> 03:00:39,005
EEG RECORDINGS, 32 CHUNKS, EACH

4761
03:00:39,005 --> 03:00:41,408
CHUNK OF 2 SECONDS, ONE MINUTE

4762
03:00:41,408 --> 03:00:43,176
OF RECORDING PER BLOCK.

4763
03:00:43,176 --> 03:00:46,113
AND THEN WE USED THIS FOR THE

4764
03:00:46,113 --> 03:00:48,081
PRE-TRAINING MODEL.

4765
03:00:48,081 --> 03:00:50,383
AND THEN AFTERWARDS WE DO

4766
03:00:50,383 --> 03:00:51,818
DOWNSTREAM TASKS, ONCE YOU'VE

4767
03:00:51,818 --> 03:00:53,987
PRE-TRAINED THIS MODEL ON THE

4768
03:00:53,987 --> 03:01:00,994
LARGE CORPUS OF DATA EEG ACROSS

4769
03:01:00,994 --> 03:01:07,667
20,000 RECORDINGS, HOW WELL DOES

4770
03:01:07,667 --> 03:01:08,769
IT PERFORM AND COMPARE TO

4771
03:01:08,769 --> 03:01:10,170
SCRATCH ON THE TASK WE'RE TRYING

4772
03:01:10,170 --> 03:01:11,404
TO DO.

4773
03:01:11,404 --> 03:01:15,675
SO WHAT WAS DOWNSTREAM TASKS,

4774
03:01:15,675 --> 03:01:17,544
BCI, BRAIN-COMPUTER INTERFACE

4775
03:01:17,544 --> 03:01:18,945
COMPOSITION IV, PROVIDED BY

4776
03:01:18,945 --> 03:01:23,216
UNIVERSITY OF TECHNOLOGY IN

4777
03:01:23,216 --> 03:01:25,085
AUSTRIA, ONLY NINE SUBJECTS,

4778
03:01:25,085 --> 03:01:26,820
FOUR MOTOR IMAGERY TASKS,

4779
03:01:26,820 --> 03:01:28,155
IMAGINING LEFT HAND VERSUS RIGHT

4780
03:01:28,155 --> 03:01:34,294
HAND VERSUS FEET VERSUS TONGUE

4781
03:01:34,294 --> 03:01:35,061
MOVEMENTS.

4782
03:01:35,061 --> 03:01:36,496
SAMPLING FREQUENCY OF 250, 72

4783
03:01:36,496 --> 03:01:39,232
TRIALS PER TASK, TOTAL OF 288,

4784
03:01:39,232 --> 03:01:42,636
EMPHASIZING THIS IS A SMALL

4785
03:01:42,636 --> 03:01:44,838
DATASET, IT'S A CHALLENGE, ALSO

4786
03:01:44,838 --> 03:01:47,040
A BENCHMARK WITH DIFFERENT

4787
03:01:47,040 --> 03:01:47,808
MODELS FOR CLASSIFICATION USED

4788
03:01:47,808 --> 03:01:49,776
TO SEE HOW WELL THEY CAN

4789
03:01:49,776 --> 03:01:50,010
PERFORM.

4790
03:01:50,010 --> 03:01:52,612
SO THIS SERVES FOR US AS A GOOD

4791
03:01:52,612 --> 03:01:54,514
EXAMPLE, WELL, OKAY, IF I CAN

4792
03:01:54,514 --> 03:01:56,917
TRAIN A MODEL ON THOUSANDS OF

4793
03:01:56,917 --> 03:01:58,552
INDIVIDUALS, AND TRY TO LEARN

4794
03:01:58,552 --> 03:02:00,854
THE GIST OR REPRESENTATION OF

4795
03:02:00,854 --> 03:02:03,924
WHAT EEG DATA LOOKS LIKE, IF I

4796
03:02:03,924 --> 03:02:06,626
START NOW WITH THIS PRE-TRAINED

4797
03:02:06,626 --> 03:02:10,197
MODEL AND TAKE IT TO A TASK OF

4798
03:02:10,197 --> 03:02:13,133
TRYING TO GUESS HERE IN THIS

4799
03:02:13,133 --> 03:02:14,201
CASE MOTOR IMAGERY OF

4800
03:02:14,201 --> 03:02:16,503
PARTICIPANTS HOW WELL WOULD THAT

4801
03:02:16,503 --> 03:02:19,005
PERFORM COMPARED TO NOT AT ALL

4802
03:02:19,005 --> 03:02:20,974
HAVING A MODEL, STARTING

4803
03:02:20,974 --> 03:02:23,510
DIRECTLY BY THE DATA.

4804
03:02:23,510 --> 03:02:26,446
AND SO NOW VERY BRIEF OVERVIEW,

4805
03:02:26,446 --> 03:02:30,050
I WILL NOT GO INTO DETAIL OF THE

4806
03:02:30,050 --> 03:02:33,119
ARCHITECTURE BUT BASICALLY TWO

4807
03:02:33,119 --> 03:02:34,888
COMPONENTS, EEG ENCODER THAT'S

4808
03:02:34,888 --> 03:02:40,794
GOING TO TAKE IN THE EEG, RAW 

4809
03:02:40,794 --> 03:02:43,196
EEG DATA AND GENERATE

4810
03:02:43,196 --> 03:02:48,535
EMBEDDINGS, THESE ARE SERVE AS

4811
03:02:48,535 --> 03:02:53,673
LOWER DIMENSIONAL AND DENOISED

4812
03:02:53,673 --> 03:02:54,341
REPRESENTATION, EXTRACTING

4813
03:02:54,341 --> 03:02:56,109
FEATURES, SIMPLIFYING PREDICTION

4814
03:02:56,109 --> 03:02:59,045
OF MASS CHUNK FOR GPT MODEL

4815
03:02:59,045 --> 03:03:04,084
RUNNING AFTERWARDS, ENHANCES ITS

4816
03:03:04,084 --> 03:03:07,954
PERFORMANCE THINKING ABOUT THIS

4817
03:03:07,954 --> 03:03:10,457
AS DENOISING, INJECTING TEMPORAL

4818
03:03:10,457 --> 03:03:11,491
CORRELATIONS, SECOND COMPONENT

4819
03:03:11,491 --> 03:03:15,195
OF THE PIPELINE IS THE ACTUAL

4820
03:03:15,195 --> 03:03:18,331
GPT MODEL ITSELF, AND AS THE

4821
03:03:18,331 --> 03:03:20,000
NAME INDICATES EMPLOYS

4822
03:03:20,000 --> 03:03:20,967
TRANSFORMER ARCHITECTURE.

4823
03:03:20,967 --> 03:03:23,136
AND IT'S GOING TO BE EXACTLY

4824
03:03:23,136 --> 03:03:28,608
LIKE, YOU CAN THINK OF THIS IN

4825
03:03:28,608 --> 03:03:30,477
ChatGPT, PREDICTING OR GPT

4826
03:03:30,477 --> 03:03:33,213
MODEL FOR LANGUAGE PREDICTING

4827
03:03:33,213 --> 03:03:35,515
THE NEXT WORD, IN THIS CASE

4828
03:03:35,515 --> 03:03:37,284
PREDICTING THE NEXT TOKEN, IN

4829
03:03:37,284 --> 03:03:39,886
THIS CASE EEG DATA, CAN I USE

4830
03:03:39,886 --> 03:03:43,623
PREVIOUS EEG CHUNKS OF DATA TO

4831
03:03:43,623 --> 03:03:45,592
PREDICT THE NEXT ONE LIKE A

4832
03:03:45,592 --> 03:03:47,460
LANGUAGE MODEL WOULD PREDICT THE

4833
03:03:47,460 --> 03:03:48,862
NEXT WORDS.

4834
03:03:48,862 --> 03:03:50,931
THIS SOMEHOW GUARANTEES

4835
03:03:50,931 --> 03:03:58,038
PREDICTION OF EEG EMBEDDINGS,

4836
03:03:58,038 --> 03:04:01,207
ALSO CONSIDERING CAUSAL TEMPORAL

4837
03:04:01,207 --> 03:04:04,277
RELATIONSHIPS, DIFFERS FROM BERT

4838
03:04:04,277 --> 03:04:06,246
AS OPPOSED TO GPT FOR INSTANCE.

4839
03:04:06,246 --> 03:04:13,019
SO, IF WE TAKE A QUICK LOOK AT

4840
03:04:13,019 --> 03:04:14,688
THE RESULTS, WE COMPARED RESULTS

4841
03:04:14,688 --> 03:04:16,756
TO OTHER METHODS USED FOR

4842
03:04:16,756 --> 03:04:18,525
CLASSIFICATION, AGAIN THIS IS A

4843
03:04:18,525 --> 03:04:18,959
MOTOR IMAGERY TASK.

4844
03:04:18,959 --> 03:04:20,827
IN OTHER WORDS THIS MEANS THE

4845
03:04:20,827 --> 03:04:22,462
BASELINE IF YOU HAD CLASSIFIER

4846
03:04:22,462 --> 03:04:24,664
THAT WOULD BE RANDOMLY GUESSING

4847
03:04:24,664 --> 03:04:27,400
WHAT TYPE OF MOVEMENTS THE

4848
03:04:27,400 --> 03:04:29,035
INDIVIDUAL WAS THINKING OF, YOUR

4849
03:04:29,035 --> 03:04:32,172
CLASSIFICATION PERCENTAGE WOULD

4850
03:04:32,172 --> 03:04:36,810
BE AT 25%, 0.25 WOULD BE CHANCE

4851
03:04:36,810 --> 03:04:37,243
LEVEL.

4852
03:04:37,243 --> 03:04:39,412
YOU CAN SEE I PRESENT THIS HERE

4853
03:04:39,412 --> 03:04:42,916
AS A PLOT ON THE RIGHT-HAND SIDE

4854
03:04:42,916 --> 03:04:47,854
HERE, THE BEST RESULT WAS

4855
03:04:47,854 --> 03:04:49,823
OBTAINED WITH NEURO-GPT, I DID

4856
03:04:49,823 --> 03:04:52,993
NOT GO INTO DETAILS, THREE WAYS

4857
03:04:52,993 --> 03:04:54,961
OF FINE TUNING, THREE FINE

4858
03:04:54,961 --> 03:05:05,505
TUNING STRATEGIES, THREE AT THE

4859
03:05:08,508 --> 03:05:12,312
BEGINNING, LINEAR, ENCODER,

4860
03:05:12,312 --> 03:05:14,280
LINEAR PLUS GPT, 64%, AGAIN THE

4861
03:05:14,280 --> 03:05:16,349
CHANCE LEVEL IS 25% HERE.

4862
03:05:16,349 --> 03:05:19,219
NOW, THE NUMBER PER SE IS NOT

4863
03:05:19,219 --> 03:05:21,054
VERY HIGH, IT IS BETTER THAN

4864
03:05:21,054 --> 03:05:23,023
STATE OF THE ART IN THE FIELD AT

4865
03:05:23,023 --> 03:05:24,991
THE TIME THIS PRE-PRINT WAS PUT

4866
03:05:24,991 --> 03:05:25,392
OUT.

4867
03:05:25,392 --> 03:05:27,727
BUT I THINK WE'LL GET BACK TO

4868
03:05:27,727 --> 03:05:28,762
THE DISCUSSION AFTERWARDS

4869
03:05:28,762 --> 03:05:30,230
BECAUSE, AGAIN, THIS IS A PROOF

4870
03:05:30,230 --> 03:05:31,231
OF PRINCIPLE.

4871
03:05:31,231 --> 03:05:32,432
THIS IS NOT TO SAY THIS IS WHERE

4872
03:05:32,432 --> 03:05:35,869
WE WANT TO BE, THIS IS A PROOF

4873
03:05:35,869 --> 03:05:37,937
OF PRINCIPLE OF TRAINING AN

4874
03:05:37,937 --> 03:05:41,341
ALGORITHM OR MODEL HERE ON LARGE

4875
03:05:41,341 --> 03:05:42,976
DATASETS AND THEN USING THAT AS

4876
03:05:42,976 --> 03:05:44,611
A STARTING POINT TO IMPROVE WHAT

4877
03:05:44,611 --> 03:05:51,084
WE CAN DO ON DOWNSTREAM TASKS.

4878
03:05:51,084 --> 03:05:54,487
SO, IN TERMS OF SUMMARIZING THE

4879
03:05:54,487 --> 03:05:58,525
MAIN MESSAGE OF THE PRE-PRINT,

4880
03:05:58,525 --> 03:06:00,927
THE STUDY SHOWS OR SHOWS HOW

4881
03:06:00,927 --> 03:06:03,563
APPLYING FOUNDATION MODEL CAN

4882
03:06:03,563 --> 03:06:04,697
IMPROVE EEG CLASSIFICATION

4883
03:06:04,697 --> 03:06:05,632
PERFORMANCE COMPARED TO MODEL

4884
03:06:05,632 --> 03:06:08,301
THAT WAS TRAINED FROM SCRATCH

4885
03:06:08,301 --> 03:06:09,269
FOR INSTANCE.

4886
03:06:09,269 --> 03:06:11,104
WE ALSO SHOW EVIDENCE FOR

4887
03:06:11,104 --> 03:06:11,871
GENERALIZABILITY OF THE

4888
03:06:11,871 --> 03:06:14,074
FOUNDATION MODEL AND ITS ABILITY

4889
03:06:14,074 --> 03:06:15,909
TO ADDRESS CHALLENGES OF DATA

4890
03:06:15,909 --> 03:06:20,313
SCARCITY AND HETEROGENEITY IN

4891
03:06:20,313 --> 03:06:21,514
EEG, SO SOMETHING, AGAIN, YOU

4892
03:06:21,514 --> 03:06:22,615
MIGHT HAVE CONDITIONS OR TASKS

4893
03:06:22,615 --> 03:06:24,984
WHERE YOU DON'T HAVE ENOUGH DATA

4894
03:06:24,984 --> 03:06:27,854
FOR YOU TO TRAIN A MODEL

4895
03:06:27,854 --> 03:06:31,224
RELIABLY TO BE ABLE TO DO ANY

4896
03:06:31,224 --> 03:06:31,591
CLASSIFICATIONS.

4897
03:06:31,591 --> 03:06:35,495
YOU 

4898
03:06:35,495 --> 03:06:37,797
JUST CAN'T TRAIN A MODEL FROM

4899
03:06:37,797 --> 03:06:39,566
THE EXISTING DATA IN THAT TASK.

4900
03:06:39,566 --> 03:06:42,735
TRANSFER LEARNING OF THIS TYPE

4901
03:06:42,735 --> 03:06:44,070
HERE WOULD BE USEFUL.

4902
03:06:44,070 --> 03:06:48,408
OBVIOUSLY THE CODE IS AVAILABLE

4903
03:06:48,408 --> 03:06:49,843
FOR THIS WORK.

4904
03:06:49,843 --> 03:06:52,045
WE STARTED THINKING ABOUT

4905
03:06:52,045 --> 03:06:53,146
ETHICAL CONSIDERATIONS THAT

4906
03:06:53,146 --> 03:06:57,951
HOPEFULLY  WE'LL BE DISCUSSING

4907
03:06:57,951 --> 03:06:59,686
THESE AND OTHERS TODAY YOU MIGHT

4908
03:06:59,686 --> 03:07:01,421
RECOGNIZE TWO SLIDES FROM

4909
03:07:01,421 --> 03:07:02,589
PATRICK'S PRESENTATION, THEY FIT

4910
03:07:02,589 --> 03:07:03,957
PERFECTLY WELL.

4911
03:07:03,957 --> 03:07:04,958
THERE ARE ETHICAL CONSIDERATIONS

4912
03:07:04,958 --> 03:07:08,261
THAT HOLD TRUE FOR CLASSIC

4913
03:07:08,261 --> 03:07:09,562
MACHINE LEARNING IN NEUROSCIENCE

4914
03:07:09,562 --> 03:07:12,832
ANYWAY NOT SPECIFIC TO THIS,

4915
03:07:12,832 --> 03:07:19,506
THEY HOLD HERE TOO,

4916
03:07:19,506 --> 03:07:20,373
CONSIDERATIONS FOR FOUNDATION

4917
03:07:20,373 --> 03:07:21,074
MODELS, PARTICULARLY IMPORTANT

4918
03:07:21,074 --> 03:07:22,909
WHERE WE NEED TO THINK ABOUT THE

4919
03:07:22,909 --> 03:07:24,644
FACT THAT OPEN WEIGHTS IS NOT

4920
03:07:24,644 --> 03:07:27,847
THE SAME AS OPEN SOURCE AND IN

4921
03:07:27,847 --> 03:07:31,618
SOME OF THESE GPT OR GPT-LIKE

4922
03:07:31,618 --> 03:07:35,955
MODELS THAT WE MIGHT BE USING IN

4923
03:07:35,955 --> 03:07:38,458
THIS ENDEAVOR MANY, FOR EXAMPLE,

4924
03:07:38,458 --> 03:07:41,060
GPT 3 OR 4, THE WEIGHTS ARE NOT

4925
03:07:41,060 --> 03:07:44,130
AVAILABLE, THIS IS AN IMPORTANT

4926
03:07:44,130 --> 03:07:46,099
ISSUE TALKING ABOUT ETHICS IF

4927
03:07:46,099 --> 03:07:48,935
YOU WANT TRANSPARENCY AT ALL

4928
03:07:48,935 --> 03:07:51,471
LEVELS OF THE PIPELINE.

4929
03:07:51,471 --> 03:07:53,673
ANOTHER -- TWO OTHER POINTS

4930
03:07:53,673 --> 03:07:55,408
WORTH DISCUSSING, I WAS HAPPY TO

4931
03:07:55,408 --> 03:07:57,677
SEE THIS IN A DISCUSSION EARLIER

4932
03:07:57,677 --> 03:08:00,313
ON TODAY, THE FACT THAT INFORMED

4933
03:08:00,313 --> 03:08:02,715
CONSENT IS A TRICKY CONCEPT, I

4934
03:08:02,715 --> 03:08:03,616
THINK HERE, BECAUSE IT'S

4935
03:08:03,616 --> 03:08:05,018
DIFFICULT TO BE INFORMED ABOUT

4936
03:08:05,018 --> 03:08:07,320
WHAT'S GOING TO COME LATER

4937
03:08:07,320 --> 03:08:08,188
SPECIFICALLY IN THE CASE WHERE

4938
03:08:08,188 --> 03:08:13,459
FIELD OF A.I. IS MOVING SO FAST.

4939
03:08:13,459 --> 03:08:14,427
SO INDIVIDUAL THAT PROVIDES

4940
03:08:14,427 --> 03:08:17,297
CONSENT FOR THE USE OF A.I. IN

4941
03:08:17,297 --> 03:08:18,364
REALITY DON'T REALLY KNOW WHAT

4942
03:08:18,364 --> 03:08:23,102
THEY ARE PROVIDING COULD  -- COT

4943
03:08:23,102 --> 03:08:26,072
FOR, WE NEED GUIDELINES TO HELP

4944
03:08:26,072 --> 03:08:29,909
US IDENTIFY WHAT A CONSENT AT

4945
03:08:29,909 --> 03:08:32,312
THE TIME IMPLIES FOR WHEN

4946
03:08:32,312 --> 03:08:35,315
TECHNOLOGY CHANGES IN THE FUTURE

4947
03:08:35,315 --> 03:08:37,016
AND WHETHER THOSE -- THAT

4948
03:08:37,016 --> 03:08:39,118
CONSENT IS STILL VALID OR NOT.

4949
03:08:39,118 --> 03:08:42,522
I'M SURE THERE ARE LEGAL EXPERTS

4950
03:08:42,522 --> 03:08:43,590
AND EXPERTS IN NEUROETHICS THAT

4951
03:08:43,590 --> 03:08:45,225
HAVE AN OPINION ON THIS.

4952
03:08:45,225 --> 03:08:49,395
ALSO MY SECOND POINT HERE IS

4953
03:08:49,395 --> 03:08:51,798
DERIVING PRE-TRAINED MODELS WILL

4954
03:08:51,798 --> 03:08:53,433
PERPETUATE AND EVEN WORSE THAN

4955
03:08:53,433 --> 03:08:55,635
THAT ENHANCE BIASES THAT EXIST

4956
03:08:55,635 --> 03:08:57,604
IN THE TRAINING SET.

4957
03:08:57,604 --> 03:08:59,472
THIS IS AN IMPORTANT ISSUE HERE.

4958
03:08:59,472 --> 03:09:00,673
WE KNOW THIS.

4959
03:09:00,673 --> 03:09:02,642
WE'VE KNOWN THAT GARBAGE IN,

4960
03:09:02,642 --> 03:09:04,944
GARBAGE OUT IN OUR MODELS, BIAS

4961
03:09:04,944 --> 03:09:09,015
IN ALSO MEANS BIAS OUT.

4962
03:09:09,015 --> 03:09:11,517
AND WHEN WE ARE THINKING WHAT

4963
03:09:11,517 --> 03:09:15,021
HAPPENS WHEN WE DERIVE

4964
03:09:15,021 --> 03:09:20,493
FOUNDATION MODELS, WE NEED TO

4965
03:09:20,493 --> 03:09:22,128
REALIZE THAT, YEAH, THAT IF

4966
03:09:22,128 --> 03:09:24,330
WE'RE TALKING ABOUT PRE-TRAINING

4967
03:09:24,330 --> 03:09:26,065
MODELS, FOUNDATION MODELS FOR

4968
03:09:26,065 --> 03:09:27,166
NEUROSCIENCE, WE'LL BE LOOKING

4969
03:09:27,166 --> 03:09:30,670
FOR THE DATA OUT THERE THAT

4970
03:09:30,670 --> 03:09:31,104
EXISTS.

4971
03:09:31,104 --> 03:09:33,273
WHAT'S NEUROSCIENTIFIC DATA,

4972
03:09:33,273 --> 03:09:35,575
WHAT BRAIN DATA, EEG, MAG,

4973
03:09:35,575 --> 03:09:37,443
fMRI, WHAT OTHER DATASETS ARE

4974
03:09:37,443 --> 03:09:39,646
AVAILABLE, AND WE'LL REALIZE

4975
03:09:39,646 --> 03:09:41,180
QUICKLY THAT THERE'S STRONG BIAS

4976
03:09:41,180 --> 03:09:43,583
IN THE WAY THE DATA HAS BEEN

4977
03:09:43,583 --> 03:09:46,319
COLLECTED AND WHAT DATA IS

4978
03:09:46,319 --> 03:09:46,753
AVAILABLE.

4979
03:09:46,753 --> 03:09:51,691
THAT'S AN IMMENSE PROBLEM.

4980
03:09:51,691 --> 03:09:53,326
ALSO BEYOND THESE LIMITATIONS

4981
03:09:53,326 --> 03:09:54,294
THERE'S THE FACT THAT IN THE

4982
03:09:54,294 --> 03:09:55,728
DATA THAT IS AVAILABLE AND THAT

4983
03:09:55,728 --> 03:09:58,131
WE CAN USE FOR PRE-TRAINING AND

4984
03:09:58,131 --> 03:10:00,433
THAT'S GOING TO BE -- SOMEHOW

4985
03:10:00,433 --> 03:10:01,401
SHAPE THE PRE-TRAINING MODELS

4986
03:10:01,401 --> 03:10:03,936
THAT WILL BE USED IN THE FUTURE

4987
03:10:03,936 --> 03:10:06,806
IN THE DOWNSTREAM TASKS, THERE'S

4988
03:10:06,806 --> 03:10:09,309
ALSO MUCH MORE DATA AVAILABLE

4989
03:10:09,309 --> 03:10:10,743
ABOUT MORE COMMON NEUROLOGICAL

4990
03:10:10,743 --> 03:10:12,145
AND PSYCHIATRIC CONDITIONS THAN

4991
03:10:12,145 --> 03:10:13,680
FOR RARE DISEASES FOR INSTANCE

4992
03:10:13,680 --> 03:10:14,781
OR RARE CONDITIONS.

4993
03:10:14,781 --> 03:10:18,618
THAT IS ALSO A PROBLEM BECAUSE

4994
03:10:18,618 --> 03:10:20,053
WE MIGHT DEVELOP PRE-TRAINED

4995
03:10:20,053 --> 03:10:22,221
MODELS THAT ACTUALLY WORK VERY

4996
03:10:22,221 --> 03:10:26,626
WELL AS LONG AS WE ARE LOOKING

4997
03:10:26,626 --> 03:10:28,561
AT CLINICAL CONDITIONS THAT ARE

4998
03:10:28,561 --> 03:10:31,731
QUITE -- FOR WHICH WE HAVE A LOT

4999
03:10:31,731 --> 03:10:34,467
OF DATA AND THAT ARE, YEAH, THAT

5000
03:10:34,467 --> 03:10:36,903
ARE AVAILABLE ACROSS MANY

5001
03:10:36,903 --> 03:10:39,205
HOSPITALS THAT MADE THE DATASET

5002
03:10:39,205 --> 03:10:40,306
AVAILABLE, BUT IN RARE

5003
03:10:40,306 --> 03:10:41,507
CONDITIONS LIKE -- AND THIS IS

5004
03:10:41,507 --> 03:10:43,142
EXACTLY THE SAME PROBLEM WE

5005
03:10:43,142 --> 03:10:45,311
DISCUSSED WHEN WE TALK ABOUT

5006
03:10:45,311 --> 03:10:47,580
LARGE LANGUAGE MODELS, LOW

5007
03:10:47,580 --> 03:10:53,753
RESOURCE LANGUAGES, AND SOME

5008
03:10:53,753 --> 03:10:55,955
LANGUAGES ACROSS HUNDREDS THAT

5009
03:10:55,955 --> 03:10:57,490
EXIST IN AFRICA, ALSO INDIGENOUS

5010
03:10:57,490 --> 03:10:58,791
LANGUAGES ARE VERY GOOD EXAMPLES

5011
03:10:58,791 --> 03:11:00,326
OF SITUATIONS WHERE THE DATA

5012
03:11:00,326 --> 03:11:01,861
AVAILABLE IS SO SMALL COMPARED

5013
03:11:01,861 --> 03:11:03,830
TO THE MORE DOMINANT LANGUAGES

5014
03:11:03,830 --> 03:11:06,232
THAT THEY GET OVERSHADOWED AND

5015
03:11:06,232 --> 03:11:10,937
IN THE TRAINING,  PRE-TRAINED

5016
03:11:10,937 --> 03:11:14,340
MODELS A BIAS GETS INJECTED INTO

5017
03:11:14,340 --> 03:11:14,674
THIS.

5018
03:11:14,674 --> 03:11:15,742
WHAT ABOUT INDIVIDUALIZED

5019
03:11:15,742 --> 03:11:16,008
MEDICINE?

5020
03:11:16,008 --> 03:11:19,379
WE'RE TALKING ABOUT DEVELOPING

5021
03:11:19,379 --> 03:11:22,548
MODELS FOR INDIVIDUALS THAT'S

5022
03:11:22,548 --> 03:11:24,083
GOING TO BE FINE TUNED, WHAT

5023
03:11:24,083 --> 03:11:25,718
GUARANTEES DO WE HAVE THAT

5024
03:11:25,718 --> 03:11:27,353
PRE-TRAINING ON A LARGE COHORT

5025
03:11:27,353 --> 03:11:29,455
AND THEN FINE TUNING TO

5026
03:11:29,455 --> 03:11:30,423
INDIVIDUAL WILL ACTUALLY BE

5027
03:11:30,423 --> 03:11:32,825
BETTER THAN JUST TRYING TO FINE

5028
03:11:32,825 --> 03:11:36,562
TUNE ON A SINGLE INDIVIDUAL? 

5029
03:11:36,562 --> 03:11:42,168
THESE ARE OPEN QUESTIONS. 

5030
03:11:42,168 --> 03:11:44,537
YES, THAT WAS ALL I WANTED TO

5031
03:11:44,537 --> 03:11:45,872
SAY TODAY.

5032
03:11:45,872 --> 03:11:47,607
THESE ARE MY ACKNOWLEDGMENTS.

5033
03:11:47,607 --> 03:11:48,941
I THANK YOU FOR YOUR ATTENTION

5034
03:11:48,941 --> 03:11:56,015
AND I HOPE I DIDN'T GO TOO MUCH

5035
03:11:56,015 --> 03:11:58,451
OVER TIME.

5036
03:11:58,451 --> 03:11:59,085
THANKS.

5037
03:11:59,085 --> 03:11:59,652
>> TERRIFIC PRESENTATION.

5038
03:11:59,652 --> 03:12:04,357
OPENED UP A LOT OF QUESTIONS FOR

5039
03:12:04,357 --> 03:12:04,957
EVERYONE.

5040
03:12:04,957 --> 03:12:08,394
BEFORE WE TAKE QUESTIONS I WANT

5041
03:12:08,394 --> 03:12:10,363
TO TURN TO DR. DEBRA MATHEWS TO

5042
03:12:10,363 --> 03:12:15,735
THE GUY  -- GUIDE US THROUGH

5043
03:12:15,735 --> 03:12:16,836
ETHICAL THINKING, YOU ALREADY

5044
03:12:16,836 --> 03:12:18,237
STARTED TO INTRODUCE THOSE

5045
03:12:18,237 --> 03:12:19,138
ETHICAL QUESTIONS AND WE'LL TAKE

5046
03:12:19,138 --> 03:12:19,972
QUESTIONS AT THE END.

5047
03:12:19,972 --> 03:12:24,043
LET ME TURN IT OVER TO DR.

5048
03:12:24,043 --> 03:12:28,748
MATHEWS NOW.

5049
03:12:28,748 --> 03:12:35,321
>> TERRIFIC. 

5050
03:12:35,321 --> 03:12:37,523
LET'S SEE IF -- ARE FOLKS SEEING

5051
03:12:37,523 --> 03:12:39,459
MY FIRST SLIDE?

5052
03:12:39,459 --> 03:12:39,826
EXCELLENT.

5053
03:12:39,826 --> 03:12:43,296
>> YES, WE SEE IT.

5054
03:12:43,296 --> 03:12:43,729
>> GREAT.

5055
03:12:43,729 --> 03:12:45,364
FIRST, THANKS.

5056
03:12:45,364 --> 03:12:48,000
I'LL ADD MY THANKS TO THE

5057
03:12:48,000 --> 03:12:49,368
ORGANIZERS FOR THE INVITATION,

5058
03:12:49,368 --> 03:12:51,771
ALSO HELLO TO MY FRIENDS AND

5059
03:12:51,771 --> 03:12:53,272
COLLEAGUES WHO ARE ON THE

5060
03:12:53,272 --> 03:12:57,109
COMMITTEE, ON THE CALL TODAY.

5061
03:12:57,109 --> 03:12:58,945
APOLOGIES, I WASN'T ABLE TO JOIN

5062
03:12:58,945 --> 03:13:03,883
UNTIL 1:00 SO I DON'T KNOW WHAT

5063
03:13:03,883 --> 03:13:06,152
I MISSED, AND THEREFORE HOW

5064
03:13:06,152 --> 03:13:08,387
REDUNDANT SOME OF WHAT I PLAN TO

5065
03:13:08,387 --> 03:13:13,326
SAY MIGHT BE BUT HOPEFULLY IT

5066
03:13:13,326 --> 03:13:15,161
WILL BE COMPLEMENTARY TO WHAT'S

5067
03:13:15,161 --> 03:13:16,696
ALREADY BEEN SHARED.

5068
03:13:16,696 --> 03:13:19,198
IT'S CERTAINLY I THINK BUILDING

5069
03:13:19,198 --> 03:13:20,733
NICELY, A LOT OF DISCUSSIONS

5070
03:13:20,733 --> 03:13:23,836
THAT HAPPENED SINCE I DID JOIN

5071
03:13:23,836 --> 03:13:24,804
THE CALL.

5072
03:13:24,804 --> 03:13:27,406
SO I'LL ALSO START WITH A LAND

5073
03:13:27,406 --> 03:13:29,942
ACKNOWLEDGMENT, MY HOME IN

5074
03:13:29,942 --> 03:13:30,576
BALTIMORE CITY ALONG WITH JOHNS

5075
03:13:30,576 --> 03:13:34,847
HOPKINS UNIVERSITY SITS ON

5076
03:13:34,847 --> 03:13:45,424
UNCEDED LANDS  PISCATAWAY AND AD

5077
03:13:48,594 --> 03:13:48,861
SUSQUEHANNAK.

5078
03:13:48,861 --> 03:13:51,564
I'M GOING TO START WITH A BIT

5079
03:13:51,564 --> 03:13:58,838
MORE BACKGROUND ON A.I., ETHICS

5080
03:13:58,838 --> 03:14:00,573
AND GOVERNANCE, I WANT TO

5081
03:14:00,573 --> 03:14:02,108
INCLUDE IT IN MY SLIDES FOR

5082
03:14:02,108 --> 03:14:04,744
REFERENCE LATER IN CASE THAT'S

5083
03:14:04,744 --> 03:14:05,945
HELPFUL TO THE COMMITTEE.

5084
03:14:05,945 --> 03:14:09,215
OF COURSE THERE ARE LOTS OF

5085
03:14:09,215 --> 03:14:13,319
AMAZING OPPORTUNITIES OFFERED BY

5086
03:14:13,319 --> 03:14:15,154
THE IMPLEMENTATION OF ML BASED

5087
03:14:15,154 --> 03:14:18,658
METHODS IN MEDICINE AND IN

5088
03:14:18,658 --> 03:14:20,760
NEUROSCIENCE IN PARTICULAR,

5089
03:14:20,760 --> 03:14:21,861
HEALTH SERVICES MANAGEMENT,

5090
03:14:21,861 --> 03:14:28,301
PREDICTIVE MEDICINE, PATIENT AND

5091
03:14:28,301 --> 03:14:30,069
DATA DIAGNOSTICS, ET CETERA,

5092
03:14:30,069 --> 03:14:33,339
THAT COMES IN DEVELOPMENT USE,

5093
03:14:33,339 --> 03:14:37,710
IMPLEMENTATION COMES WITH BOTH

5094
03:14:37,710 --> 03:14:38,945
FAMILIAR AND NEW CONSTELLATION

5095
03:14:38,945 --> 03:14:41,647
OR HEIGHTENED SET OF ETHICAL

5096
03:14:41,647 --> 03:14:42,014
ISSUES.

5097
03:14:42,014 --> 03:14:42,982
FAMILIAR CHALLENGES INCLUDE MANY

5098
03:14:42,982 --> 03:14:48,354
THAT HAVE BEEN TALKED ABOUT

5099
03:14:48,354 --> 03:14:53,926
TODAY, CONCERNS ABOUT INFORMED

5100
03:14:53,926 --> 03:14:56,228
CONSENT, PRIVACY, INCIDENTAL

5101
03:14:56,228 --> 03:14:59,298
FINDINGS, RETURN OF RESULTS,

5102
03:14:59,298 --> 03:15:02,134
DATA OWNERSHIP, AND DATA

5103
03:15:02,134 --> 03:15:04,236
SHARING, INCLUDING DATA SHARING

5104
03:15:04,236 --> 03:15:07,707
UNDER MULTIPLE DIFFERENT SORT OF

5105
03:15:07,707 --> 03:15:08,474
ETHICAL REGIMES.

5106
03:15:08,474 --> 03:15:10,042
THE MOST SORT OF THE CURRENT

5107
03:15:10,042 --> 03:15:19,819
GOLD STANDARD OF COURSE ARE FAIR

5108
03:15:19,819 --> 03:15:20,686
PRINCIPLES, AND IMPORTANT TREND

5109
03:15:20,686 --> 03:15:22,221
OR IMPORTANT FORCE I WOULD SAY

5110
03:15:22,221 --> 03:15:26,058
OVER THE LAST COUPLE YEARS IS

5111
03:15:26,058 --> 03:15:26,826
INDIGENOUS DATA SOVEREIGNTY

5112
03:15:26,826 --> 03:15:31,197
MOVEMENT, AND I'LL TALK MORE

5113
03:15:31,197 --> 03:15:32,331
ABOUT THAT LATER.

5114
03:15:32,331 --> 03:15:34,934
BUT OFFERED A DIFFERENT SET OF

5115
03:15:34,934 --> 03:15:36,602
PRINCIPLES FOR DATA SHARING,

5116
03:15:36,602 --> 03:15:39,105
CARE PRINCE TELLS, COLLECTIVE

5117
03:15:39,105 --> 03:15:40,406
BENEFIT, AUTHORITY TO CONTROL,

5118
03:15:40,406 --> 03:15:42,274
RESPONSIBILITY AND ETHICS.

5119
03:15:42,274 --> 03:15:45,978
IT'S CLEAR THAT ETHICAL

5120
03:15:45,978 --> 03:15:48,280
PRINCIPLES ARE GUIDING YOU SHAPE

5121
03:15:48,280 --> 03:15:51,484
AND BUILD VERY DIFFERENT KINDS

5122
03:15:51,484 --> 03:15:54,520
OF DATA SHARING REGIMES.

5123
03:15:54,520 --> 03:15:58,124
DUAL USE, WE HAVE A LONG HISTORY

5124
03:15:58,124 --> 03:15:59,558
OF DEVELOPING NEW TECHNOLOGIES

5125
03:15:59,558 --> 03:16:02,061
FOR THE BENEFIT FOR EXAMPLE IN

5126
03:16:02,061 --> 03:16:04,130
MEDICINE OF THOSE AMONG US MOST

5127
03:16:04,130 --> 03:16:07,299
PRIVILEGED, AND THEN THE SAME

5128
03:16:07,299 --> 03:16:10,036
TECHNOLOGIES ARE TURNED TOWARDS

5129
03:16:10,036 --> 03:16:12,271
MARGINALIZED AND MINORITIZED

5130
03:16:12,271 --> 03:16:14,540
COMMUNITIES AND USED FOR

5131
03:16:14,540 --> 03:16:15,007
SURVEILLANCE.

5132
03:16:15,007 --> 03:16:17,843
IMPACT OF EARLY PUBLIC FAILURE

5133
03:16:17,843 --> 03:16:21,447
OF SOME NEW TECHNOLOGIES AND OF

5134
03:16:21,447 --> 03:16:23,282
A.I. IN PARTICULAR, FOR EXAMPLE

5135
03:16:23,282 --> 03:16:25,718
WHEN ChatGPT ENCOURAGED THE

5136
03:16:25,718 --> 03:16:28,487
REPORTER TO LEAVE HIS WIFE, AND

5137
03:16:28,487 --> 03:16:32,058
IMPORTANTLY FOR THE U.S. CONTEXT

5138
03:16:32,058 --> 03:16:33,626
AND IMPORTANT FOR THE

5139
03:16:33,626 --> 03:16:35,795
CONVERSATIONS THAT WE'LL HAVE

5140
03:16:35,795 --> 03:16:38,097
ABOUT CONSENT, AND RISK AND

5141
03:16:38,097 --> 03:16:41,167
BENEFIT, MANY PUBLICS HAVE NO

5142
03:16:41,167 --> 03:16:41,701
REASONABLE EXPECTATION OF

5143
03:16:41,701 --> 03:16:42,568
BENEFIT, BECAUSE HERE IN THE

5144
03:16:42,568 --> 03:16:44,503
U.S. WE DON'T HAVE A NATIONAL

5145
03:16:44,503 --> 03:16:46,972
HEALTH CARE SYSTEM, DON'T HAVE A

5146
03:16:46,972 --> 03:16:50,910
RIGHT TO HEALTH CARE, AND THAT

5147
03:16:50,910 --> 03:16:52,011
DOESN'T CHANGE WITH INCREASING

5148
03:16:52,011 --> 03:16:55,815
CALLS TO INCREASE DIVERSITY IN

5149
03:16:55,815 --> 03:16:56,082
RESEARCH.

5150
03:16:56,082 --> 03:17:00,019
I DON'T THINK WE HAVE SORT OF

5151
03:17:00,019 --> 03:17:01,520
FULLY -- IT'S LONG RECOGNIZED

5152
03:17:01,520 --> 03:17:03,723
BUT NOT SOMETHING WE FULLY

5153
03:17:03,723 --> 03:17:04,590
GRAPPLED WITH.

5154
03:17:04,590 --> 03:17:07,526
NOW, IN A.I. IN PARTICULAR,

5155
03:17:07,526 --> 03:17:10,830
THERE IS NOT NECESSARILY NOVEL

5156
03:17:10,830 --> 03:17:13,999
ISSUES, BUT IN NOVEL

5157
03:17:13,999 --> 03:17:16,302
CONSTELLATION OF ISSUES, THESE

5158
03:17:16,302 --> 03:17:18,370
INCLUDE -- ONE OF THE I THINK

5159
03:17:18,370 --> 03:17:22,308
MORE NOVEL AMONG THEM IS JUST --

5160
03:17:22,308 --> 03:17:26,946
YOU'VE HEARD ABOUT ALREADY

5161
03:17:26,946 --> 03:17:27,813
TODAY, SCALE, PERVASIVENESS,

5162
03:17:27,813 --> 03:17:29,115
INVISIBILITY OF TOOLS AND

5163
03:17:29,115 --> 03:17:32,952
EVOLUTION OVER TIME AS THEY ARE

5164
03:17:32,952 --> 03:17:33,285
USED.

5165
03:17:33,285 --> 03:17:37,323
ALSO IN CONTRAST TO OTHER

5166
03:17:37,323 --> 03:17:41,193
TECHNOLOGIES THAT WERE IN B

5167
03:17:41,193 --> 03:17:43,362
IOMEDICINE THAT WE'RE FAMILIAR

5168
03:17:43,362 --> 03:17:46,565
WITH, WHERE THERE ARE ETHICAL

5169
03:17:46,565 --> 03:17:48,300
ISSUES, MANY ETHICAL ISSUES OF

5170
03:17:48,300 --> 03:17:52,004
COURSE, AND THEY ARE OFTEN

5171
03:17:52,004 --> 03:17:53,839
AROUND HOW THOSE DATA OR

5172
03:17:53,839 --> 03:17:55,407
TECHNOLOGIES ARE USED.

5173
03:17:55,407 --> 03:17:59,545
IN A.I., THE ETHICS IS BAKED IN.

5174
03:17:59,545 --> 03:18:01,947
IT'S OFTEN BAKED IN

5175
03:18:01,947 --> 03:18:02,314
UNINTENTIONALLY.

5176
03:18:02,314 --> 03:18:05,251
AND IT CAN BE BAKED IN IN MANY

5177
03:18:05,251 --> 03:18:06,986
DIFFERENT WAYS, INCLUDING BIAS

5178
03:18:06,986 --> 03:18:11,457
TRAINING DATA WHICH I KNOW HAS

5179
03:18:11,457 --> 03:18:12,358
ALREADY BEEN DISCUSSED,

5180
03:18:12,358 --> 03:18:13,225
INAPPROPRIATE LABELING WHEN THE

5181
03:18:13,225 --> 03:18:14,660
PROXY YOU CHOOSE FOR THE THING

5182
03:18:14,660 --> 03:18:17,062
YOU'RE TRYING TO PREDICT IS NOT

5183
03:18:17,062 --> 03:18:17,830
THE RIGHT PROXY.

5184
03:18:17,830 --> 03:18:22,434
I'LL COME TO AN EXAMPLE IN A

5185
03:18:22,434 --> 03:18:23,736
SECOND.

5186
03:18:23,736 --> 03:18:25,070
NARROW OR INADEQUATE

5187
03:18:25,070 --> 03:18:26,705
FUNCTIONALITY, APPLE WATCH

5188
03:18:26,705 --> 03:18:28,674
INITIALLY HAD NO MENSTRUAL

5189
03:18:28,674 --> 03:18:31,610
TRACKER DESPITE THE FACT ABOUT

5190
03:18:31,610 --> 03:18:36,315
HALF THE POPULATION MENSTRUATES,

5191
03:18:36,315 --> 03:18:38,384
FAIRNESS HAS BEEN MENTIONED

5192
03:18:38,384 --> 03:18:44,757
TODAY, AND THEN THE

5193
03:18:44,757 --> 03:18:45,558
INSCRUTABILITY AND BLACK BOX

5194
03:18:45,558 --> 03:18:47,459
THAT COMES WITH SOME OF THESE

5195
03:18:47,459 --> 03:18:48,994
TECHNOLOGIES.

5196
03:18:48,994 --> 03:18:52,631
IN THE CONTEXT OF HEALTH CARE,

5197
03:18:52,631 --> 03:18:56,235
I'VE INCLUDED TWO EXAMPLES NOT

5198
03:18:56,235 --> 03:18:56,902
DIRECTLY NEURO-EXAMPLES BUT

5199
03:18:56,902 --> 03:18:58,838
STILL GIVE A SENSE OF KINDS OF

5200
03:18:58,838 --> 03:19:01,240
ISSUES A.I. PRESENTS IN THE

5201
03:19:01,240 --> 03:19:02,775
HEALTHCARE SPACE.

5202
03:19:02,775 --> 03:19:04,543
FIRST IS ABOUT THE LABELING

5203
03:19:04,543 --> 03:19:07,479
ISSUE THAT I MENTIONED, A VERY

5204
03:19:07,479 --> 03:19:09,148
FAMOUS PAPER, MANY OF YOU ARE

5205
03:19:09,148 --> 03:19:13,752
FAMILIAR WITH IT.

5206
03:19:13,752 --> 03:19:17,890
IN WHICH THEY -- OBERMEYER

5207
03:19:17,890 --> 03:19:19,091
DEMONSTRATED THIS TOOL BEING

5208
03:19:19,091 --> 03:19:24,029
USED BY LARGE HEALTH SYSTEMS WAS

5209
03:19:24,029 --> 03:19:25,998
USING HEALTH CARE COSTS AS A

5210
03:19:25,998 --> 03:19:28,934
PROXY FOR HEALTH CARE NEED AND

5211
03:19:28,934 --> 03:19:29,935
BECAUSE OF STRUCTURAL INEQUITY

5212
03:19:29,935 --> 03:19:33,873
IN THE U.S. HEALTH CARE SYSTEM,

5213
03:19:33,873 --> 03:19:37,142
IT WAS -- BLACK PATIENTS WERE

5214
03:19:37,142 --> 03:19:39,879
CONSIDERABLY SICKER THAN WHITE

5215
03:19:39,879 --> 03:19:40,579
PATIENTS.

5216
03:19:40,579 --> 03:19:48,220
THE SECOND EXAMPLE BY GOODMAN ET

5217
03:19:48,220 --> 03:19:50,422
AL., A.I.-GENERATED CLINICAL

5218
03:19:50,422 --> 03:19:52,725
SUMMARIES, ASKED AN LLM TO

5219
03:19:52,725 --> 03:19:57,096
GENERATE CLINICAL SUM  SUMMARIEN

5220
03:19:57,096 --> 03:19:59,465
THE SAME RECORD REPEATEDLY,

5221
03:19:59,465 --> 03:20:00,466
SUMMARIES PRODUCED WERE

5222
03:20:00,466 --> 03:20:02,668
DIFFERENT, VARIED BY

5223
03:20:02,668 --> 03:20:03,302
ORGANIZATION, PHRASING,

5224
03:20:03,302 --> 03:20:06,805
INCLUSION OR EXCLUSION OF

5225
03:20:06,805 --> 03:20:09,575
PARTICULAR CLINICAL DETAILS,

5226
03:20:09,575 --> 03:20:11,877
WHICH COULD REASONABLY SHAPE

5227
03:20:11,877 --> 03:20:13,612
CLINICAL DECISION MAKING IN WAYS

5228
03:20:13,612 --> 03:20:15,114
THAT MIGHT BE CONCERNING.

5229
03:20:15,114 --> 03:20:16,682
AND THESE ARE THE KINDS OF TOOLS

5230
03:20:16,682 --> 03:20:19,752
THAT WILL ARE UNLIKELY TO FALL

5231
03:20:19,752 --> 03:20:22,821
UNDER FDA'S AUTHORITY. 

5232
03:20:22,821 --> 03:20:24,356
OR UNDERGO CLINICAL TRIALS.

5233
03:20:24,356 --> 03:20:30,896
SO, AT THIS POINT WE HAVE MANY

5234
03:20:30,896 --> 03:20:33,098
EFFORTS AT GOVERNANCE OF A.I.,

5235
03:20:33,098 --> 03:20:34,533
SETS OF RECOMMENDATIONS, BUT

5236
03:20:34,533 --> 03:20:37,369
WE -- AGAIN I'M JUST BREEZING

5237
03:20:37,369 --> 03:20:38,804
THROUGH THESE BUT THESE ARE SOME

5238
03:20:38,804 --> 03:20:41,874
THAT MIGHT BE OF INTEREST TO THE

5239
03:20:41,874 --> 03:20:43,208
COMMITTEE IN PARTICULAR.

5240
03:20:43,208 --> 03:20:45,844
THIS ONE, THIS CAME OUT OF NIH,

5241
03:20:45,844 --> 03:20:54,687
OUT OF A WORKING GROUP OF THE

5242
03:20:54,687 --> 03:20:56,789
NEXTRAC, FOCUSED ON DATA SCIENCE

5243
03:20:56,789 --> 03:20:59,925
AND EMERGING TECH IN BIOMEDICAL

5244
03:20:59,925 --> 03:21:00,492
RESEARCH.

5245
03:21:00,492 --> 03:21:02,461
THIS REPORT CAME OUT A YEAR AGO,

5246
03:21:02,461 --> 03:21:05,864
AGAIN I HAVE THE RECOMMENDATIONS

5247
03:21:05,864 --> 03:21:08,400
HERE, I'M NOT GOING INTO DETAIL

5248
03:21:08,400 --> 03:21:09,902
THERE BUT IT'S AVAILABLE FOR

5249
03:21:09,902 --> 03:21:12,104
YOUR REFERENCE, AND I'M HAPPY TO

5250
03:21:12,104 --> 03:21:13,839
ANSWER QUESTIONS.

5251
03:21:13,839 --> 03:21:17,009
AND THEN OF COURSE THIS

5252
03:21:17,009 --> 03:21:17,876
DISCUSSION DRAFT THAT WAS

5253
03:21:17,876 --> 03:21:22,581
CIRCULATED TO THE COMMITTEE AND

5254
03:21:22,581 --> 03:21:24,450
IDENTIFIED CONDUCT, CODE OF

5255
03:21:24,450 --> 03:21:25,884
CONDUCT PRINCIPLES FOR USE OF

5256
03:21:25,884 --> 03:21:27,987
A.I. IN HEALTH CARE AND I'LL

5257
03:21:27,987 --> 03:21:31,123
MENTION ONE OF MY HOPKINS

5258
03:21:31,123 --> 03:21:34,093
COLLEAGUES WAS INVOLVED WITH

5259
03:21:34,093 --> 03:21:35,194
THIS WORK.

5260
03:21:35,194 --> 03:21:37,396
AND WE'RE -- THERE ARE MANY SUCH

5261
03:21:37,396 --> 03:21:39,131
DOCUMENT BUS WE'RE STILL IN THE

5262
03:21:39,131 --> 03:21:41,867
EARLY DAYS AND STILL VERY MUCH

5263
03:21:41,867 --> 03:21:45,904
IN THE SPACE OF ETHICS, NORMS,

5264
03:21:45,904 --> 03:21:47,239
GOVERNANCE, TRYING TO CATCH UP

5265
03:21:47,239 --> 03:21:49,742
EVEN A LITTLE BIT TO WHERE THE

5266
03:21:49,742 --> 03:21:52,144
SCIENCE IS BUT THE SCIENCE IS

5267
03:21:52,144 --> 03:21:57,850
NOT STOPPING. 

5268
03:21:57,850 --> 03:22:00,452
IT JUST IS ACCELERATING.

5269
03:22:00,452 --> 03:22:01,553
AND GIVEN THE EXPERTISE AND ROLE

5270
03:22:01,553 --> 03:22:02,855
OF THIS COMMITTEE THERE'S NO

5271
03:22:02,855 --> 03:22:05,724
NEED FOR ME TO REHEARSE THE

5272
03:22:05,724 --> 03:22:07,926
ISSUES RAISED BY

5273
03:22:07,926 --> 03:22:08,594
NEUROTECHNOLOGY.

5274
03:22:08,594 --> 03:22:10,863
NEUROSCIENCE INDEPENDENT OF ITS

5275
03:22:10,863 --> 03:22:14,166
CONVERGENCE WITH A.I., LISTED A

5276
03:22:14,166 --> 03:22:14,933
FEW RELEVANT CONSIDERATIONS

5277
03:22:14,933 --> 03:22:21,073
HERE, SOME OF WHICH HAVE ALREADY

5278
03:22:21,073 --> 03:22:24,810
BEEN BROUGHT UP.

5279
03:22:24,810 --> 03:22:26,345
SO, BECAUSE OF LIMITATIONS OF

5280
03:22:26,345 --> 03:22:28,647
TIME, THERE'S A WHOLE BUNCH OF

5281
03:22:28,647 --> 03:22:31,583
ARGUE MENTATION AND SUPPORT THAT

5282
03:22:31,583 --> 03:22:33,352
I HAVEN'T SHOWED.

5283
03:22:33,352 --> 03:22:35,087
BUT I WANTED TO SUMMARIZE WHERE

5284
03:22:35,087 --> 03:22:37,723
I SEE US BEING IN TERMS OF THE

5285
03:22:37,723 --> 03:22:39,792
PATIENT AND PARTICIPANT SIDE OF

5286
03:22:39,792 --> 03:22:42,828
THINGS IN THIS RESEARCH AREA.

5287
03:22:42,828 --> 03:22:45,497
AND THE SCIENCE AND GOVERNANCE

5288
03:22:45,497 --> 03:22:49,435
FOR ETHICS SIDE. 

5289
03:22:49,435 --> 03:22:52,371
SO, AS HAS MENTIONED, PATIENTS

5290
03:22:52,371 --> 03:22:53,238
AND RESEARCH PARTICIPANTS

5291
03:22:53,238 --> 03:22:55,774
GENERALLY SUPPORT AND ARE HAPPY

5292
03:22:55,774 --> 03:22:57,242
TO CONTRIBUTE DATA TO RESEARCH.

5293
03:22:57,242 --> 03:23:01,246
BUT THEY DO WANT TO BE ASKED.

5294
03:23:01,246 --> 03:23:03,649
THEY ARE ALSO LESS WILLING TO

5295
03:23:03,649 --> 03:23:04,850
SUPPORT AND CONTRIBUTE THEIR

5296
03:23:04,850 --> 03:23:08,821
DATA TO RESEARCH AND DEVELOPMENT

5297
03:23:08,821 --> 03:23:09,755
INVOLVING COMMERCIAL ENTITIES,

5298
03:23:09,755 --> 03:23:12,424
THIS WAS FOUND NOT ONLY IN

5299
03:23:12,424 --> 03:23:18,897
PUBLIC ENGAGEMENT THAT WE DID AS

5300
03:23:18,897 --> 03:23:20,199
PART OF THE NEXTRAC WORKING

5301
03:23:20,199 --> 03:23:22,167
GROUP BUT ALSO BY OTHERS.

5302
03:23:22,167 --> 03:23:25,571
I WANT TO POINT OUT, THIS WAS

5303
03:23:25,571 --> 03:23:27,206
ALLUDED TO, AS WELL, PATIENTS

5304
03:23:27,206 --> 03:23:33,112
WITH SERIOUS DISEASE AND PARENTS

5305
03:23:33,112 --> 03:23:34,746
OF KIDS WITH SERIOUS DISEASE

5306
03:23:34,746 --> 03:23:37,816
OFTEN HAVE A DIFFERENT

5307
03:23:37,816 --> 03:23:39,918
RISK/BENEFIT ANALYSIS OF

5308
03:23:39,918 --> 03:23:41,854
CONTRIBUTING TO -- CONTRIBUTING

5309
03:23:41,854 --> 03:23:42,855
DATA TO RESEARCH PARTICULARLY

5310
03:23:42,855 --> 03:23:45,057
WHERE DATA IS AT RISK OF

5311
03:23:45,057 --> 03:23:46,024
RE-IDENTIFICATION, ET CETERA.

5312
03:23:46,024 --> 03:23:47,559
LIKE THEY DON'T CARE, RIGHT?

5313
03:23:47,559 --> 03:23:50,395
THEY JUST WANT BENEFIT.

5314
03:23:50,395 --> 03:23:53,365
THEY JUST WANT FORWARD MOTION TO

5315
03:23:53,365 --> 03:23:55,334
BE MADE, PARTICULARLY OR OFTEN

5316
03:23:55,334 --> 03:23:58,303
PARTICULARLY IN RARE DISEASE.

5317
03:23:58,303 --> 03:24:01,340
SO, THIS DIFFERS ACROSS PEOPLE,

5318
03:24:01,340 --> 03:24:03,876
OF COURSE.

5319
03:24:03,876 --> 03:24:07,813
WE KNOW FROM QUITE A BIT OF

5320
03:24:07,813 --> 03:24:10,415
RESEARCH THAT THE PRIMARY DRIVER

5321
03:24:10,415 --> 03:24:13,118
OF PEOPLE'S WILLINGNESS TO

5322
03:24:13,118 --> 03:24:15,454
CONTRIBUTE IS TRUST IN THE

5323
03:24:15,454 --> 03:24:17,222
PERSON OR INSTITUTION ASKING

5324
03:24:17,222 --> 03:24:18,423
THEM AND ALTRUISM.

5325
03:24:18,423 --> 03:24:22,027
PEOPLE WANT TO CONTRIBUTE TO

5326
03:24:22,027 --> 03:24:23,028
BENEFIT FOR OTHERS.

5327
03:24:23,028 --> 03:24:26,832
FOR A VARIETY OF BOTH HISTORICAL

5328
03:24:26,832 --> 03:24:28,367
AND CONTEMPORARY REASONS, A BIG

5329
03:24:28,367 --> 03:24:31,203
SEGMENT OF OUR POPULATION IS

5330
03:24:31,203 --> 03:24:32,838
LESS WILLING TO CONTRIBUTE, AND

5331
03:24:32,838 --> 03:24:36,008
I MENTIONED BEFORE HAVE LESS

5332
03:24:36,008 --> 03:24:37,442
ACCESS TO BENEFIT THAT COMES

5333
03:24:37,442 --> 03:24:41,180
FROM THIS KIND OF RESEARCH.

5334
03:24:41,180 --> 03:24:43,148
AND THERE ARE INCREASING CALLS

5335
03:24:43,148 --> 03:24:44,216
FROM INDIVIDUALS AND COMMUNITIES

5336
03:24:44,216 --> 03:24:45,450
FOR RETURN OF VALUE.

5337
03:24:45,450 --> 03:24:53,525
SO WE ALSO HEARD THIS IN OUR

5338
03:24:53,525 --> 03:24:58,330
PUBLIC ENGAGEMENT WITH THE

5339
03:24:58,330 --> 03:25:01,400
NEXTRAC WORKING GROUP,

5340
03:25:01,400 --> 03:25:02,100
COLLEAGUES WORKING IN GENETICS

5341
03:25:02,100 --> 03:25:04,803
AND A.I., ET CETERA.

5342
03:25:04,803 --> 03:25:08,640
ON THE SCIENCE AND GOVERNANCE

5343
03:25:08,640 --> 03:25:10,943
SIDE OUR RULES WERE BUILD FOR

5344
03:25:10,943 --> 03:25:13,579
INTERVENTIONAL RESEARCH IN A

5345
03:25:13,579 --> 03:25:16,949
CONTEXT OF RARE EXPENSIVE

5346
03:25:16,949 --> 03:25:18,150
SEQUENCING AND DATA COLLECTION,

5347
03:25:18,150 --> 03:25:19,985
WHERE THERE WAS A REASONABLE

5348
03:25:19,985 --> 03:25:21,320
EXPECTATION THAT YOU COULD

5349
03:25:21,320 --> 03:25:23,956
DE-IDENTIFY DATA AND THAT DATA

5350
03:25:23,956 --> 03:25:27,125
COULD REMAIN PRIVATE AND SECURE,

5351
03:25:27,125 --> 03:25:32,898
SO INFORMATIONAL RISKS COULD BE

5352
03:25:32,898 --> 03:25:33,365
ELIMINATED.

5353
03:25:33,365 --> 03:25:37,603
AND CONSENT WAS THEREFORE NOT

5354
03:25:37,603 --> 03:25:38,303
NEEDED.

5355
03:25:38,303 --> 03:25:40,472
I THINK INFORMATION ECOSYSTEM

5356
03:25:40,472 --> 03:25:42,708
HAS CHANGED, THAT TECHNOLOGY HAS

5357
03:25:42,708 --> 03:25:45,377
CHANGED, AND WE NEED TO START

5358
03:25:45,377 --> 03:25:46,178
RESPONDING TO THAT.

5359
03:25:46,178 --> 03:25:49,681
ALSO IN THE U.S. THERE'S A

5360
03:25:49,681 --> 03:25:51,316
STRONG FOCUS ON AUTONOMY IN THE

5361
03:25:51,316 --> 03:25:53,385
U.S. AND WEST GENERALLY NOT

5362
03:25:53,385 --> 03:25:58,090
EVERYWHERE IN THE WORLD, BUT

5363
03:25:58,090 --> 03:25:59,625
AUTONOMY IS PRIZED OFTEN ABOVE

5364
03:25:59,625 --> 03:26:02,127
CONCERNS ABOUT JUSTICE AND

5365
03:26:02,127 --> 03:26:02,561
EQUITY.

5366
03:26:02,561 --> 03:26:08,066
AND WE DON'T HAVE GREAT WAYS OF

5367
03:26:08,066 --> 03:26:10,469
MEASURING IMPACTS OF RESEARCH

5368
03:26:10,469 --> 03:26:11,670
AND TECHNOLOGY AT THE POPULATION

5369
03:26:11,670 --> 03:26:13,772
LEVEL, AND IT'S NOT SOMETHING WE

5370
03:26:13,772 --> 03:26:17,709
GENERALLY TAKE INTO ACCOUNT IN

5371
03:26:17,709 --> 03:26:22,180
THE REGULATORY SPACE.

5372
03:26:22,180 --> 03:26:22,848
SCIENCE HAS EVOLVED

5373
03:26:22,848 --> 03:26:23,181
DRAMATICALLY.

5374
03:26:23,181 --> 03:26:27,452
THIS A.I. IS MOVING INCREDIBLY

5375
03:26:27,452 --> 03:26:27,686
RAPIDLY.

5376
03:26:27,686 --> 03:26:29,521
SO OUR ETHICAL AND GOVERNANCE

5377
03:26:29,521 --> 03:26:32,557
FRAMEWORKS NEED TO AS WELL.

5378
03:26:32,557 --> 03:26:37,296
THE SCIENCE DOESN'T EXIST IN A

5379
03:26:37,296 --> 03:26:37,629
VACUUM.

5380
03:26:37,629 --> 03:26:38,830
ALONDRA NELSON'S OF THE SOCIAL

5381
03:26:38,830 --> 03:26:41,033
LIFE OF DNA, THE FACT THERE ARE

5382
03:26:41,033 --> 03:26:43,969
MANY DIRECT TO CONSUMER

5383
03:26:43,969 --> 03:26:46,138
NEUROTECHNOLOGIES AVAILABLE, WE

5384
03:26:46,138 --> 03:26:49,441
ALL HAVE ACCESS TO ChatGPT, ET

5385
03:26:49,441 --> 03:26:49,675
CETERA.

5386
03:26:49,675 --> 03:26:52,010
PEOPLE'S  UNDERSTANDING OF WHATS

5387
03:26:52,010 --> 03:26:53,945
GOING ON IN THE SCIENCE IS GOING

5388
03:26:53,945 --> 03:26:57,316
TO BE SHAPED BY THEIR EXPERIENCE

5389
03:26:57,316 --> 03:27:01,153
OF THESE TECHNOLOGIES IN THE

5390
03:27:01,153 --> 03:27:01,386
WILD.

5391
03:27:01,386 --> 03:27:02,621
FINALLY EQUITY DISTRIBUTION OF

5392
03:27:02,621 --> 03:27:09,594
RISKS AND BENEFITS IS IMPORTANT

5393
03:27:09,594 --> 03:27:13,899
NOT JUST AMONG RACIAL AND

5394
03:27:13,899 --> 03:27:15,667
ETHNICS GROUPS BUT RESEARCHERS,

5395
03:27:15,667 --> 03:27:16,935
AS ENVIRONMENT HAS SHIFTED RISKS

5396
03:27:16,935 --> 03:27:18,070
AND SHIFT AND WE HAVEN'T

5397
03:27:18,070 --> 03:27:21,340
NECESSARILY KEPT UP IN TERMS OF

5398
03:27:21,340 --> 03:27:23,208
THINKING ABOUT THE BENEFITS SIDE

5399
03:27:23,208 --> 03:27:27,346
OF THAT, ESPECIALLY WHEN YOU'RE

5400
03:27:27,346 --> 03:27:31,616
TALKING ABOUT UNCONSENTED DATA. 

5401
03:27:31,616 --> 03:27:34,686
SO, THAT BRINGS ME TO THE PAPER.

5402
03:27:34,686 --> 03:27:37,422
AND I THINK IT'S BECAUSE OF THE

5403
03:27:37,422 --> 03:27:39,958
NATURE OF A.I. ML-BASED

5404
03:27:39,958 --> 03:27:40,892
TECHNOLOGIES AND OUR DUTIES

5405
03:27:40,892 --> 03:27:42,728
WITHIN THE HEALTH CARE SYSTEM TO

5406
03:27:42,728 --> 03:27:45,297
IMPROVE WELL-BEING WE NEED TO BE

5407
03:27:45,297 --> 03:27:46,398
THINKING NOT JUST ABOUT

5408
03:27:46,398 --> 03:27:48,467
INDIVIDUAL EXPERIMENTS OR MODELS

5409
03:27:48,467 --> 03:27:52,637
BUT THE LIFE CYCLE AND LIFESPAN

5410
03:27:52,637 --> 03:27:55,407
OF WHAT WE'RE CREATING.

5411
03:27:55,407 --> 03:27:58,643
SO, FROM -- AND THIS IS NOT

5412
03:27:58,643 --> 03:28:02,047
NECESSARILY EVERY SINGLE STEP.

5413
03:28:02,047 --> 03:28:05,650
THIS IS JUST SORT OF ONE SET

5414
03:28:05,650 --> 03:28:10,389
OF -- ONE WAY OF SLICING UP THE

5415
03:28:10,389 --> 03:28:11,056
PIE.

5416
03:28:11,056 --> 03:28:13,325
AND SO STARTING WITH DATA

5417
03:28:13,325 --> 03:28:15,227
COLLECTION AND, YOU KNOW, HERE

5418
03:28:15,227 --> 03:28:17,295
AGAIN THERE'S BEEN A FAIR AMOUNT

5419
03:28:17,295 --> 03:28:18,697
OF CONVERSATION ABOUT THIS, BUT

5420
03:28:18,697 --> 03:28:21,133
CONSENT, WHAT ARE THE DATA

5421
03:28:21,133 --> 03:28:24,936
TYPES, WHAT IS BEING DONE ABOUT

5422
03:28:24,936 --> 03:28:26,138
IDENTIFIABILITY, HOW ARE DATA

5423
03:28:26,138 --> 03:28:28,673
BEING STORED, HOW ARE DATA BEING

5424
03:28:28,673 --> 03:28:30,642
SHARED, ARE THEY BEING SOLD, WHO

5425
03:28:30,642 --> 03:28:33,278
OWNS THE DATA, WHO HAS ACCESS TO

5426
03:28:33,278 --> 03:28:35,480
THE DATA, HOW ARE THEY BEING

5427
03:28:35,480 --> 03:28:36,248
USED, ET CETERA.

5428
03:28:36,248 --> 03:28:38,483
I THINK EVEN THOUGH THERE ARE

5429
03:28:38,483 --> 03:28:41,686
INCREASING CALLS FOR A.I.-READY

5430
03:28:41,686 --> 03:28:44,423
DATASETS, THERE IS NOT CONSENSUS

5431
03:28:44,423 --> 03:28:45,991
ABOUT WHAT CONSTITUTES ETHICALLY

5432
03:28:45,991 --> 03:28:50,562
SOURCED DATA FOR A.I.

5433
03:28:50,562 --> 03:28:52,431
DATA PROCESSING, RIGHT, WHERE IS

5434
03:28:52,431 --> 03:28:53,632
THAT HAPPENING, AND THE SECURITY

5435
03:28:53,632 --> 03:28:55,267
OF THAT, HOW IT'S BEING CLEANED,

5436
03:28:55,267 --> 03:28:56,468
HOW IT'S BEING LABELED, OR IN

5437
03:28:56,468 --> 03:28:58,136
THE CASE OF THE PAPER CHUNKED,

5438
03:28:58,136 --> 03:29:01,840
I'LL COME BACK TO THAT IN A SEC.

5439
03:29:01,840 --> 03:29:04,910
MODEL DEVELOPMENT INCLUDING

5440
03:29:04,910 --> 03:29:08,513
ENVIRONMENTAL IMPACT OF THAT.

5441
03:29:08,513 --> 03:29:09,614
TESTING, APPLICATION

5442
03:29:09,614 --> 03:29:10,282
DEVELOPMENT, THESE ARE

5443
03:29:10,282 --> 03:29:12,350
DOWNSTREAM FROM WHERE THIS PAPER

5444
03:29:12,350 --> 03:29:15,086
IS BUT APPLICATION DEVELOPMENT,

5445
03:29:15,086 --> 03:29:17,923
IMPLEMENTATION, AUDITING FOR NOT

5446
03:29:17,923 --> 03:29:20,325
ONLY THE IMPACT OF THE ALGORITHM

5447
03:29:20,325 --> 03:29:21,960
ON INDIVIDUALS AND POPULATIONS

5448
03:29:21,960 --> 03:29:25,497
BUT ALSO DRIFT OVER TIME,

5449
03:29:25,497 --> 03:29:28,467
ENSURING THAT THE TOOL IS

5450
03:29:28,467 --> 03:29:29,334
STILL -- THE ALGORITHM IS STILL

5451
03:29:29,334 --> 03:29:30,869
DOING WHAT IT WAS INTENDED TO DO

5452
03:29:30,869 --> 03:29:33,371
IN THE FIRST PLACE.

5453
03:29:33,371 --> 03:29:36,007
AND THEN HIGHER LEVEL QUESTION,

5454
03:29:36,007 --> 03:29:37,742
WHOSE VALUES ARE INFORMING THE

5455
03:29:37,742 --> 03:29:41,146
DECISIONS THAT ARE MADE AT EACH

5456
03:29:41,146 --> 03:29:42,047
OF THESE STEPS.

5457
03:29:42,047 --> 03:29:44,015
SO IN THE PARTICULAR CASE OF THE

5458
03:29:44,015 --> 03:29:45,550
PAPER, FIRST IT'S IMPORTANT TO

5459
03:29:45,550 --> 03:29:48,753
NOTE I'M NOT SUGGESTING THERE'S

5460
03:29:48,753 --> 03:29:49,387
ANYTHING FUNDAMENTALLY FLAWED

5461
03:29:49,387 --> 03:29:51,256
ABOUT THIS WORK OR IT DOESN'T

5462
03:29:51,256 --> 03:29:55,727
CONFORM TO CURRENT STANDARDS BUT

5463
03:29:55,727 --> 03:29:59,564
WAS ASKED TO RESPOND TO THIS

5464
03:29:59,564 --> 03:30:02,834
PAPER AND THE IDEA OF TRAINING

5465
03:30:02,834 --> 03:30:06,571
MODELS ON HUMAN BRAIN DATA.

5466
03:30:06,571 --> 03:30:11,977
SO, GOING FROM THAT SORT OF LIFE

5467
03:30:11,977 --> 03:30:14,913
CYCLE OR LIFESPAN APPROACH THAT

5468
03:30:14,913 --> 03:30:17,549
I JUST DESCRIBED, STARTING WITH

5469
03:30:17,549 --> 03:30:20,952
THE DATA, THE TRAINING DATA CAME

5470
03:30:20,952 --> 03:30:26,291
FROM THE TEMPLE UNIVERSITY EEG

5471
03:30:26,291 --> 03:30:26,858
CORPUS.

5472
03:30:26,858 --> 03:30:31,029
THIS IS CLINICAL DATA, THESE ARE

5473
03:30:31,029 --> 03:30:34,299
DATA THAT WERE COLLECTED FROM

5474
03:30:34,299 --> 03:30:37,168
CLINICAL TESTING FROM 2002 TO

5475
03:30:37,168 --> 03:30:39,704
2013 AND BEYOND.

5476
03:30:39,704 --> 03:30:42,207
IT HAD THE 18 HIPAA IDENTIFIERS

5477
03:30:42,207 --> 03:30:42,440
REMOVED.

5478
03:30:42,440 --> 03:30:46,278
THE DATA WERE DE-IDENTIFIED IN A

5479
03:30:46,278 --> 03:30:47,679
VARIETY OF WAYS.

5480
03:30:47,679 --> 03:30:50,515
I BELIEVE IT'S THE CASE THAT

5481
03:30:50,515 --> 03:30:55,554
THIS TEAM ONLY USED THE EEG DATA

5482
03:30:55,554 --> 03:30:57,622
FROM THIS CORPUS, THE CORPUS

5483
03:30:57,622 --> 03:30:59,591
ALSO INCLUDES SUMMARY OF

5484
03:30:59,591 --> 03:31:01,226
CLINICAL HISTORY AND MEDICATION

5485
03:31:01,226 --> 03:31:01,993
HISTORY.

5486
03:31:01,993 --> 03:31:04,195
AND THE MEDICAL FINDINGS.

5487
03:31:04,195 --> 03:31:09,200
THE COHORT ITSELF IS PRIMARILY

5488
03:31:09,200 --> 03:31:12,170
AND EPILEPSY COHORT, SOMETHING

5489
03:31:12,170 --> 03:31:18,543
LIKE 87% OF THE RECORDS HAD

5490
03:31:18,543 --> 03:31:19,744
LABEL OF EPILEPSY, A SMALL

5491
03:31:19,744 --> 03:31:30,221
PERCENTAGE HAD STROKE AS AN

5492
03:31:32,157 --> 03:31:32,390
INDICATION.

5493
03:31:32,390 --> 03:31:37,028
I WAS UNABLE TO LEARN WHAT THE

5494
03:31:37,028 --> 03:31:39,731
DIVERSITY OF THE PATIENTS IN

5495
03:31:39,731 --> 03:31:50,208
THIS DATABASE ARE, NOT JUST

5496
03:31:56,881 --> 03:31:58,883
RACIAL, ETHIC, SOCIOECONOMIC BUT

5497
03:31:58,883 --> 03:32:01,086
BAKING IN A NEUROTYPICAL BIAS

5498
03:32:01,086 --> 03:32:02,821
TRAINED ON THESE DATA.

5499
03:32:02,821 --> 03:32:04,889
AGAIN, THESE WERE CLINICAL DATA

5500
03:32:04,889 --> 03:32:07,959
SO THIS HAPPENED WITH IRB REVIEW

5501
03:32:07,959 --> 03:32:10,495
BUT I IMAGINE THEY DIDN'T STATE

5502
03:32:10,495 --> 03:32:12,564
IT EXPLICITLY IN THE PAPER, I

5503
03:32:12,564 --> 03:32:18,336
IMAGINE A WAIVER OF CONSENT.

5504
03:32:18,336 --> 03:32:19,904
WITH REGARD TO CONTENT, AGAIN, I

5505
03:32:19,904 --> 03:32:23,041
THINK THE TEAM ONLY USED THE EEG

5506
03:32:23,041 --> 03:32:27,412
DATA BUT HERE IS ONE OF THE --

5507
03:32:27,412 --> 03:32:31,683
I'M A HUMAN GENETICIST BY

5508
03:32:31,683 --> 03:32:33,785
TRAINING, A LOT OF THINGS IS

5509
03:32:33,785 --> 03:32:35,420
SHAPED BY THAT.

5510
03:32:35,420 --> 03:32:38,590
ONE OF THOSE WAYS IS WITH REGARD

5511
03:32:38,590 --> 03:32:40,125
TO THE CONTENT, AND WHETHER OR

5512
03:32:40,125 --> 03:32:43,194
NOT WE'RE USING JUST BRAIN DATA

5513
03:32:43,194 --> 03:32:49,768
OR WE'RE ALSO USING THE OTHER

5514
03:32:49,768 --> 03:32:51,002
KINDS OF

5515
03:32:51,002 --> 03:32:51,536
MEDICAL/BEHAVIORAL/SOCIAL

5516
03:32:51,536 --> 03:32:53,705
TORRENTS OF HEALTH DATA THAT

5517
03:32:53,705 --> 03:32:57,342
MIGHT INFORM WHAT THE BRAIN DATA

5518
03:32:57,342 --> 03:32:59,210
ARE ACTUALLY TELLING US, RIGHT?

5519
03:32:59,210 --> 03:33:05,884
AND HOW MUCH DATA DO WE NEED TO

5520
03:33:05,884 --> 03:33:07,285
UNDERSTAND OR IMBUE BROADER

5521
03:33:07,285 --> 03:33:08,687
MEANING INTO THE BRAIN DATA.

5522
03:33:08,687 --> 03:33:12,991
ON THE CHUNKING, THIS IS MORE OF

5523
03:33:12,991 --> 03:33:15,927
A TECHNICAL QUESTION, AND

5524
03:33:15,927 --> 03:33:18,463
REVEALS MY LACK OF EXPERTISE IN

5525
03:33:18,463 --> 03:33:20,432
THIS PARTICULAR SPACE, BUT THE

5526
03:33:20,432 --> 03:33:22,400
CHUNKING THAT WAS DONE, RIGHT,

5527
03:33:22,400 --> 03:33:26,971
THE ANALOGY IN MY BRAIN WAS TO

5528
03:33:26,971 --> 03:33:29,374
DNA, RIGHT?

5529
03:33:29,374 --> 03:33:30,709
AND INTRONS AND EXONS.

5530
03:33:30,709 --> 03:33:32,010
IF YOU'RE CHUNKING IN A

5531
03:33:32,010 --> 03:33:33,545
PARTICULAR WAY ARE YOU

5532
03:33:33,545 --> 03:33:35,080
DISRUPTING MEANING IN THE DATA?

5533
03:33:35,080 --> 03:33:38,149
I KNOW THE CHUNKS EACH HAD 10%

5534
03:33:38,149 --> 03:33:42,087
OVERLAP IN THE FINAL GO, BUT HOW

5535
03:33:42,087 --> 03:33:45,256
DO WE THINK ABOUT THOSE KINDS OF

5536
03:33:45,256 --> 03:33:46,257
DECISIONS, RIGHT?

5537
03:33:46,257 --> 03:33:47,358
LABELING WASN'T USED HERE BUT

5538
03:33:47,358 --> 03:33:47,792
CHUNKING WAS.

5539
03:33:47,792 --> 03:33:49,427
HOW DO WE THINK ABOUT THE IMPACT

5540
03:33:49,427 --> 03:33:53,465
OF THOSE KINDS OF DECISIONS ON

5541
03:33:53,465 --> 03:33:56,534
THE MEANING WE'RE GETTING OUT OF

5542
03:33:56,534 --> 03:33:58,837
THE ALGORITHM AND ITS DOWNSTREAM

5543
03:33:58,837 --> 03:33:59,037
USES?

5544
03:33:59,037 --> 03:34:01,239
OF COURSE CONCERNS ABOUT PRIVACY

5545
03:34:01,239 --> 03:34:01,906
AND IDENTIFIABILITY, AND THIS

5546
03:34:01,906 --> 03:34:03,742
CAME UP IN THE CHAT EARLIER,

5547
03:34:03,742 --> 03:34:03,942
RIGHT?

5548
03:34:03,942 --> 03:34:05,376
THE QUESTION OF IF YOU PUT THE

5549
03:34:05,376 --> 03:34:07,345
DATA IN, CAN THE DATA THEN LATER

5550
03:34:07,345 --> 03:34:08,346
BE EXTRACTED, RIGHT?

5551
03:34:08,346 --> 03:34:10,515
THIS WAS A FOUNDATION MODEL.

5552
03:34:10,515 --> 03:34:14,018
WE KNOW THAT THEY CAN SPIT OUT

5553
03:34:14,018 --> 03:34:16,321
TRAINING DATA.

5554
03:34:16,321 --> 03:34:21,593
SO CONCERNS ABOUT THAT IN THIS

5555
03:34:21,593 --> 03:34:22,026
CASE.

5556
03:34:22,026 --> 03:34:23,361
RISK/BENEFIT AND RETURN OF

5557
03:34:23,361 --> 03:34:26,831
VALUE, WHAT ARE QUESTIONS

5558
03:34:26,831 --> 03:34:29,134
DRIVING THIS RESEARCH, WHOSE

5559
03:34:29,134 --> 03:34:30,101
VALUES AND EXPERIENCE ARE

5560
03:34:30,101 --> 03:34:33,104
INFORMING THE QUESTIONS AND HOW

5561
03:34:33,104 --> 03:34:34,939
IS THE MODEL ULTIMATELY -- AGAIN

5562
03:34:34,939 --> 03:34:38,443
THIS IS PROOF OF PRINCIPLE BUT

5563
03:34:38,443 --> 03:34:43,681
DOWNSTREAM HOW MIGHT IT BE USED,

5564
03:34:43,681 --> 03:34:44,682
WHAT KINDS OF APPLICATIONS,

5565
03:34:44,682 --> 03:34:48,853
AGAIN GIVEN WE KNOW IN THIS CASE

5566
03:34:48,853 --> 03:34:50,388
THAT THE FOLKS WHOSE DATA

5567
03:34:50,388 --> 03:34:53,124
TRAINED THE MODEL DID NOT GIVE

5568
03:34:53,124 --> 03:34:55,093
CONSENT, HOW DO WE THINK ABOUT

5569
03:34:55,093 --> 03:34:56,828
RETURN OF VALUE GIVEN THE

5570
03:34:56,828 --> 03:34:58,830
POTENTIAL RISKS THEY HAVE TAKEN

5571
03:34:58,830 --> 03:35:03,201
ON WITH REGARD TO

5572
03:35:03,201 --> 03:35:03,601
RE-IDENTIFICATION.

5573
03:35:03,601 --> 03:35:06,237
I'VE INCLUDED A NUMBER OF

5574
03:35:06,237 --> 03:35:07,572
CITATIONS FOR THINGS THAT I

5575
03:35:07,572 --> 03:35:11,843
MENTIONED IN THE TALK AND THAT

5576
03:35:11,843 --> 03:35:13,678
OTHERWISE INFORMED MY

5577
03:35:13,678 --> 03:35:14,145
PRESENTATION.

5578
03:35:14,145 --> 03:35:22,987
AND WITH THAT I LOOK FORWARD TO

5579
03:35:22,987 --> 03:35:23,555
THE DISCUSSION.

5580
03:35:23,555 --> 03:35:26,157
>> THAT WAS A GREAT PLACE
FOR US

5581
03:35:26,157 --> 03:35:28,026
TO BEGIN THE CONVERSATION.

5582
03:35:28,026 --> 03:35:30,428
I KNOW THERE ARE A LOT OF

5583
03:35:30,428 --> 03:35:30,695
QUESTIONS.

5584
03:35:30,695 --> 03:35:33,164
GRACE AND I WILL LAUNCH THE

5585
03:35:33,164 --> 03:35:35,900
CONVERSATION, AND I THINK I'LL

5586
03:35:35,900 --> 03:35:38,636
START WITH A QUESTION FOR DR.

5587
03:35:38,636 --> 03:35:41,072
JERBI, PICKING UP ON WHAT SOME

5588
03:35:41,072 --> 03:35:43,775
OF WHAT DR. MATHEWS WAS

5589
03:35:43,775 --> 03:35:44,342
DISCUSSING.

5590
03:35:44,342 --> 03:35:48,813
I FOUND THIS PAPER FASCINATING,

5591
03:35:48,813 --> 03:35:51,249
BUT I WANT TO START WITH JUST

5592
03:35:51,249 --> 03:35:54,619
THE BASIC ISSUE OF BIAS, OF

5593
03:35:54,619 --> 03:35:56,921
UNDERSTANDING, YOU KNOW, HOW

5594
03:35:56,921 --> 03:35:59,657
GIVEN THAT EEG DATA

5595
03:35:59,657 --> 03:36:01,626
TRADITIONALLY IS INHERENTLY

5596
03:36:01,626 --> 03:36:04,362
BIASED IN MANY CASE, THAT OVER

5597
03:36:04,362 --> 03:36:06,130
TIME WE'VE SEEN THAT CERTAIN

5598
03:36:06,130 --> 03:36:07,465
GROUPS OF INDIVIDUALS ARE

5599
03:36:07,465 --> 03:36:08,800
EXCLUDED, THAT THIS IS CLINICAL

5600
03:36:08,800 --> 03:36:13,037
DATA SO LIMITED ALREADY TO A

5601
03:36:13,037 --> 03:36:19,010
SUBSET OF INDIVIDUALS, HOW ARE

5602
03:36:19,010 --> 03:36:20,378
YOU ADDRESSING BIAS IN THE MODEL

5603
03:36:20,378 --> 03:36:22,313
OR HOW DO YOU PLAN TO ADDRESS

5604
03:36:22,313 --> 03:36:24,315
BIAS IN THE MODEL GOING FORWARD.

5605
03:36:24,315 --> 03:36:27,352
I'LL START THERE.

5606
03:36:27,352 --> 03:36:29,320
I HAVE AUDITIONAL QUESTIONS BUT

5607
03:36:29,320 --> 03:36:34,792
WANT TO MAKE SURE EVERYBODY ELSE

5608
03:36:34,792 --> 03:36:35,560
HAS A CHANCE.

5609
03:36:35,560 --> 03:36:37,328
>> THANK YOU SO MUCH.

5610
03:36:37,328 --> 03:36:38,596
THANK YOU, DEBRA, ALSO.

5611
03:36:38,596 --> 03:36:41,332
THERE WERE A LOT OF QUESTIONS.

5612
03:36:41,332 --> 03:36:44,736
I HOPE WE GET TIME TO GO THROUGH

5613
03:36:44,736 --> 03:36:46,271
ALL THESE.

5614
03:36:46,271 --> 03:36:49,374
SO I'LL START WITH YOUR

5615
03:36:49,374 --> 03:36:50,575
QUESTION, NITA.

5616
03:36:50,575 --> 03:36:52,310
SO, FOR NOW NOTHING HAS BEEN

5617
03:36:52,310 --> 03:36:53,511
DONE.

5618
03:36:53,511 --> 03:36:54,712
THAT'S THE PROBLEM.

5619
03:36:54,712 --> 03:36:58,316
THE REASON PARTLY IS THAT -- SO

5620
03:36:58,316 --> 03:37:00,418
IN TERMS OF GETTING A

5621
03:37:00,418 --> 03:37:01,486
PROOF-OF-CONCEPT THAT THESE

5622
03:37:01,486 --> 03:37:03,388
THINGS CAN WORK, THAT YOU CAN

5623
03:37:03,388 --> 03:37:07,392
TAKE LARGE AMOUNTS OF DATA,

5624
03:37:07,392 --> 03:37:08,593
GENERATE GENERIC INFORMED MODEL

5625
03:37:08,593 --> 03:37:11,229
THAT BASICALLY IS ABLE TO

5626
03:37:11,229 --> 03:37:12,530
GENERATE SOMETHING LIKE EEG,

5627
03:37:12,530 --> 03:37:17,368
BECAUSE IT'S LEARNING TO

5628
03:37:17,368 --> 03:37:18,903
REGENERATE EEG, IT'S TRYING TO

5629
03:37:18,903 --> 03:37:20,838
CREATE THAT SO LEARNING

5630
03:37:20,838 --> 03:37:21,606
REPRESENTATIONS, WHAT DOES EEG

5631
03:37:21,606 --> 03:37:22,840
DATA LOOK LIKE?

5632
03:37:22,840 --> 03:37:26,444
JUST THE MERE FACT OF TRYING TO

5633
03:37:26,444 --> 03:37:27,745
GENERATE THAT AND THEN SAY CAN

5634
03:37:27,745 --> 03:37:29,280
THIS BE A STARTING POINT I CAN

5635
03:37:29,280 --> 03:37:33,017
FINE TUNE AND MAKE IT OTHER ON

5636
03:37:33,017 --> 03:37:34,452
OTHER CLASS

5637
03:37:34,452 --> 03:37:40,558
CLINICAL APPLICATIONS, PCI, --

5638
03:37:40,558 --> 03:37:41,993
BCI AND SO ON.

5639
03:37:41,993 --> 03:37:45,063
THERE'S INCREASING INTEREST TO

5640
03:37:45,063 --> 03:37:46,597
BUILD THESE MODELS, COMPARE

5641
03:37:46,597 --> 03:37:47,999
THEM, FIND FUNCTIONAL TRICKS TO

5642
03:37:47,999 --> 03:37:51,936
MAKE THEM MORE ROBUST, MORE

5643
03:37:51,936 --> 03:37:55,006
RELIABLE, AND IDEALLY MORE FAIR

5644
03:37:55,006 --> 03:37:57,742
DOWN THE LINE.

5645
03:37:57,742 --> 03:38:00,144
BUT IN GENERAL PEOPLE GO RIGHT

5646
03:38:00,144 --> 03:38:03,314
TO LARGER DATASETS THAT EXIST

5647
03:38:03,314 --> 03:38:10,855
AND THIS IS ONE OF THE LARGER

5648
03:38:10,855 --> 03:38:13,491
ONES THERE, WHICH COMES WITH

5649
03:38:13,491 --> 03:38:15,994
LIMITATIONS AND I'M 200% IN

5650
03:38:15,994 --> 03:38:18,429
AGREEMENT WITH EVERYTHING SAID

5651
03:38:18,429 --> 03:38:19,430
ABOUT LIMITATIONS THAT ACTUALLY

5652
03:38:19,430 --> 03:38:21,065
IN THIS DATASET BUT APPLIES TO

5653
03:38:21,065 --> 03:38:25,536
THE WHOLE FILED OF COLLECTED EEG

5654
03:38:25,536 --> 03:38:25,970
DATA.

5655
03:38:25,970 --> 03:38:31,242
WE'RE VERY FAR FROM HAVING

5656
03:38:31,242 --> 03:38:33,978
SUFFICIENT DATA COLLECTED ACROSS

5657
03:38:33,978 --> 03:38:36,714
DISEASES, ALSO ACROSS GEOGRAPHIC

5658
03:38:36,714 --> 03:38:37,582
GROUPS, UNDERREPRESENTED

5659
03:38:37,582 --> 03:38:39,250
MARGINALIZED GROUPS, SO THERE'S

5660
03:38:39,250 --> 03:38:43,287
A PROBLEM INHERENTLY WITH

5661
03:38:43,287 --> 03:38:44,288
EXISTING DATA.

5662
03:38:44,288 --> 03:38:46,791
SO WHETHER WE'RE TALKING ABOUT

5663
03:38:46,791 --> 03:38:48,993
JUST DOING REGULAR RESEARCH ON

5664
03:38:48,993 --> 03:38:51,963
EXISTING EEG DATA OR DEVELOPING

5665
03:38:51,963 --> 03:38:54,032
MODELS THAT'S INFORMED BY EEG

5666
03:38:54,032 --> 03:38:57,001
DATA THAT EXISTS THIS SAME ISSUE

5667
03:38:57,001 --> 03:38:59,070
STILL PERSISTS, IT'S NOT

5668
03:38:59,070 --> 03:39:00,405
SPECIFIC TO I THINK BUILDING

5669
03:39:00,405 --> 03:39:01,906
THESE MODELS BUT WE HAVE TO BE

5670
03:39:01,906 --> 03:39:05,977
AWARE OF IT BECAUSE IT CAN HAVE

5671
03:39:05,977 --> 03:39:06,611
DOWNSTREAM IMPLICATIONS.

5672
03:39:06,611 --> 03:39:08,212
WHEN WE CLAIM OR WHEN SOME

5673
03:39:08,212 --> 03:39:12,083
RESEARCHERS CLAIM NOW THEY HAVE

5674
03:39:12,083 --> 03:39:13,818
A VERY POWERFUL MODEL TRAINED ON

5675
03:39:13,818 --> 03:39:15,686
DATASETS, YOU CAN FINE TUNE AND

5676
03:39:15,686 --> 03:39:17,622
RUN AND APPLY TO MANY

5677
03:39:17,622 --> 03:39:19,857
APPLICATIONS, WE ALWAYS NEED TO

5678
03:39:19,857 --> 03:39:22,126
KNOW THAT DATA THAT WAS USED HAS

5679
03:39:22,126 --> 03:39:23,995
ITS OWN BIAS AND WE NEED TO TRY

5680
03:39:23,995 --> 03:39:24,695
AND MITIGATE.

5681
03:39:24,695 --> 03:39:27,498
ONE WAY OF DOING THAT IS

5682
03:39:27,498 --> 03:39:30,334
OBVIOUSLY TRY AND HAVE LESS --

5683
03:39:30,334 --> 03:39:31,869
MORE -- THAT DATA THAT IS USED

5684
03:39:31,869 --> 03:39:34,705
FOR TRAINING THAT WE CAN HAVE

5685
03:39:34,705 --> 03:39:36,908
ACCESS TO DATA THAT SOURCES FROM

5686
03:39:36,908 --> 03:39:41,712
THE BEGINNING, AT LEAST REDUCES

5687
03:39:41,712 --> 03:39:42,146
BIAS.

5688
03:39:42,146 --> 03:39:43,481
I WOULD SAY THAT'S THE PROBLEM

5689
03:39:43,481 --> 03:39:45,116
OF THE WHOLE FIELD OF

5690
03:39:45,116 --> 03:39:46,417
NEUROSCIENCE IN GENERAL WHICH

5691
03:39:46,417 --> 03:39:47,985
GOES BEYOND THESE MODELS.

5692
03:39:47,985 --> 03:39:52,123
NOT SURE THAT ANSWERS THE

5693
03:39:52,123 --> 03:39:52,390
QUESTION.

5694
03:39:52,390 --> 03:39:52,924
>> VERY HELPFUL.

5695
03:39:52,924 --> 03:39:56,527
IT WILL BECOME I THINK A BIG

5696
03:39:56,527 --> 03:39:58,362
CHALLENGE PARTICULARLY FOR EEG

5697
03:39:58,362 --> 03:40:00,131
DATA GIVEN IT'S OFTEN EXCLUDED

5698
03:40:00,131 --> 03:40:04,502
PEOPLE WITH DIFFERENT HAIR

5699
03:40:04,502 --> 03:40:07,472
TYPES, SO IT WILL BECOME MORE

5700
03:40:07,472 --> 03:40:09,207
PROBLEMATIC IN THIS SETTING THAN

5701
03:40:09,207 --> 03:40:09,774
OTHERS.

5702
03:40:09,774 --> 03:40:13,778
CAROLYN, LET ME TURN TO THE

5703
03:40:13,778 --> 03:40:15,213
CO-MODERATOR FIRST TO JUMP IN

5704
03:40:15,213 --> 03:40:16,647
WITH A QUESTION OR IF SHE'S

5705
03:40:16,647 --> 03:40:18,983
AHEAD WITH US TAKING QUESTIONS

5706
03:40:18,983 --> 03:40:20,151
FROM OTHERS.

5707
03:40:20,151 --> 03:40:21,452
>> I HAVE A QUESTION.

5708
03:40:21,452 --> 03:40:24,822
I COULDN'T HELP BUT NOTICE ON

5709
03:40:24,822 --> 03:40:30,228
YOUR RESULT SLIDE THERE WERE

5710
03:40:30,228 --> 03:40:31,629
NON-NEUROGPT METHODS THAT DID

5711
03:40:31,629 --> 03:40:35,466
REACH 60% ACCURACY IN PREDICTION

5712
03:40:35,466 --> 03:40:40,171
OF EEG DATA, WONDERING IF THEY

5713
03:40:40,171 --> 03:40:43,674
WERE TRADITIONAL APPROACHES THAT

5714
03:40:43,674 --> 03:40:45,109
DOESN'T REQUIRE THE CARBON

5715
03:40:45,109 --> 03:40:46,844
FOOTPRINT OF TRAINING LARGE

5716
03:40:46,844 --> 03:40:48,179
LANGUAGE MODELS.

5717
03:40:48,179 --> 03:40:51,115
DOES IT MAKE SENSE TO GO IN THE

5718
03:40:51,115 --> 03:40:53,184
DIRECTION NEEDING MORE DATA,

5719
03:40:53,184 --> 03:40:55,686
CREATING MORE BIAS, CONTINUING

5720
03:40:55,686 --> 03:40:56,354
EVENTUAL SUSTAINABILITY ISSUE,

5721
03:40:56,354 --> 03:40:57,555
USING LARGE LANGUAGE MODEL, OR

5722
03:40:57,555 --> 03:40:59,624
SHOULD WE TAKE A STEP BACK AND

5723
03:40:59,624 --> 03:41:04,328
SAY, YOU KNOW, GIVEN THE

5724
03:41:04,328 --> 03:41:05,863
HETEROGENEITY OF THIS EEG DATA,

5725
03:41:05,863 --> 03:41:14,939
YOU SAID 60% ACCURACY WITH

5726
03:41:14,939 --> 03:41:18,342
GPT-2, WHAT IS THE TRADEOFF,

5727
03:41:18,342 --> 03:41:20,545
WHEN DO YOU LOOK AT ALTERNATIVE

5728
03:41:20,545 --> 03:41:21,846
TECHNIQUES THAT DON'T REQUIRE

5729
03:41:21,846 --> 03:41:23,481
ALL THIS DATA?

5730
03:41:23,481 --> 03:41:24,682
>> GREAT QUESTION, GRACE.

5731
03:41:24,682 --> 03:41:27,084
I AGREE FULLY THAT IT HAS TO BE

5732
03:41:27,084 --> 03:41:29,620
WORTH IT AND WE NEED TO MAKE

5733
03:41:29,620 --> 03:41:31,789
SURE GAINING FROM IT MORE THAN

5734
03:41:31,789 --> 03:41:36,260
ANY RISKS THAT WE INCUR BY DOING

5735
03:41:36,260 --> 03:41:36,727
SO.

5736
03:41:36,727 --> 03:41:38,596
SAME RATIONALE APPLIES TO, FOR

5737
03:41:38,596 --> 03:41:39,897
EXAMPLE, DEVELOPING LARGE

5738
03:41:39,897 --> 03:41:40,865
LANGUAGE MODELS.

5739
03:41:40,865 --> 03:41:43,067
PEOPLE CONTINUE TO DEVELOP THESE

5740
03:41:43,067 --> 03:41:45,603
HUGE MODELS, IT'S COSTLY, HAS A

5741
03:41:45,603 --> 03:41:47,238
LARGE CARBON FOOTPRINT.

5742
03:41:47,238 --> 03:41:50,541
THERE'S INCREASING INTEREST IN

5743
03:41:50,541 --> 03:41:54,245
LOOKING AT SLMs RATHER THAN

5744
03:41:54,245 --> 03:41:55,646
LLMs, SMALL LANGUAGE MODELS,

5745
03:41:55,646 --> 03:41:57,748
BEING MORE EFFICIENT.

5746
03:41:57,748 --> 03:42:01,118
AT OUR RESEARCH CENTER AT UNIQUE

5747
03:42:01,118 --> 03:42:02,486
WE'RE PUSHING FOR EXPLORING

5748
03:42:02,486 --> 03:42:04,088
MAYBE LEARNING MORE HOW IS THE

5749
03:42:04,088 --> 03:42:04,855
BRAIN SO EFFICIENT, SOMETHING

5750
03:42:04,855 --> 03:42:07,258
THE BRAIN IS REALLY GOOD AT,

5751
03:42:07,258 --> 03:42:10,861
IT'S A COMPLEX NEURAL NETWORK,

5752
03:42:10,861 --> 03:42:12,830
ALSO BIOLOGICAL, BUT SUPER

5753
03:42:12,830 --> 03:42:13,097
EFFICIENT.

5754
03:42:13,097 --> 03:42:14,765
WE'RE STILL LACKING THAT

5755
03:42:14,765 --> 03:42:16,534
EFFICIENCY WHEN WE TALK ABOUT

5756
03:42:16,534 --> 03:42:17,368
ARTIFICIAL NEURAL NETWORKS.

5757
03:42:17,368 --> 03:42:19,704
THERE'S A LOT TO BE LEARNED FROM

5758
03:42:19,704 --> 03:42:26,844
THE BRAIN, BUT ALSO FOR SURE A

5759
03:42:26,844 --> 03:42:29,780
WHOLE SUB, OR NICHE IN THE

5760
03:42:29,780 --> 03:42:31,749
FIELD, LOOKING INTO OPTIMIZING,

5761
03:42:31,749 --> 03:42:32,883
INCREASING PERFORMANCE WITHOUT

5762
03:42:32,883 --> 03:42:34,952
INCREASING THE SIZE AND

5763
03:42:34,952 --> 03:42:35,553
COMPUTES.

5764
03:42:35,553 --> 03:42:37,154
ALSO BECAUSE EVEN IF IT'S

5765
03:42:37,154 --> 03:42:39,023
UNFORTUNATE, IF IT COMES FROM

5766
03:42:39,023 --> 03:42:39,757
INDUSTRIAL AND FROM COMPANIES

5767
03:42:39,757 --> 03:42:41,759
THAT ARE TRYING TO MAKE MONEY,

5768
03:42:41,759 --> 03:42:43,394
EVEN THAT'S GOING TO BE A FIELD

5769
03:42:43,394 --> 03:42:47,665
WHERE FOR THESE THINGS TO BE

5770
03:42:47,665 --> 03:42:49,834
FUNCTIONAL AND FOR THEM TO BE

5771
03:42:49,834 --> 03:42:51,902
DEPLOYED ON CELL PHONES AND

5772
03:42:51,902 --> 03:42:53,371
PEOPLE TO TRAIN MODELS LOCALLY,

5773
03:42:53,371 --> 03:42:56,874
IT'S GOING TO BE A PUSH TOWARDS

5774
03:42:56,874 --> 03:42:57,408
MORE EFFICIENT MODELS.

5775
03:42:57,408 --> 03:42:58,709
I THINK THAT'S GOING TO COME

5776
03:42:58,709 --> 03:43:00,378
DOWN THE LINE.

5777
03:43:00,378 --> 03:43:03,848
THIS IS WHY STUDIES LIKE OURS

5778
03:43:03,848 --> 03:43:05,616
AND A FEW OTHERS OUT THERE AND

5779
03:43:05,616 --> 03:43:08,119
MORE IN THE NEXT YEARS I THINK,

5780
03:43:08,119 --> 03:43:10,321
YES, THE CURRENT RESULTS IF THAT

5781
03:43:10,321 --> 03:43:11,956
WAS THE END GOAL, PROBABLY IT'S

5782
03:43:11,956 --> 03:43:13,491
NOT WORTH IT.

5783
03:43:13,491 --> 03:43:14,925
INCREASE IN TWO, THREE, FOUR

5784
03:43:14,925 --> 03:43:15,993
PERCENT, I DON'T THINK IT'S

5785
03:43:15,993 --> 03:43:16,661
WORTH IT.

5786
03:43:16,661 --> 03:43:17,995
THE IDEA IS THAT WE'RE FIGURING

5787
03:43:17,995 --> 03:43:24,001
OUT HOW TO DO THIS IN THE BEST

5788
03:43:24,001 --> 03:43:27,605
WAY POSSIBLE SO THAT -- IN A FEW

5789
03:43:27,605 --> 03:43:31,409
I DON'T KNOW, WEEKS, MONTHS,

5790
03:43:31,409 --> 03:43:33,311
A.I. GOES SO QUICKLY WE'LL HAVE

5791
03:43:33,311 --> 03:43:34,512
SOMETHING THAT ACHIEVES

5792
03:43:34,512 --> 03:43:38,115
SUBSTANTIAL IMPROVEMENT THAT

5793
03:43:38,115 --> 03:43:39,650
ULTIMATELY BENEFITS THE HUMANITY

5794
03:43:39,650 --> 03:43:41,185
AND ALSO CLINICAL APPLICATIONS

5795
03:43:41,185 --> 03:43:49,393
AS WELL AS BASIC RESEARCH.

5796
03:43:49,393 --> 03:43:54,865
>> MAYBE THERE WILL BE
MORE WAYS

5797
03:43:54,865 --> 03:43:56,300
TO BOOTSTRAP ETHICS.

5798
03:43:56,300 --> 03:43:57,835
>> I SEE CAROLYN WAITED

5799
03:43:57,835 --> 03:44:01,439
PATIENTLY SO LET ME TURN TO HER.

5800
03:44:01,439 --> 03:44:03,407
>> THANK YOU, NITA.

5801
03:44:03,407 --> 03:44:06,811
GOOD TO SEE  YOU HERE TODAY.

5802
03:44:06,811 --> 03:44:08,446
I HAVE QUESTIONS ABOUT

5803
03:44:08,446 --> 03:44:09,647
ALGORITHMIC BIAS, WHICH IS A

5804
03:44:09,647 --> 03:44:11,549
TOPIC THAT WAS BROUGHT UP BY

5805
03:44:11,549 --> 03:44:12,583
SEVERAL SPEAKERS THROUGHOUT THE

5806
03:44:12,583 --> 03:44:13,184
DAY.

5807
03:44:13,184 --> 03:44:16,754
ONE OF THEM IS RELATED TO NITA'S

5808
03:44:16,754 --> 03:44:20,458
QUESTION, AND ALSO EXTENDS FROM

5809
03:44:20,458 --> 03:44:21,659
CHRISTINE'S EARLIER PROMPT ABOUT

5810
03:44:21,659 --> 03:44:22,993
FOCUSING ON ASPECTS THAT MIGHT

5811
03:44:22,993 --> 03:44:25,930
BE VERY SORT OF UNIQUE OR

5812
03:44:25,930 --> 03:44:27,131
SPECIFIC TO NEUROSCIENCE.

5813
03:44:27,131 --> 03:44:30,634
SO MY FIRST QUESTION IS ARE

5814
03:44:30,634 --> 03:44:31,936
THERE ASPECTS OF ALGORITHMIC

5815
03:44:31,936 --> 03:44:34,705
BIAS IN AN A.I. MODEL THAT ARE

5816
03:44:34,705 --> 03:44:36,440
UNIQUE TO BRAIN DATA?

5817
03:44:36,440 --> 03:44:39,377
WE TALKED ABOUT SOME

5818
03:44:39,377 --> 03:44:40,911
EEG-SPECIFIC AREAS BUT THINKING

5819
03:44:40,911 --> 03:44:43,748
MORE BROADLY THAN EEG ARE THERE

5820
03:44:43,748 --> 03:44:46,183
AREAS THAT ARE UNIQUE TO BRAIN

5821
03:44:46,183 --> 03:44:48,786
DATA BEING FED INTO A MODEL AS

5822
03:44:48,786 --> 03:44:51,188
COMPARED TO OTHER KINDS OF HUMAN

5823
03:44:51,188 --> 03:44:51,555
BIOLOGICAL DATA?

5824
03:44:51,555 --> 03:44:54,358
I THINK IT WAS INTERESTING TO

5825
03:44:54,358 --> 03:44:55,926
HEAR DEBRA MENTION NEUROTYPICAL

5826
03:44:55,926 --> 03:44:59,196
SIGNATURES, I DON'T KNOW, MAYBE

5827
03:44:59,196 --> 03:45:00,398
SPATIOTEMPORAL NATURE OF BRAIN

5828
03:45:00,398 --> 03:45:05,002
DATA BUT CURIOUS ON TO HEAR

5829
03:45:05,002 --> 03:45:08,506
SPEAKER THOUGHTS ON THAT.

5830
03:45:08,506 --> 03:45:09,473
SECOND QUESTION, RESEARCH

5831
03:45:09,473 --> 03:45:11,675
APPROACHES TO CLEANING UP OR

5832
03:45:11,675 --> 03:45:13,744
MITIGATING UNWANTED BIAS IN A

5833
03:45:13,744 --> 03:45:17,014
TRAINING DATASET, PERHAPS

5834
03:45:17,014 --> 03:45:17,815
INSERTING ETHICAL CONSIDERATION

5835
03:45:17,815 --> 03:45:18,983
AND ADJUSTMENT FROM THE

5836
03:45:18,983 --> 03:45:20,985
PIPELINE, I THINK THE SECOND

5837
03:45:20,985 --> 03:45:22,953
QUESTION MAYBE RELATES MORE TO

5838
03:45:22,953 --> 03:45:23,954
BIASES THAT ARE CHARACTERISTIC

5839
03:45:23,954 --> 03:45:26,791
IN A GROUP DATASET AS OPPOSED TO

5840
03:45:26,791 --> 03:45:29,760
INDIVIDUAL DATASETS, THINGS LIKE

5841
03:45:29,760 --> 03:45:30,394
BALANCING DEMOGRAPHIC

5842
03:45:30,394 --> 03:45:32,163
REPRESENTATION IN A TRAINING SET

5843
03:45:32,163 --> 03:45:34,565
OR I THINK ANOTHER SPEAKER

5844
03:45:34,565 --> 03:45:37,635
MENTIONED THE DOMINANT VERSUS

5845
03:45:37,635 --> 03:45:42,406
RARE NEUROLOGICAL QUESTIONS.

5846
03:45:42,406 --> 03:45:44,074
QUESTIONS ABOUT UNIQUENESS AND

5847
03:45:44,074 --> 03:45:45,676
SECOND QUESTION ABOUT MITIGATION

5848
03:45:45,676 --> 03:45:50,281
PERHAPS BY TARGETING DATA

5849
03:45:50,281 --> 03:45:50,948
PREPARATION PIPELINE.

5850
03:45:50,948 --> 03:45:55,753
>> THANK YOU FOR THAT
QUESTION.

5851
03:45:55,753 --> 03:45:57,922
SO, I AGREE THESE ARE IMPORTANT

5852
03:45:57,922 --> 03:45:58,823
TOPICS.

5853
03:45:58,823 --> 03:46:04,061
FOR THE FIRST QUESTION, WHICH IS

5854
03:46:04,061 --> 03:46:08,332
MITIGATION, SECOND WAS THE

5855
03:46:08,332 --> 03:46:09,433
UNIQUENESS, YES.

5856
03:46:09,433 --> 03:46:10,301
SO, THERE'S AN OVERLAP.

5857
03:46:10,301 --> 03:46:12,069
THERE ARE A FEW THINGS THAT ARE

5858
03:46:12,069 --> 03:46:13,704
UNIQUE, THAT ARE SPECIFIC TO

5859
03:46:13,704 --> 03:46:15,906
THIS TYPE OF DATA, NEUROSCIENCE

5860
03:46:15,906 --> 03:46:18,309
DATA OR BRAIN DATA, I WOULD SAY

5861
03:46:18,309 --> 03:46:20,277
FOR EXAMPLE ACCESS TO DATA

5862
03:46:20,277 --> 03:46:20,578
COLLECTION.

5863
03:46:20,578 --> 03:46:23,314
SOME COUNTRIES WHERE YOU DON'T

5864
03:46:23,314 --> 03:46:25,749
HAVE fMRI MACHINE, YOU WON'T

5865
03:46:25,749 --> 03:46:27,284
HAVE fMRI DATA COLLECTED IN

5866
03:46:27,284 --> 03:46:29,153
SOME CASES ACROSS THE WORLD, AT

5867
03:46:29,153 --> 03:46:33,324
LEAST NOT AS OFTEN AS ELSEWHERE.

5868
03:46:33,324 --> 03:46:34,291
GEOGRAPHICALLY ACROSS THE PLANET

5869
03:46:34,291 --> 03:46:40,297
LOOKING AT WHAT BRAIN DATA WE

5870
03:46:40,297 --> 03:46:42,366
HAVE, IT WON'T BE THE SAME

5871
03:46:42,366 --> 03:46:43,234
COVERAGE EVERYWHERE.

5872
03:46:43,234 --> 03:46:45,636
IF WE'RE RELYING ON EXISTING MRI

5873
03:46:45,636 --> 03:46:48,606
DATASETS ACROSS THE PLANET TO

5874
03:46:48,606 --> 03:46:50,674
DERIVE A MODEL OF MRI DATA THAT

5875
03:46:50,674 --> 03:46:54,745
WE'RE USING AS A PRE-TRAINING

5876
03:46:54,745 --> 03:46:57,481
MODEL THAT'S GOING TO COME FROM

5877
03:46:57,481 --> 03:47:01,418
THERE, OTHER AP MILES  -- APPLIS

5878
03:47:01,418 --> 03:47:07,458
TO OTHER TIMES OF DATA, RICH

5879
03:47:07,458 --> 03:47:07,958
COUNTRIES.

5880
03:47:07,958 --> 03:47:09,593
SO, YEAH, MAYBE I'LL ROLL BACK A

5881
03:47:09,593 --> 03:47:09,994
BIT.

5882
03:47:09,994 --> 03:47:13,497
IT'S NOT SO MUCH IT'S SO UNIQUE

5883
03:47:13,497 --> 03:47:16,033
I WOULD SAY ABOUT THE BIAS THAT

5884
03:47:16,033 --> 03:47:17,968
COMES INTO THE BRAIN SCIENCE BUT

5885
03:47:17,968 --> 03:47:22,339
WHAT MIGHT BE UNIQUE WHEN WE'RE

5886
03:47:22,339 --> 03:47:24,542
TRYING TO BUILD MODELS THAT

5887
03:47:24,542 --> 03:47:26,944
WOULD SERVE AS STARTING POINT,

5888
03:47:26,944 --> 03:47:28,479
DIFFICULT TO KNOW WHAT WENT INTO

5889
03:47:28,479 --> 03:47:30,447
THE MODELS BECAUSE IT MIGHT BE

5890
03:47:30,447 --> 03:47:31,749
USEFUL FOR SPECIFIC DOWNSTREAM

5891
03:47:31,749 --> 03:47:33,517
APPLICATIONS BUT NOT AT ALL

5892
03:47:33,517 --> 03:47:35,719
USEFUL AND ACTUALLY DANGEROUS

5893
03:47:35,719 --> 03:47:38,355
FOR OTHER DOWNSTREAM

5894
03:47:38,355 --> 03:47:39,256
APPLICATIONS, THAT'S SOMETHING

5895
03:47:39,256 --> 03:47:40,991
WHERE THERE'S SOME KNOWLEDGE

5896
03:47:40,991 --> 03:47:43,494
ABOUT NEUROSCIENCE OR CLINICAL

5897
03:47:43,494 --> 03:47:44,595
APPLICATIONS THAT GETS LOST,

5898
03:47:44,595 --> 03:47:46,897
MAYBE WHEN THESE MODELS ARE

5899
03:47:46,897 --> 03:47:47,765
BEING BUILT.

5900
03:47:47,765 --> 03:47:49,099
AND TO YOUR SECOND QUESTION OF

5901
03:47:49,099 --> 03:47:52,870
WHETHER THERE ARE WAYS OF

5902
03:47:52,870 --> 03:47:54,271
MITIGATING, IN PRE-PROCESSING, I

5903
03:47:54,271 --> 03:47:59,810
THINK SO, APPLIES TO IN GENERAL

5904
03:47:59,810 --> 03:48:00,911
PROCESSING EEG OR NEUROSCIENCE

5905
03:48:00,911 --> 03:48:01,345
DATA IN GENERAL.

5906
03:48:01,345 --> 03:48:04,415
A LOT OF WORK NEEDS TO BE DONE,

5907
03:48:04,415 --> 03:48:06,050
IN PRE-PROCESSING, IDEALLY FROM

5908
03:48:06,050 --> 03:48:07,484
THE DATA COLLECTION IN GENERAL.

5909
03:48:07,484 --> 03:48:08,285
THAT'S SORT OF LIKE WHERE I

5910
03:48:08,285 --> 03:48:12,756
THINK MOST OF THE PROBLEM COMES.

5911
03:48:12,756 --> 03:48:14,058
AND SO DESIGN OF STUDIES,

5912
03:48:14,058 --> 03:48:21,498
EXPERIMENTS.

5913
03:48:21,498 --> 03:48:24,335
>> YES, I'LL ADD MY
THOUGHTS AS

5914
03:48:24,335 --> 03:48:24,568
WELL.

5915
03:48:24,568 --> 03:48:29,707
ON THE QUESTION OF UNIQUENESS, I

5916
03:48:29,707 --> 03:48:31,775
DO -- I DON'T THINK THIS IS --

5917
03:48:31,775 --> 03:48:32,876
THIS ISN'T BRAND NEW.

5918
03:48:32,876 --> 03:48:36,580
THERE ARE AS AMY MENTIONED,

5919
03:48:36,580 --> 03:48:38,682
THERE ARE ANALOGIES IN OTHER

5920
03:48:38,682 --> 03:48:41,619
SPACES, DNA IN PARTICULAR,

5921
03:48:41,619 --> 03:48:44,121
RIGHT, BOTH GENETICS AND OUR

5922
03:48:44,121 --> 03:48:47,891
BRAINS WE FEEL DETERMINISTIC

5923
03:48:47,891 --> 03:48:48,092
ABOUT.

5924
03:48:48,092 --> 03:48:49,727
AND FEEL LIKE THEY ARE SPECIAL

5925
03:48:49,727 --> 03:48:52,830
IN WAYS THAT OTHER ORGANS OR

5926
03:48:52,830 --> 03:48:55,566
OTHER PARTS OF OUR BODIES ARE

5927
03:48:55,566 --> 03:49:01,372
NOT.

5928
03:49:01,372 --> 03:49:05,409
AND THAT SORT OF EMPHASIS ON THE

5929
03:49:05,409 --> 03:49:07,044
TWO KINDS OF DATA IS DIFFERENT

5930
03:49:07,044 --> 03:49:11,215
THAN OTHER KINDS OF DATA.

5931
03:49:11,215 --> 03:49:12,950
THE CHANGING OVER TIME IS ONE

5932
03:49:12,950 --> 03:49:15,586
WAY IN WHICH BRAIN DATA ARE

5933
03:49:15,586 --> 03:49:19,723
QUITE DIFFERENT THAN DNA, RIGHT?

5934
03:49:19,723 --> 03:49:23,227
ON THE QUESTION OF MITIGATION,

5935
03:49:23,227 --> 03:49:25,229
AND THE DATA PREPARATION

5936
03:49:25,229 --> 03:49:29,600
PIPELINE, OF COURSE HUGE

5937
03:49:29,600 --> 03:49:31,468
QUESTIONS ABOUT DATA COLLECTION,

5938
03:49:31,468 --> 03:49:35,606
FROM INDIVIDUALS, RIGHT, WHO ARE

5939
03:49:35,606 --> 03:49:36,807
DIVERSE ACROSS MULTIPLE SPECTRA,

5940
03:49:36,807 --> 03:49:39,476
BUT ALSO THE QUESTION OF WHICH

5941
03:49:39,476 --> 03:49:43,480
DATA ARE WE COLLECTING, RIGHT?

5942
03:49:43,480 --> 03:49:44,848
THIS IS THE GENETICS COMMUNITY

5943
03:49:44,848 --> 03:49:49,420
IS FACING THIS RIGHT NOW, RIGHT,

5944
03:49:49,420 --> 03:49:51,622
WITH THE NATIONAL ACADEMY'S

5945
03:49:51,622 --> 03:49:56,794
REPORT ON RACE, ETHNICITY, AND

5946
03:49:56,794 --> 03:49:58,429
POPULATION, DESIGNATORS USED IN

5947
03:49:58,429 --> 03:49:59,863
THE DEVELOPMENT OF GENETIC

5948
03:49:59,863 --> 03:50:00,831
RESEARCH, I THINK THERE'S

5949
03:50:00,831 --> 03:50:04,234
ANALOGY HERE 

5950
03:50:04,234 --> 03:50:07,271
ANALOGY HERE TO WHAT OTHER KINDS

5951
03:50:07,271 --> 03:50:10,541
OF DATA THAT I ALLUDED TO IN MY

5952
03:50:10,541 --> 03:50:11,675
TALK, NECESSARY TO UNDERSTAND

5953
03:50:11,675 --> 03:50:13,310
WHAT IS GOING WITH THE BRAIN

5954
03:50:13,310 --> 03:50:14,178
DATA, RIGHT?

5955
03:50:14,178 --> 03:50:15,713
YOU CAN'T ISOLATE THE DATA FROM

5956
03:50:15,713 --> 03:50:16,914
EVERYTHING ELSE.

5957
03:50:16,914 --> 03:50:20,217
YOU NEED TO UNDERSTAND THE

5958
03:50:20,217 --> 03:50:21,418
ENVIRONMENTAL FACTORS, ET

5959
03:50:21,418 --> 03:50:22,720
CETERA, THAT ARE SHAPING WHAT'S

5960
03:50:22,720 --> 03:50:27,958
HAPPENING IN THE BRAIN.

5961
03:50:27,958 --> 03:50:34,531
YEAH, I THINK I'LL STOP THERE.

5962
03:50:34,531 --> 03:50:35,332
>> THANK YOU.

5963
03:50:35,332 --> 03:50:36,567
>> THANKS, CAROLYN.

5964
03:50:36,567 --> 03:50:38,702
FEEL FREE TO RAISE HANDS.

5965
03:50:38,702 --> 03:50:40,003
OTHERWISE WE HAVE LOTS OF

5966
03:50:40,003 --> 03:50:41,405
QUESTIONS AND YOU'LL END UP

5967
03:50:41,405 --> 03:50:46,777
HEARING GRACE AND MY QUESTIONS

5968
03:50:46,777 --> 03:50:48,846
ALL DAY LONG HERE.

5969
03:50:48,846 --> 03:50:51,482
A SECOND QUESTION FOR DR. JERBI,

5970
03:50:51,482 --> 03:50:52,816
BRIDGING BETWEEN THE FIRST CASE

5971
03:50:52,816 --> 03:50:54,218
STUDY AND THIS ONE TO SEE IF

5972
03:50:54,218 --> 03:50:56,186
THERE'S A DIFFERENCE BETWEEN

5973
03:50:56,186 --> 03:50:57,821
APPLYING, YOU KNOW, GENERAL

5974
03:50:57,821 --> 03:50:59,690
FOUNDATIONAL MODEL TO QUESTIONS

5975
03:50:59,690 --> 03:51:02,526
OF NEUROSCIENCE AND DIGITAL

5976
03:51:02,526 --> 03:51:03,427
PHENOTYPING VERSUS CREATING

5977
03:51:03,427 --> 03:51:05,829
CUSTOM MODEL THAT'S TRAINED ON

5978
03:51:05,829 --> 03:51:09,233
EEG OR OTHER BRAIN DATA AND

5979
03:51:09,233 --> 03:51:09,833
SIMILARLY TO MAKE PREDICTIONS

5980
03:51:09,833 --> 03:51:11,301
ABOUT THE FUTURE.

5981
03:51:11,301 --> 03:51:14,138
YOU MENTIONED IT'S PREDICTING

5982
03:51:14,138 --> 03:51:16,173
THE NEXT TOKEN, SO TRYING TO

5983
03:51:16,173 --> 03:51:18,342
IMAGINE PREDICTING THE NEXT

5984
03:51:18,342 --> 03:51:20,377
TOKEN IN EEG SEEMS LIKE YOU

5985
03:51:20,377 --> 03:51:22,813
COULD ACHIEVE THE SAME THING

5986
03:51:22,813 --> 03:51:24,915
FROM THE DIGITAL PHENOTYPING, IF

5987
03:51:24,915 --> 03:51:26,216
NOT MORE PRECISELY OR

5988
03:51:26,216 --> 03:51:28,552
DIFFERENTLY BY BEING ABLE TO SEE

5989
03:51:28,552 --> 03:51:31,155
PATTERNS OVER TIME THROUGH

5990
03:51:31,155 --> 03:51:33,690
LARGER DATASETS OF WHAT COMES

5991
03:51:33,690 --> 03:51:36,426
NEXT, NEURODEGENERATION OR IF

5992
03:51:36,426 --> 03:51:37,961
THAT IS MENTAL ILLNESSES OR

5993
03:51:37,961 --> 03:51:40,264
OTHER TIMES OF PATTERNS THAT

5994
03:51:40,264 --> 03:51:42,833
COULD BE DETECTED FROM RECORDING

5995
03:51:42,833 --> 03:51:49,540
OF EEG THAT WOULD BE UPLOADED

5996
03:51:49,540 --> 03:51:50,607
SAYING WHAT COMES NEXT.

5997
03:51:50,607 --> 03:51:53,210
IS THAT RIGHT THAT EVENTUALLY

5998
03:51:53,210 --> 03:51:55,746
THESE MODELS COULD ACHIEVE THAT

5999
03:51:55,746 --> 03:51:57,181
SAME CLINICAL SIGNIFICANCE?

6000
03:51:57,181 --> 03:52:00,117
AND LET ME START THERE, WHETHER

6001
03:52:00,117 --> 03:52:01,418
OR NOT THAT'S ACCURATE, YOU

6002
03:52:01,418 --> 03:52:03,387
COULD DO THE SAME THING AND

6003
03:52:03,387 --> 03:52:04,922
ACHIEVE IT BY PREDICTING WHAT

6004
03:52:04,922 --> 03:52:07,257
COMES NEXT IN THE EEG DATA

6005
03:52:07,257 --> 03:52:10,093
POTENTIALLY IN WAYS YOU WOULDN'T

6006
03:52:10,093 --> 03:52:11,495
DO WITH DIGITAL PHENOTYPING

6007
03:52:11,495 --> 03:52:17,601
BECAUSE IT'S MORE FORECASTING IN

6008
03:52:17,601 --> 03:52:19,036
THE FUTURE, MORE ACCURATELY THAN

6009
03:52:19,036 --> 03:52:19,703
THE FIRST CASE STUDY.

6010
03:52:19,703 --> 03:52:21,772
>> THANK YOU FOR THE
QUESTION.

6011
03:52:21,772 --> 03:52:23,507
THIS ALLOWS ME TO CLARIFY ONE

6012
03:52:23,507 --> 03:52:25,576
THING ABOUT THE NEXT TOKEN

6013
03:52:25,576 --> 03:52:27,878
PREDICTION OR NEXT WORD

6014
03:52:27,878 --> 03:52:30,714
PREDICTION, LIKE LANGUAGE MODEL,

6015
03:52:30,714 --> 03:52:31,181
FOR INSTANCE.

6016
03:52:31,181 --> 03:52:32,482
TO THE QUESTION WHETHER ON

6017
03:52:32,482 --> 03:52:36,753
LARGER SCALE WE CAN USE THESE

6018
03:52:36,753 --> 03:52:40,157
MODELS TO EXPLORE CONVERSION IN

6019
03:52:40,157 --> 03:52:40,757
NEURODEGENERATIVE DISEASES, I

6020
03:52:40,757 --> 03:52:44,194
THINK THIS IS PART OF THE

6021
03:52:44,194 --> 03:52:47,965
APPLICATION THAT CAN BE

6022
03:52:47,965 --> 03:52:50,701
DOWNSTREAM TASK BUT THE NEXT

6023
03:52:50,701 --> 03:52:51,668
PART, CHUNKING DATA, THAT'S USED

6024
03:52:51,668 --> 03:52:55,172
IN THE FIRST PART OF THE

6025
03:52:55,172 --> 03:52:56,940
PRE-TRAINING, TO TRAIN A MODEL

6026
03:52:56,940 --> 03:52:59,009
TO LEARN REPRESENTATIONS OF WHAT

6027
03:52:59,009 --> 03:53:01,078
EEG DATA IS LIKE YOU REMOVE

6028
03:53:01,078 --> 03:53:03,614
PARTS OF THE DATA AND YOU TRY TO

6029
03:53:03,614 --> 03:53:05,015
LET IT, ITSELF, PREDICT WHAT

6030
03:53:05,015 --> 03:53:06,550
WOULD BE THE CONTINUATION OF

6031
03:53:06,550 --> 03:53:09,052
LET'S SAY EEG SENTENCE IN A WAY,

6032
03:53:09,052 --> 03:53:11,688
PREDICT THE NEXT WORK OF EEG.

6033
03:53:11,688 --> 03:53:13,023
BUT THE MORE IT BECOMES CAPABLE

6034
03:53:13,023 --> 03:53:15,292
OF DOING THAT IN THAT PROCESS

6035
03:53:15,292 --> 03:53:16,627
IT'S NOW LEARNING REPRESENTATION

6036
03:53:16,627 --> 03:53:17,594
OF EEG DATA.

6037
03:53:17,594 --> 03:53:20,130
NOW YOU HAVE A MODEL THAT'S

6038
03:53:20,130 --> 03:53:21,331
PRE-TRAINED BUT NOT PREDICTING

6039
03:53:21,331 --> 03:53:24,167
WHAT'S GOING TO HAPPEN TO THE

6040
03:53:24,167 --> 03:53:26,603
INDIVIDUAL IN THAT DATASET.

6041
03:53:26,603 --> 03:53:33,911
IT'S JUST THAT HAS LEARNED BY --

6042
03:53:33,911 --> 03:53:37,314
AS IF YOU'RE LEARNING A GAME AND

6043
03:53:37,314 --> 03:53:38,949
SOME COMPONENTS ARE REMOVED AND

6044
03:53:38,949 --> 03:53:40,150
YOU'RE TRYING TO GUESS MISSING

6045
03:53:40,150 --> 03:53:43,887
PARTS AND END UP WITH A HOLISTIC

6046
03:53:43,887 --> 03:53:44,121
VIEW.

6047
03:53:44,121 --> 03:53:45,622
>> I UNDERSTOOD THAT FOR
THE

6048
03:53:45,622 --> 03:53:46,023
TRAINING.

6049
03:53:46,023 --> 03:53:48,825
I WAS ASKING THEN FOR THE

6050
03:53:48,825 --> 03:53:54,731
APPLICATION PART, ONCE IT HAS

6051
03:53:54,731 --> 03:53:56,500
LEARNED THAT, PRESUMABLY COULD

6052
03:53:56,500 --> 03:53:58,468
TAKE AN INDIVIDUAL NEW, FEED IT

6053
03:53:58,468 --> 03:54:03,840
IN AN EEG REPORTING FROM

6054
03:54:03,840 --> 03:54:04,308
SOMEONE, PARTICULARLY

6055
03:54:04,308 --> 03:54:05,242
LONGITUDINAL DATA, TO SAY, YOU

6056
03:54:05,242 --> 03:54:07,511
KNOW, BOTH PICKING UP SUBTLETIES

6057
03:54:07,511 --> 03:54:08,512
WE COULDN'T NECESSARILY PICK UP

6058
03:54:08,512 --> 03:54:11,815
BY HAVING BEEN TRAINED ON HUGE

6059
03:54:11,815 --> 03:54:13,250
DATASETS ABOUT THE CURRENT

6060
03:54:13,250 --> 03:54:14,685
INFORMATION THAT IS BEING

6061
03:54:14,685 --> 03:54:17,621
INFORMED BY THAT EEG BUT ALSO

6062
03:54:17,621 --> 03:54:21,224
BEING ABLE TO FORECAST BY HAVING

6063
03:54:21,224 --> 03:54:23,393
SEEN IN, YOU KNOW, IF IT'S BEEN

6064
03:54:23,393 --> 03:54:25,162
TRAINED ON LONGITUDINAL DATA,

6065
03:54:25,162 --> 03:54:29,433
WHAT COMES NEXT, RIGHT, IN

6066
03:54:29,433 --> 03:54:32,169
PREDICTING AND FORECASTING

6067
03:54:32,169 --> 03:54:33,904
NEURODEGENERATION OR ONSET OF

6068
03:54:33,904 --> 03:54:35,539
MENTAL ILLNESS OR OTHER

6069
03:54:35,539 --> 03:54:36,540
FEATURES.

6070
03:54:36,540 --> 03:54:37,207
>> IT'S POSSIBLE AND
EXCITING

6071
03:54:37,207 --> 03:54:38,308
VENUE FOR THE FUTURE.

6072
03:54:38,308 --> 03:54:39,943
TIME SCALE AT THE MOMENT DOING

6073
03:54:39,943 --> 03:54:41,144
NEXT PREDICTION ARE NOT THE

6074
03:54:41,144 --> 03:54:41,345
SAME.

6075
03:54:41,345 --> 03:54:42,779
WE'RE NOT AT TIME SCALES OF

6076
03:54:42,779 --> 03:54:43,213
YEARS.

6077
03:54:43,213 --> 03:54:44,748
THEIR ARE TIME SCALES OF SECONDS

6078
03:54:44,748 --> 03:54:45,615
OF DATA.

6079
03:54:45,615 --> 03:54:47,384
BUT ABSOLUTELY.

6080
03:54:47,384 --> 03:54:50,120
THAT WOULD BE SOMETHING THAT MAY

6081
03:54:50,120 --> 03:54:51,755
BE A SOURCE, IDEA FOR RESEARCH

6082
03:54:51,755 --> 03:54:53,824
TRYING TO USE LONGITUDINAL DATA,

6083
03:54:53,824 --> 03:54:56,126
TRYING TO DEVELOP THE SAME

6084
03:54:56,126 --> 03:54:57,561
ALGORITHMS DOING PREDICTIONS BUT

6085
03:54:57,561 --> 03:54:59,529
ON WAY LARGER TIME SCALES, THAT

6086
03:54:59,529 --> 03:55:04,101
WOULD INVOLVE A WHOLE OTHER

6087
03:55:04,101 --> 03:55:10,240
SCHEME FOR TOKENIZATION, MIGHT

6088
03:55:10,240 --> 03:55:11,441
BE INTERESTING TO PURSUE.

6089
03:55:11,441 --> 03:55:12,776
>> A FOLLOW-UP QUESTION.

6090
03:55:12,776 --> 03:55:13,143
>> GO AHEAD.

6091
03:55:13,143 --> 03:55:14,711
>> ABOUT THE PARTICULAR
DATA.

6092
03:55:14,711 --> 03:55:19,016
AS I MENTIONED, THE DATA ARE

6093
03:55:19,016 --> 03:55:20,984
PRIMARILY IN EPILEPSY DATASET,

6094
03:55:20,984 --> 03:55:25,455
SO YOU ALREADY HAVE DATA THAT

6095
03:55:25,455 --> 03:55:27,891
PRESUMABLY HAVE MEDICAL EVENTS

6096
03:55:27,891 --> 03:55:29,192
IN THEM.

6097
03:55:29,192 --> 03:55:31,028
ARE YOU ABLE TO PREDICT THOSE

6098
03:55:31,028 --> 03:55:32,129
MEDICAL EVENTS GIVEN THESE ARE

6099
03:55:32,129 --> 03:55:34,698
THE DATA YOU'VE ALREADY TRAINED

6100
03:55:34,698 --> 03:55:37,167
ON? 

6101
03:55:37,167 --> 03:55:38,702
>> UM, SO, WE DIDN'T TRY
THAT

6102
03:55:38,702 --> 03:55:40,670
BUT THE POINT HERE IS THAT SO

6103
03:55:40,670 --> 03:55:42,205
THE LENGTH OF THE DATA, THEY

6104
03:55:42,205 --> 03:55:43,173
WERE REALLY SHORT.

6105
03:55:43,173 --> 03:55:43,707
>> RIGHT.

6106
03:55:43,707 --> 03:55:49,546
>> YOU HAVE TWO SECONDS OF
DATA,

6107
03:55:49,546 --> 03:55:50,180
MULTIPLE CHUNKS.

6108
03:55:50,180 --> 03:55:53,583
SO -- AND THE FACT THAT IT HELPS

6109
03:55:53,583 --> 03:55:56,186
IMPROVE, NOT TO A LARGE EXTENT,

6110
03:55:56,186 --> 03:55:57,387
MODELS COME PRIMARILY FROM

6111
03:55:57,387 --> 03:55:59,156
CLINICAL CONDITIONS, THEY DID

6112
03:55:59,156 --> 03:56:01,324
HELP IMPROVE ON THE

6113
03:56:01,324 --> 03:56:02,859
CLASSIFICATION TASKS UNRELATED

6114
03:56:02,859 --> 03:56:04,728
TO ANY CLINICAL CONDITIONS,

6115
03:56:04,728 --> 03:56:05,796
ACTUALLY SHOWING WE'RE CAPTURING

6116
03:56:05,796 --> 03:56:07,030
SOMETHING THAT REPRESENTS

6117
03:56:07,030 --> 03:56:08,565
SOMETHING ABOUT THE INHERENT

6118
03:56:08,565 --> 03:56:12,702
LATENT REPRESENTATION OF EEG

6119
03:56:12,702 --> 03:56:13,036
DATA.

6120
03:56:13,036 --> 03:56:15,238
>> COULD YOU CLARIFY THAT
POINT,

6121
03:56:15,238 --> 03:56:16,540
SINCE THE QUESTION ON CHUNKING

6122
03:56:16,540 --> 03:56:20,710
DID COME UP IN DEBRA'S TALK?

6123
03:56:20,710 --> 03:56:25,415
DID YOU EXHAUSTIVELY RUN YOUR

6124
03:56:25,415 --> 03:56:27,150
ANALYSIS OF CHUNKS AND DIDN'T

6125
03:56:27,150 --> 03:56:28,151
MATTER HOW MUCH OVERLAP THERE

6126
03:56:28,151 --> 03:56:30,220
WAS IN TERMS OF ACCURACY?

6127
03:56:30,220 --> 03:56:34,291
>> IF I REMEMBER CORRECT 
CORRE,

6128
03:56:34,291 --> 03:56:35,792
MIGHT HAVE BEEN MENTIONED, MAYBE

6129
03:56:35,792 --> 03:56:39,663
NOT. 

6130
03:56:39,663 --> 03:56:41,631
WE TRIED, MAYBE SIX SECONDS OF

6131
03:56:41,631 --> 03:56:43,800
DATA WITH DIFFERENT DEGREES OF

6132
03:56:43,800 --> 03:56:45,435
OVERLAP, ONE USED FOR THE

6133
03:56:45,435 --> 03:56:47,737
REPRESENTATION WAS THE ONE THAT

6134
03:56:47,737 --> 03:56:49,272
WORKED BEST BUT DIDN'T SEEM TO

6135
03:56:49,272 --> 03:56:50,574
MAKE A BIG DIFFERENCE.

6136
03:56:50,574 --> 03:56:52,576
THIS TOUCHES UPON A VERY

6137
03:56:52,576 --> 03:56:53,577
IMPORTANT QUESTION NOT RESOLVED

6138
03:56:53,577 --> 03:56:56,379
YET FOR THE FIELD OF EEG AND

6139
03:56:56,379 --> 03:56:57,147
BUILDING FOUNDATION MODELS, WHAT

6140
03:56:57,147 --> 03:57:05,021
IS THE BEST TOKENIZATION SCHEME.

6141
03:57:05,021 --> 03:57:10,293
WE KNOW BUILDING UNITS MAKE UP

6142
03:57:10,293 --> 03:57:11,161
LANGUAGE, A SENTENCE, WE CAN

6143
03:57:11,161 --> 03:57:12,028
LOOK FOR THAT.

6144
03:57:12,028 --> 03:57:15,031
WHAT IS A BUILDING UNIT FOR EEG

6145
03:57:15,031 --> 03:57:15,398
TIME SERIES?

6146
03:57:15,398 --> 03:57:18,368
I THINK THIS IS SOMETHING FOR

6147
03:57:18,368 --> 03:57:20,670
NOW PEOPLE ARE TAKING WILD

6148
03:57:20,670 --> 03:57:24,741
GUESSES, IT WORKS OKAY WITH ONE

6149
03:57:24,741 --> 03:57:30,647
SECONDS, MAYBE 200 MILLISECONDS

6150
03:57:30,647 --> 03:57:33,283
WORKS BEST, WE'RE LOOKING WITH

6151
03:57:33,283 --> 03:57:34,684
STUDENTS INTO INFORMED WAYS

6152
03:57:34,684 --> 03:57:37,754
EXPLORING BUILDING BLOCKS OVER

6153
03:57:37,754 --> 03:57:39,923
TIME FOR EEG, AND SO WE'LL SEE

6154
03:57:39,923 --> 03:57:44,094
IF THAT IMPROVES THE RESULTS

6155
03:57:44,094 --> 03:57:44,394
DOWNSTREAM.

6156
03:57:44,394 --> 03:57:47,898
>> NOW, SHIFTING SLIGHT
GEARS

6157
03:57:47,898 --> 03:57:51,401
HERE BUT STAYING ON THE WHAT CAN

6158
03:57:51,401 --> 03:57:54,271
BE DONE WITH IT, EARLIER TODAY

6159
03:57:54,271 --> 03:57:55,839
WE WERE TALKING ABOUT EXTENSION

6160
03:57:55,839 --> 03:58:04,981
OF THE  WORK FROM PLACES LIKE

6161
03:58:04,981 --> 03:58:09,686
ALEX HUTH'S LAB, TO BE ABLE TO

6162
03:58:09,686 --> 03:58:11,788
TRANSLATE EEG TO TEXT, AND I

6163
03:58:11,788 --> 03:58:14,057
KNOW THAT THE LAB IS WORKING,

6164
03:58:14,057 --> 03:58:15,492
THEY PRESENTED EARLIER, ABOUT

6165
03:58:15,492 --> 03:58:17,127
THE ABILITY TO TRANSLATE ACROSS

6166
03:58:17,127 --> 03:58:18,461
PEOPLE WITH MINIMAL TRAINING

6167
03:58:18,461 --> 03:58:20,864
DATA FROM ANY INDIVIDUAL PERSON.

6168
03:58:20,864 --> 03:58:23,366
SO MY QUESTION IS ASSUME YOU GET

6169
03:58:23,366 --> 03:58:25,001
TO THE PLACE WITH MUCH HIGHER

6170
03:58:25,001 --> 03:58:27,837
ACCURACY WITH MODEL YOU THAT YOU

6171
03:58:27,837 --> 03:58:29,506
CREATED, CAN YOU THEN SAY, YOU

6172
03:58:29,506 --> 03:58:34,211
KNOW, HERE IS A CHUNK OF EEG,

6173
03:58:34,211 --> 03:58:36,446
TRANSLATE THIS INTO WHAT THE

6174
03:58:36,446 --> 03:58:37,914
PERSON WAS THINKING AT THE TIME,

6175
03:58:37,914 --> 03:58:41,151
GIVEN THAT IT IS AN EEG LANGUAGE

6176
03:58:41,151 --> 03:58:44,621
MODEL THAT IS, YOU KNOW, IT

6177
03:58:44,621 --> 03:58:45,155
SPEAKS EEG?

6178
03:58:45,155 --> 03:58:46,690
>> SO I THINK WE'LL SEE
WORK

6179
03:58:46,690 --> 03:58:47,757
LIKE THIS THAT COMES OUT.

6180
03:58:47,757 --> 03:58:48,825
THE QUESTION IS ALWAYS GOING TO

6181
03:58:48,825 --> 03:58:50,293
BE HOW RELIABLE IT IS.

6182
03:58:50,293 --> 03:58:51,828
WHEN YOU HAVE THE GROUND TRUTH,

6183
03:58:51,828 --> 03:58:54,564
YOU HAVE THE LABELS, YOU CAN

6184
03:58:54,564 --> 03:58:55,632
CHECK BACK.

6185
03:58:55,632 --> 03:58:56,866
BUT YOU ALWAYS HAVE THE SAME

6186
03:58:56,866 --> 03:58:58,835
ISSUE WITH MACHINE LEARNING AND

6187
03:58:58,835 --> 03:59:00,704
PREDICTIONS.

6188
03:59:00,704 --> 03:59:03,740
IF YOU HAVE AN 80% -- WE NEED TO

6189
03:59:03,740 --> 03:59:05,809
FIND MORE SHORT WAYS OF

6190
03:59:05,809 --> 03:59:08,011
IDENTIFYING ACCURACY IN THIS

6191
03:59:08,011 --> 03:59:10,847
SPECIFIC CONDITION -- SMART WAYS

6192
03:59:10,847 --> 03:59:14,684
OF IDENTIFYING ACCURACY.

6193
03:59:14,684 --> 03:59:15,318
>> THEORETICALLY POSSIBLE
NOT

6194
03:59:15,318 --> 03:59:16,152
HIGHLY ACCURATE.

6195
03:59:16,152 --> 03:59:17,954
>> YES, I SEE THE QUESTION
IN

6196
03:59:17,954 --> 03:59:19,723
THE CHAT, DEBRA'S QUESTION ABOUT

6197
03:59:19,723 --> 03:59:22,292
THE LABELS AND GROUND TRUTH, IN

6198
03:59:22,292 --> 03:59:23,760
THIS CASE LABELS WOULD BE FOR

6199
03:59:23,760 --> 03:59:25,328
EXAMPLE IF YOU COLLECT DATA

6200
03:59:25,328 --> 03:59:27,264
WHERE YOU ARE ASKING INDIVIDUALS

6201
03:59:27,264 --> 03:59:31,768
TO EITHER SAY SOMETHING OR THINK

6202
03:59:31,768 --> 03:59:32,969
ABOUT A STATEMENT, SO TO THE

6203
03:59:32,969 --> 03:59:34,271
EXTENT THAT THEY ARE FOLLOWING

6204
03:59:34,271 --> 03:59:35,705
INSTRUCTIONS YOU COULD BUILD A

6205
03:59:35,705 --> 03:59:37,974
DATASET WHERE YOU HAVE THAT TYPE

6206
03:59:37,974 --> 03:59:39,309
OF LABELING WHERE THEY WERE

6207
03:59:39,309 --> 03:59:40,410
INSTRUCTED TO THINK ABOUT THIS

6208
03:59:40,410 --> 03:59:43,246
OR THINK ABOUT THAT OR WERE

6209
03:59:43,246 --> 03:59:47,651
SHOWN IMAGES OF ANIMALS VERSUS

6210
03:59:47,651 --> 03:59:50,053
HOUSES, THEN YOU'D BE OUT OF THE

6211
03:59:50,053 --> 03:59:50,186
--

6212
03:59:50,186 --> 03:59:51,921
>> WHOSE MODEL HAS GROUND
TRUTH?

6213
03:59:51,921 --> 03:59:53,757
HERE IS ALL OF THE, YOU KNOW,

6214
03:59:53,757 --> 03:59:56,059
PODCASTS A PERSON IS LISTENING

6215
03:59:56,059 --> 03:59:59,362
TO, AND HERE IS, YOU KNOW, THE

6216
03:59:59,362 --> 04:00:01,665
DATA, SO THEN THEY CREATED

6217
04:00:01,665 --> 04:00:03,733
TRANSFORMER THAT IS ABLE TO

6218
04:00:03,733 --> 04:00:04,834
TRANSFORM THAT FROM ONE

6219
04:00:04,834 --> 04:00:06,803
INDIVIDUAL TO THE NEXT

6220
04:00:06,803 --> 04:00:07,103
INDIVIDUAL.

6221
04:00:07,103 --> 04:00:09,539
SO I'M ASSUMING A WORLD IN WHICH

6222
04:00:09,539 --> 04:00:14,144
WE HAVE EEG WITH GROUND  TRUTH

6223
04:00:14,144 --> 04:00:15,378
FOR LANGUAGE WONDERING THE

6224
04:00:15,378 --> 04:00:17,347
EXTEND TO WHICH YOUR MODEL

6225
04:00:17,347 --> 04:00:22,185
ALLOWS ME TO TAKE PUBLICLY

6226
04:00:22,185 --> 04:00:24,454
AVAILABLE EEG DATA, FEED IT IN,

6227
04:00:24,454 --> 04:00:25,689
SAY WHAT WAS THE RESEARCH

6228
04:00:25,689 --> 04:00:29,826
SUBJECT THINKING AT THE TIME

6229
04:00:29,826 --> 04:00:31,061
THIS WAS RECORDED. 

6230
04:00:31,061 --> 04:00:33,997
>> IT'S A TOUGH QUESTION.

6231
04:00:33,997 --> 04:00:36,499
I DON'T KNOW IF THAT'S -- SO I

6232
04:00:36,499 --> 04:00:37,600
DON'T KNOW HOW RELIABLE THAT WAS

6233
04:00:37,600 --> 04:00:40,036
GOING TO BE. 

6234
04:00:40,036 --> 04:00:44,407
SO WHETHER WE'LL BE, YOU KNOW --

6235
04:00:44,407 --> 04:00:46,576
WE'LL GET A ROUGH ESTIMATE, IT'S

6236
04:00:46,576 --> 04:00:52,615
SOMETHING LIKE ONE OF THE

6237
04:00:52,615 --> 04:00:53,817
REASONS I'M IN NEUROSCIENCE AS A

6238
04:00:53,817 --> 04:00:57,087
KID MY DESIRE WAS TO BUILD A

6239
04:00:57,087 --> 04:01:00,924
MACHINE TO IN REAL TIME BUILD A

6240
04:01:00,924 --> 04:01:02,759
MACHINE OF IMAGES WHILE PEOPLE

6241
04:01:02,759 --> 04:01:03,993
ARE DREAMING.

6242
04:01:03,993 --> 04:01:07,697
>> MY NIGHTMARE, YOUR
DREAM, I

6243
04:01:07,697 --> 04:01:09,132
LOVE IT.

6244
04:01:09,132 --> 04:01:12,302
>> SO, YES, I THINK WITH
ENOUGH

6245
04:01:12,302 --> 04:01:13,503
DATA, HIGH QUALITY DATA, WE

6246
04:01:13,503 --> 04:01:14,938
MIGHT BE GETTING CLOSER TO

6247
04:01:14,938 --> 04:01:16,806
THINGS LIKE THIS, WITH THE MODEL

6248
04:01:16,806 --> 04:01:18,441
WE'RE BUILDING, YOU CAN SEE THE

6249
04:01:18,441 --> 04:01:23,146
SCALE, HOW MUCH DATA THIS WILL

6250
04:01:23,146 --> 04:01:23,713
REQUIRE.

6251
04:01:23,713 --> 04:01:26,616
>> YOURS DOESN'T TRANSLATE
FROM

6252
04:01:26,616 --> 04:01:27,984
EEG TO TEXT, RIGHT?

6253
04:01:27,984 --> 04:01:29,052
MEANING I COULDN'T UPLOAD IT

6254
04:01:29,052 --> 04:01:31,020
INTO THE -- IT'S GOING TO

6255
04:01:31,020 --> 04:01:32,756
PREDICT OUT IN EEG, NOT IN

6256
04:01:32,756 --> 04:01:35,191
LANGUAGE, SO YOU NEED SOMETHING

6257
04:01:35,191 --> 04:01:41,531
THAT WOULD TRANSLATE BETWEEN THE

6258
04:01:41,531 --> 04:01:43,933
NEURO-GPT AND LANGUAGE MODEL.

6259
04:01:43,933 --> 04:01:46,903
JOHN, I'M SORRY. 

6260
04:01:46,903 --> 04:01:48,972
I SEE YOUR HAND IS UP.

6261
04:01:48,972 --> 04:01:52,275
>> THIS IS SOMETHING WE
TOUCHED

6262
04:01:52,275 --> 04:01:56,346
ON BEFORE.

6263
04:01:56,346 --> 04:01:59,182
NITA, YOUR NIGHTMARE, KARIM'S

6264
04:01:59,182 --> 04:01:59,382
DREAM.

6265
04:01:59,382 --> 04:02:01,050
>> I THINK IT'S
INCREDIBLE, IN

6266
04:02:01,050 --> 04:02:05,622
THAT SENSE.

6267
04:02:05,622 --> 04:02:08,158
>> WHAT'S SOBERING, WHEN
HUTH

6268
04:02:08,158 --> 04:02:09,526
PUBLISHED HIS FIRST PAPER IT

6269
04:02:09,526 --> 04:02:10,427
TOOK FOUR OR FIVE DAYS OF

6270
04:02:10,427 --> 04:02:12,529
TRAINING IN THE SCANNER, WHO IS

6271
04:02:12,529 --> 04:02:14,597
GOING TO DO THAT?

6272
04:02:14,597 --> 04:02:16,900
IF THEY DISTRACTED THE PERSON OR

6273
04:02:16,900 --> 04:02:18,201
DIDN'T HAVE THEM CONCENTRATE IT

6274
04:02:18,201 --> 04:02:20,170
DIDN'T WORK.

6275
04:02:20,170 --> 04:02:22,472
AND THEN NEXT THING, WE TURN

6276
04:02:22,472 --> 04:02:26,643
AROUND,  HE HAD A TRANSFORMER

6277
04:02:26,643 --> 04:02:27,844
THAT COULD GET IT.

6278
04:02:27,844 --> 04:02:29,813
THIS SPEAKS TO THINGS ARE MOVING

6279
04:02:29,813 --> 04:02:38,655
SO FAST AND HOW DO WE - PROTECT

6280
04:02:38,655 --> 04:02:41,090
IS THE WRONG WORD BUT PREPARE

6281
04:02:41,090 --> 04:02:45,795
OURSELVES FOR THAT

6282
04:02:45,795 --> 04:02:46,229
INEVITABILITY.

6283
04:02:46,229 --> 04:02:51,501
A BCI BRAIN TO SPEECH NEURAL

6284
04:02:51,501 --> 04:02:54,003
PROSTHESIS THIS WEEK, MANY YEARS

6285
04:02:54,003 --> 04:02:55,405
WORK, GREAT WORK, ALWAYS

6286
04:02:55,405 --> 04:02:58,274
REQUIRED EXTENSIVE TRAINING AND

6287
04:02:58,274 --> 04:03:01,010
VERY LIMITED VOCABULARY, GOOD

6288
04:03:01,010 --> 04:03:01,277
ACCURACY.

6289
04:03:01,277 --> 04:03:03,680
BOOM, THIS THING WAS PLUG AND

6290
04:03:03,680 --> 04:03:04,714
PLAY, REQUIRED ESSENTIALLY NO

6291
04:03:04,714 --> 04:03:05,081
TRAINING.

6292
04:03:05,081 --> 04:03:07,016
AFTER THE FIRST HALF HOUR THEY

6293
04:03:07,016 --> 04:03:09,619
GOT 50 WORDS AT CLOSE TO 100%

6294
04:03:09,619 --> 04:03:13,122
ACCURACY, THEY GOT UP TO 125,000

6295
04:03:13,122 --> 04:03:14,123
WORDS, 95% ACCURACY WITH MAYBE A

6296
04:03:14,123 --> 04:03:15,558
DAY OR TWO OF TRAINING.

6297
04:03:15,558 --> 04:03:18,628
THIS IS MIND BOGGLING.

6298
04:03:18,628 --> 04:03:21,030
IT SPEAKS TO THE POWER OF

6299
04:03:21,030 --> 04:03:22,765
ALGORITHMS AND THESE MODELS AND

6300
04:03:22,765 --> 04:03:24,100
IT'S JUST INTERESTING TO ME,

6301
04:03:24,100 --> 04:03:25,735
WE'RE GOING TO HAVE TO FIND A

6302
04:03:25,735 --> 04:03:27,904
WAY TO NOT GET IN THE WAY OF THE

6303
04:03:27,904 --> 04:03:29,539
PROGRESS BUT ALSO PREPARE

6304
04:03:29,539 --> 04:03:30,773
OURSELVES WHAT IT'S GOING TO

6305
04:03:30,773 --> 04:03:31,441
BRING US.

6306
04:03:31,441 --> 04:03:35,578
THESE KIND OF THINGS BRING IT TO

6307
04:03:35,578 --> 04:03:35,912
LIFE.

6308
04:03:35,912 --> 04:03:39,082
>> WHAT WE NEED TO
TRANSITION

6309
04:03:39,082 --> 04:03:41,384
THE CONVERSATION TO NEXT AND

6310
04:03:41,384 --> 04:03:43,019
EXACTLY ON TIME, SO THANK YOU.

6311
04:03:43,019 --> 04:03:44,354
>> WE'VE DONE THIS BEFORE,
NITA.

6312
04:03:44,354 --> 04:03:46,956
>> I KNOW YOU HAVE.

6313
04:03:46,956 --> 04:03:49,392
NEXT STEPS FOR NEWG AND TAKE

6314
04:03:49,392 --> 04:03:51,995
THIS EXCITING RESEARCH WHICH IS

6315
04:03:51,995 --> 04:03:55,832
EXTRAORDINARY, HOW QUICKLY IT'S

6316
04:03:55,832 --> 04:03:57,567
TRANSFORMING, THIS PAPER, WHEN

6317
04:03:57,567 --> 04:04:00,103
IT CAME OUT, INSTANTLY GRABBED

6318
04:04:00,103 --> 04:04:02,405
MY ATTENTION, FASCINATED BY THE

6319
04:04:02,405 --> 04:04:02,605
WORK.

6320
04:04:02,605 --> 04:04:05,842
THANK YOU BOTH FOR A RICH AND

6321
04:04:05,842 --> 04:04:08,978
INTERESTING CONVERSATION. 

6322
04:04:08,978 --> 04:04:11,014
THANK YOU, DR. MATHEWS, FOR

6323
04:04:11,014 --> 04:04:13,883
HELPING US TEE UP NEXT STEPS,

6324
04:04:13,883 --> 04:04:15,752
THANK YOU FOR GROUNDBREAKING

6325
04:04:15,752 --> 04:04:16,419
RESEARCH BOTH EXCITING,

6326
04:04:16,419 --> 04:04:17,987
IMPORTANT, KEEPS US IN BUSINESS

6327
04:04:17,987 --> 04:04:20,256
OF TRYING TO FIGURE OUT WHAT

6328
04:04:20,256 --> 04:04:20,690
NEXT.

6329
04:04:20,690 --> 04:04:22,158
AND LET'S TRANSITION THE

6330
04:04:22,158 --> 04:04:23,526
CONVERSATION THERE.

6331
04:04:23,526 --> 04:04:28,331
I THINK NINA HAS A SLIDE TO GET

6332
04:04:28,331 --> 04:04:29,532
US STARTED.

6333
04:04:29,532 --> 04:04:33,002
LET'S START HERE WHICH IS TO

6334
04:04:33,002 --> 04:04:35,004
REALLY THINK ABOUT WHAT ARE THE

6335
04:04:35,004 --> 04:04:39,475
NEXT STEPS AND WHERE DO WE AS

6336
04:04:39,475 --> 04:04:43,313
NEWG WANT TO JUMP IN, LIKE IF

6337
04:04:43,313 --> 04:04:46,282
IT'S DEVELOPING A SET OF POINTS

6338
04:04:46,282 --> 04:04:47,684
TO CONSIDER FOR BRAIN-FUNDED

6339
04:04:47,684 --> 04:04:48,251
RESEARCH?

6340
04:04:48,251 --> 04:04:50,320
IS IT, YOU KNOW, POTENTIALLY

6341
04:04:50,320 --> 04:04:52,655
OPENING UP AREAS FOR FUTURE

6342
04:04:52,655 --> 04:04:54,257
FUNDING OPPORTUNITIES?

6343
04:04:54,257 --> 04:04:55,358
IS IT WORKING TOGETHER TO

6344
04:04:55,358 --> 04:04:57,660
DEVELOP A PAPER THAT GOES BEYOND

6345
04:04:57,660 --> 04:04:58,595
POINTS TO CONSIDER?

6346
04:04:58,595 --> 04:05:00,830
DO WE THINK THERE ARE UNIQUE

6347
04:05:00,830 --> 04:05:02,899
ISSUES TO GRAPPLE WITH HERE THAT

6348
04:05:02,899 --> 04:05:05,435
ARE DIFFERENT AND WARRANT

6349
04:05:05,435 --> 04:05:06,936
JUMPING INTO THE NEXT STEPS?

6350
04:05:06,936 --> 04:05:08,338
LET ME OPEN THE FLOOR TO

6351
04:05:08,338 --> 04:05:13,710
CONVERSATION.

6352
04:05:13,710 --> 04:05:14,811
I DON'T SEE ANYONE BECAUSE OF

6353
04:05:14,811 --> 04:05:15,345
THE SLIDES.

6354
04:05:15,345 --> 04:05:16,679
MAYBE WE COULD TAKE THE SLIDE

6355
04:05:16,679 --> 04:05:17,213
DOWN.

6356
04:05:17,213 --> 04:05:20,083
THERE WE GO.

6357
04:05:20,083 --> 04:05:24,153
CHRISTINE, PLEASE.

6358
04:05:24,153 --> 04:05:25,755
>> I WANTED TO ASK DEBRA
SINCE

6359
04:05:25,755 --> 04:05:28,057
SHE WENT THROUGH SLIDES WITH

6360
04:05:28,057 --> 04:05:30,360
FRAMEWORKS AND GUIDELINES AND

6361
04:05:30,360 --> 04:05:31,594
MEETINGS AND THINGS THAT HAD

6362
04:05:31,594 --> 04:05:34,097
COME UP WITH LOTS OF I DON'T

6363
04:05:34,097 --> 04:05:35,398
KNOW PRINCIPLES AND ETHICAL

6364
04:05:35,398 --> 04:05:37,400
FRAMEWORKS, THINGS TO THINK

6365
04:05:37,400 --> 04:05:39,302
ABOUT, I GUESS GIVEN ALL OF

6366
04:05:39,302 --> 04:05:42,538
THAT, I KNOW ALL OF US ARE --

6367
04:05:42,538 --> 04:05:43,640
I'LL SPEAK FOR MYSELF, ALL OF US

6368
04:05:43,640 --> 04:05:47,110
I THINK ARE TRYING TO GRASP SOME

6369
04:05:47,110 --> 04:05:49,545
OF THE SPEED OF THIS RESEARCH

6370
04:05:49,545 --> 04:05:51,047
AND WHAT IT MEANS FOR ALL THE

6371
04:05:51,047 --> 04:05:55,018
KINDS OF THINGS WE THINK ABOUT

6372
04:05:55,018 --> 04:05:57,654
BUT GIVEN THE SORT OF WEALTH OF

6373
04:05:57,654 --> 04:05:58,588
ATTENTION TO THIS MATTER, THE

6374
04:05:58,588 --> 04:06:00,590
NUMBER OF PEOPLE COMING UP WITH

6375
04:06:00,590 --> 04:06:03,426
FRAMEWORKS, DO YOU THINK THERE'S

6376
04:06:03,426 --> 04:06:05,662
SOMETHING THAT THE NEUROETHICS

6377
04:06:05,662 --> 04:06:07,330
WORKING GROUP OF BRAIN CAN

6378
04:06:07,330 --> 04:06:08,598
UNIQUELY CONTRIBUTE?

6379
04:06:08,598 --> 04:06:10,033
WE'RE NOT GOING TO HOLD YOU TO

6380
04:06:10,033 --> 04:06:10,333
IT.

6381
04:06:10,333 --> 04:06:13,636
I JUST WANT YOUR OPINION ON IT.

6382
04:06:13,636 --> 04:06:13,903
>> SURE.

6383
04:06:13,903 --> 04:06:17,674
I MEAN, YES, I THINK THERE IS.

6384
04:06:17,674 --> 04:06:20,843
I'LL GIVE A COUPLE REASONS.

6385
04:06:20,843 --> 04:06:24,681
ONE, THERE HAS BEEN A TON OF

6386
04:06:24,681 --> 04:06:26,416
WORK, CONSENSUS BUILDING ABOUT

6387
04:06:26,416 --> 04:06:28,951
RELEVANT PRINCIPLES, THERE

6388
04:06:28,951 --> 04:06:30,153
CONTINUE TO BE DISAGREEMENTS

6389
04:06:30,153 --> 04:06:31,788
ABOUT TO WHOM AND WHICH STAGES

6390
04:06:31,788 --> 04:06:34,323
OF THE PROCESS THE PRINCIPLES

6391
04:06:34,323 --> 04:06:36,726
APPLY, WHO IS RESPONSIBLE FOR

6392
04:06:36,726 --> 04:06:38,895
UPHOLDING THEM, RIGHT?

6393
04:06:38,895 --> 04:06:40,863
IT'S GOING TO -- BECAUSE A.I. IS

6394
04:06:40,863 --> 04:06:45,635
NOT ONE THING, IT'S MANY

6395
04:06:45,635 --> 04:06:47,003
DIFFERENT THINGS, THERE IS WORK

6396
04:06:47,003 --> 04:06:49,839
THAT NEEDS TO BE DONE IN EACH

6397
04:06:49,839 --> 04:06:50,940
INDIVIDUAL CONTEXT TO HELP GUIDE

6398
04:06:50,940 --> 04:06:54,210
PEOPLE TO FIGURE OUT, OKAY, HERE

6399
04:06:54,210 --> 04:06:56,079
ARE PRINCIPLES THAT SHOULD BE

6400
04:06:56,079 --> 04:06:59,582
GUIDING THIS STUFF, WHAT DOES IT

6401
04:06:59,582 --> 04:07:02,552
MEAN TO TRANSLATE INTO DECISIONS

6402
04:07:02,552 --> 04:07:05,288
FOR THIS TECHNOLOGY AT THIS

6403
04:07:05,288 --> 04:07:05,955
STAGE OF DEVELOPMENT?

6404
04:07:05,955 --> 04:07:10,226
I THINK THAT'S THE KIND OF SORT

6405
04:07:10,226 --> 04:07:13,062
OF MORE GRANULAR GUIDANCE WE

6406
04:07:13,062 --> 04:07:17,467
NEED.

6407
04:07:17,467 --> 04:07:20,770
>> WOULD YOU SAY THEN SORT
OF I

6408
04:07:20,770 --> 04:07:23,039
DON'T KNOW SIMILAR TO WHAT WE

6409
04:07:23,039 --> 04:07:25,508
DID TODAY, YOU KNOW, PICK A

6410
04:07:25,508 --> 04:07:27,276
COUPLE VERY SPECIFIC CASES AND

6411
04:07:27,276 --> 04:07:30,747
THINK ABOUT MAYBE WHAT THE

6412
04:07:30,747 --> 04:07:31,948
PRINCIPLES ARE, WHAT FRAMEWORKS

6413
04:07:31,948 --> 04:07:34,350
ARE, HOW THEY DO OR DON'T APPLY,

6414
04:07:34,350 --> 04:07:35,685
USE THAT AS ILLUSTRATIVE TO THE

6415
04:07:35,685 --> 04:07:36,452
QUESTIONS THAT YOU'RE SAYING

6416
04:07:36,452 --> 04:07:38,788
STILL NEED TO BE ADDRESSED LIKE

6417
04:07:38,788 --> 04:07:40,590
HOW DOES, YOU KNOW, WHO DO THEY

6418
04:07:40,590 --> 04:07:42,859
APPLY TO, WHO IS SUPPOSED TO

6419
04:07:42,859 --> 04:07:43,159
UPHOLD THEM?

6420
04:07:43,159 --> 04:07:45,094
>> NO, I THINK THAT WOULD
BE A

6421
04:07:45,094 --> 04:07:47,830
GREAT STARTING PLACE.

6422
04:07:47,830 --> 04:07:52,602
THE FIRST CASE, THE SET OF

6423
04:07:52,602 --> 04:07:53,169
QUESTIONS EACH INVESTIGATOR

6424
04:07:53,169 --> 04:07:55,071
SHOULD BE ASKING ABOUT VARIOUS

6425
04:07:55,071 --> 04:07:56,139
ASPECTS OF RESEARCH, I THINK

6426
04:07:56,139 --> 04:08:00,076
THAT'S THE KIND OF GUIDANCE

6427
04:08:00,076 --> 04:08:00,943
FOLKS NEED.

6428
04:08:00,943 --> 04:08:03,045
OKAY, I GET THIS IS CONTENTIOUS

6429
04:08:03,045 --> 04:08:06,883
BUT WHAT DO I DO ABOUT THAT?

6430
04:08:06,883 --> 04:08:07,183
>> RIGHT.

6431
04:08:07,183 --> 04:08:08,651
THAT'S HELPFUL.

6432
04:08:08,651 --> 04:08:10,720
THANK YOU.

6433
04:08:10,720 --> 04:08:11,854
>> YEP.

6434
04:08:11,854 --> 04:08:14,090
>> I SEE ONE TAKER FOR
INTEREST

6435
04:08:14,090 --> 04:08:15,958
IN THE POTENTIAL TO DEVELOP

6436
04:08:15,958 --> 04:08:16,259
COMMENTARY.

6437
04:08:16,259 --> 04:08:17,827
SOUNDS LIKE THERE'S OPEN SPACES

6438
04:08:17,827 --> 04:08:19,495
TO BE DEVELOPED.

6439
04:08:19,495 --> 04:08:21,464
TOR, YOU HAVE YOUR REAL HAND UP

6440
04:08:21,464 --> 04:08:22,665
WHICH IS CONFUSING BUT I'M HAPPY

6441
04:08:22,665 --> 04:08:24,734
TO CALL ON YOUR REAL HAND

6442
04:08:24,734 --> 04:08:27,270
INSTEAD OF THE VIRTUAL ONE.

6443
04:08:27,270 --> 04:08:29,138
>> I HAVEN'T FIGURED OUT
HOW TO

6444
04:08:29,138 --> 04:08:31,340
RAISE MY HAND IN THIS INTERFACE.

6445
04:08:31,340 --> 04:08:36,479
>> FAIR. 

6446
04:08:36,479 --> 04:08:38,481
REAL HANDS WORK TOO.

6447
04:08:38,481 --> 04:08:39,982
>> ONE OF THE TEAMS THAT'S

6448
04:08:39,982 --> 04:08:42,518
EMERGED THAT I WAS INTERESTED IN

6449
04:08:42,518 --> 04:08:46,022
IS THE TRADEOFFS AMONG ETHICAL

6450
04:08:46,022 --> 04:08:47,657
PRINCIPLES, AND I SORT OF SEE

6451
04:08:47,657 --> 04:08:51,160
THIS AS A GRAPH IN MY MIND,

6452
04:08:51,160 --> 04:08:53,996
MULTIPLE PRINCIPLES, YOU KNOW,

6453
04:08:53,996 --> 04:08:56,332
OPEN SCIENCE CONFLICT WAS

6454
04:08:56,332 --> 04:08:58,167
PRIVACY, MAYBE FOUR, FIVE, SIX

6455
04:08:58,167 --> 04:09:01,370
AT LEAST DIFFERENT PRINCIPLES WE

6456
04:09:01,370 --> 04:09:01,971
DISCUSSED.

6457
04:09:01,971 --> 04:09:03,339
I WOULD ADVOCATE FOR THAT AS,

6458
04:09:03,339 --> 04:09:05,975
YOU KNOW, SOMETHING THAT COULD

6459
04:09:05,975 --> 04:09:07,510
BE MADE EXPLICIT BECAUSE OFTEN

6460
04:09:07,510 --> 04:09:09,879
IT'S, YOU KNOW, WE HAVE THESE

6461
04:09:09,879 --> 04:09:10,513
COMPETING PRINCIPLES, ONLY ONE

6462
04:09:10,513 --> 04:09:12,748
OR TWO IS IN OUR MIND OR

6463
04:09:12,748 --> 04:09:14,183
ATTENTION AT ANY GIVEN TIME.

6464
04:09:14,183 --> 04:09:17,220
IF YOU COULD SEE HOW THEY TRADE

6465
04:09:17,220 --> 04:09:18,654
OFF, RIGHT?

6466
04:09:18,654 --> 04:09:21,390
AND BE EXPLICIT OR CONSCIOUS

6467
04:09:21,390 --> 04:09:23,693
THOUSAND TO CHOOSE A TRADEOFF

6468
04:09:23,693 --> 04:09:25,027
BETWEEN CONFLICTING PRINCIPLES

6469
04:09:25,027 --> 04:09:27,563
IN ANY GIVEN SPACE, WHERE I

6470
04:09:27,563 --> 04:09:30,733
THINK A.I.-RELATED THINGS, THERE

6471
04:09:30,733 --> 04:09:31,801
WILL BE A DIFFERENT SOLUTION

6472
04:09:31,801 --> 04:09:34,437
THAN OTHER KINDS OF DATA SO I

6473
04:09:34,437 --> 04:09:35,204
LIKE THAT.

6474
04:09:35,204 --> 04:09:38,841
>> NOT ALWAYS EASY BUT
WITH MORE

6475
04:09:38,841 --> 04:09:39,909
PRACTICAL GUIDANCE THAT

6476
04:09:39,909 --> 04:09:41,143
IMPLEMENTS PRINCIPLES AND WHERE

6477
04:09:41,143 --> 04:09:42,778
CONFLICTS CAN REVOLVE CAN BE

6478
04:09:42,778 --> 04:09:44,413
INSTRUMENTAL AND HELPFUL AND HOW

6479
04:09:44,413 --> 04:09:47,917
THEY ARE DIFFERENT ACROSS

6480
04:09:47,917 --> 04:09:48,918
CONTEXTS.

6481
04:09:48,918 --> 04:09:49,085
AMY?

6482
04:09:49,085 --> 04:09:50,887
>> YEAH, GOING BACK TO
THIS IDEA

6483
04:09:50,887 --> 04:09:53,155
OF LIKE WHAT'S DIFFERENT HERE,

6484
04:09:53,155 --> 04:09:55,258
WHERE SHOULD WE SEE COORDINATING

6485
04:09:55,258 --> 04:09:57,326
EFFORTS VERSUS WHERE WE NEED

6486
04:09:57,326 --> 04:09:58,628
SORT OF SILOED EFFORTS WHICH

6487
04:09:58,628 --> 04:10:00,196
THERE'S PROBABLY A PLACE FOR

6488
04:10:00,196 --> 04:10:01,697
BOTH OF THEM, I WONDER IF

6489
04:10:01,697 --> 04:10:03,900
THERE'S A WAY TO GET LIKE

6490
04:10:03,900 --> 04:10:04,934
INVENTORY OF WHAT'S HAPPENING

6491
04:10:04,934 --> 04:10:07,503
ACROSS NIH WITH THIS SO I KNOW

6492
04:10:07,503 --> 04:10:10,373
FOR EXAMPLE LIKE I THINK NHGRI,

6493
04:10:10,373 --> 04:10:12,441
OR MAYBE THE COMMON FUND, THEY

6494
04:10:12,441 --> 04:10:14,510
ARE FUNDING GRANTS ON

6495
04:10:14,510 --> 04:10:17,480
DEVELOPMENT OF A.I. TOOLS, AS

6496
04:10:17,480 --> 04:10:19,782
PART OF IT CONSORTIUM OR EACH

6497
04:10:19,782 --> 04:10:21,517
GROUP HAS TO DEVELOP ETHICAL

6498
04:10:21,517 --> 04:10:22,718
FRAMEWORK FOR A.I., I IMAGINE

6499
04:10:22,718 --> 04:10:24,287
THERE ARE OTHER SORT OF CALLS

6500
04:10:24,287 --> 04:10:26,689
LIKE THIS IN OTHER INSTITUTES

6501
04:10:26,689 --> 04:10:27,757
AND CENTERS AND WONDER IF

6502
04:10:27,757 --> 04:10:30,726
THERE'S A WAY TO AT LEAST HAVE

6503
04:10:30,726 --> 04:10:31,494
AN UNDERSTANDING OF WHAT'S

6504
04:10:31,494 --> 04:10:33,562
HAPPENING AT THE NIH LEVEL WITH

6505
04:10:33,562 --> 04:10:36,866
REGARD TO THIS AND WHERE SORT OF

6506
04:10:36,866 --> 04:10:39,135
WE NEED TO BE PARTICIPATING IN

6507
04:10:39,135 --> 04:10:40,670
THOSE CONVERSATIONS VERSUS WHERE

6508
04:10:40,670 --> 04:10:42,438
THERE'S LIKE UNIQUE OPPORTUNITY

6509
04:10:42,438 --> 04:10:50,980
FOR US RELATED TO BRAIN DATA?

6510
04:10:50,980 --> 04:10:52,381
>> GOOD QUESTIONS.

6511
04:10:52,381 --> 04:10:56,218
JIM AND WALTER.

6512
04:10:56,218 --> 04:11:00,056
>> THANKS, NITA. 

6513
04:11:00,056 --> 04:11:01,023
DISCUSSION, FOCUSING MY THOUGHTS

6514
04:11:01,023 --> 04:11:03,259
ON THE IDEA OF CLINICAL VERSUS

6515
04:11:03,259 --> 04:11:08,030
FUNDAMENTAL RESEARCH A.I. AND

6516
04:11:08,030 --> 04:11:12,635
THOSE APPLICATIONS.

6517
04:11:12,635 --> 04:11:14,270
WE HEARD ABOUT LANGUAGE, EEG,

6518
04:11:14,270 --> 04:11:15,805
BUT SO MUCH MORE TIMES OF BRAIN

6519
04:11:15,805 --> 04:11:18,407
DATA THAT WILL IMPACT US, IF WE

6520
04:11:18,407 --> 04:11:20,609
GET SYNAPTIC QUESTIONS THERE MAY

6521
04:11:20,609 --> 04:11:24,213
BE PROPENSITIES TO PARTICULAR

6522
04:11:24,213 --> 04:11:25,681
TYPES OF PRECISION MAKING,

6523
04:11:25,681 --> 04:11:26,649
THINGS THAT COME OUT OF THE

6524
04:11:26,649 --> 04:11:27,049
THIS.

6525
04:11:27,049 --> 04:11:28,517
WHAT KINDS OF BRAIN DATA SHOULD

6526
04:11:28,517 --> 04:11:33,756
WE THINK ABOUT IN TERMS OF

6527
04:11:33,756 --> 04:11:34,757
PERHAPS FEDERATED DATABASES, HOY

6528
04:11:34,757 --> 04:11:37,360
IT'S GOING TO BE IMPACTED BY

6529
04:11:37,360 --> 04:11:39,328
A.I., AND SO I THINK SOMETHING

6530
04:11:39,328 --> 04:11:40,963
ALONG THOSE LINES WHERE WE SORT

6531
04:11:40,963 --> 04:11:44,700
OF THINK ABOUT CLINICAL VERSUS

6532
04:11:44,700 --> 04:11:45,668
FUNDAMENTAL SCIENCE, AND THE

6533
04:11:45,668 --> 04:11:47,536
BROAD EXPANSE OF THE TYPES OF

6534
04:11:47,536 --> 04:11:50,272
NEURODATA THAT'S COMING FROM THE

6535
04:11:50,272 --> 04:11:52,141
NEUROTECHNOLOGIES, NOT JUST

6536
04:11:52,141 --> 04:11:53,342
RELATED NECESSARILY, AND NOBODY

6537
04:11:53,342 --> 04:11:54,310
IS SUGGESTING THIS, BUT I

6538
04:11:54,310 --> 04:11:57,279
WOULDN'T WANT TO SEE IT RELATED

6539
04:11:57,279 --> 04:11:59,682
SOLELY TO, SAY, EEG OR fMRI OR

6540
04:11:59,682 --> 04:12:00,883
THAT BECAUSE THERE'S SO MANY

6541
04:12:00,883 --> 04:12:05,154
MORE TIMES OF DATA THAT WILL BE

6542
04:12:05,154 --> 04:12:06,389
IMPACTFUL PARTICULARLY

6543
04:12:06,389 --> 04:12:07,456
MULTI-MODAL DATA.

6544
04:12:07,456 --> 04:12:07,890
>> FAIR.

6545
04:12:07,890 --> 04:12:13,696
GOOD AND IMPORTANT POINTS.

6546
04:12:13,696 --> 04:12:13,929
WALTER?

6547
04:12:13,929 --> 04:12:15,031
>> I WAS WONDERING, I'M
REALLY

6548
04:12:15,031 --> 04:12:18,734
GOOD AT MAKING WORK FOR OTHER

6549
04:12:18,734 --> 04:12:20,169
PEOPLE, IN TERMS OF, YOU KNOW,

6550
04:12:20,169 --> 04:12:24,340
THE PAPER THAT WENT AROUND ON,

6551
04:12:24,340 --> 04:12:29,311
YOU KNOW, WHAT WAS IT, CODE OF

6552
04:12:29,311 --> 04:12:33,182
CONDUCT, PRINCIPLES FOR

6553
04:12:33,182 --> 04:12:34,316
ARTIFICIAL INTELLIGENCE, HEALTH,

6554
04:12:34,316 --> 04:12:35,351
HEALTHCARE, BIOMEDICAL SCIENCES,

6555
04:12:35,351 --> 04:12:36,685
IT WAS PRETTY GENERIC.

6556
04:12:36,685 --> 04:12:39,655
I WAS WONDERING IF OVER TIME WE

6557
04:12:39,655 --> 04:12:43,325
SHOULD BE LOOKING AT CREATING

6558
04:12:43,325 --> 04:12:46,195
SUCH A CODE OF CONDUCT FOR BRAIN

6559
04:12:46,195 --> 04:12:47,663
INITIATIVE WORK THAT CERTAINLY

6560
04:12:47,663 --> 04:12:52,334
WOULD INFORM YOU WILLS HERE AT

6561
04:12:52,334 --> 04:12:55,638
NIH BUT ALSO IF WE CAN WORK OUT

6562
04:12:55,638 --> 04:12:59,141
THE KINKS AS WE GO ALONG COULD

6563
04:12:59,141 --> 04:13:01,677
SERVE AS SOMETHING THAT WOULD BE

6564
04:13:01,677 --> 04:13:04,914
VALUABLE GLOBALLY.

6565
04:13:04,914 --> 04:13:07,983
AND ANYWAY THAT WAS ON MY TO-DO

6566
04:13:07,983 --> 04:13:09,852
LIST FOR YOU GUYS.

6567
04:13:09,852 --> 04:13:13,122
BUT THE OTHER THING WAS

6568
04:13:13,122 --> 04:13:14,457
MENTIONED, YOU KNOW, I'M

6569
04:13:14,457 --> 04:13:16,525
CO-CHAIRING THE A.I. PROJECT,

6570
04:13:16,525 --> 04:13:23,632
THE COMMON FUND ONE WITH GRACE

6571
04:13:23,632 --> 04:13:25,501
HERE, LOOKING AT INCORPORATING

6572
04:13:25,501 --> 04:13:29,438
IMAGING INTO OTHER TIMES OF --

6573
04:13:29,438 --> 04:13:35,377
TYPES OF DATA TO MAKE HEALTH

6574
04:13:35,377 --> 04:13:36,011
PREDICTIONS, WE HAVEN'T GOTTEN

6575
04:13:36,011 --> 04:13:38,514
OFF THE GROUND LOOKING AT

6576
04:13:38,514 --> 04:13:40,716
ETHICAL PRINCIPLE, AGAIN ON

6577
04:13:40,716 --> 04:13:43,219
ANOTHER TO-DO LIST, SO, YEAH,

6578
04:13:43,219 --> 04:13:44,687
THERE'S A LOT -- WE'RE GOING TO

6579
04:13:44,687 --> 04:13:45,754
LEARN A LOT.

6580
04:13:45,754 --> 04:13:48,491
HOPEFULLY WE DON'T MAKE TOO MANY

6581
04:13:48,491 --> 04:13:49,191
BAD MISTAKES.

6582
04:13:49,191 --> 04:13:54,296
WE'LL LEARN A LOT AS WE MOVE

6583
04:13:54,296 --> 04:13:58,033
INTO THIS SPACE.

6584
04:13:58,033 --> 04:13:58,801
>> IT'S GREAT TO THINK
ABOUT

6585
04:13:58,801 --> 04:14:00,769
DEVELOPING A SET OF POINTS TO

6586
04:14:00,769 --> 04:14:03,706
THAT ARE COULD BE USEFUL NOT

6587
04:14:03,706 --> 04:14:05,774
JUST FOR BRAIN BUT GLOBALLY, I

6588
04:14:05,774 --> 04:14:08,210
WOULD SAY GIVEN HOW RAPIDLY THE

6589
04:14:08,210 --> 04:14:08,844
FIELD IS PROGRESSING THAT'S

6590
04:14:08,844 --> 04:14:12,047
PROBABLY GOT TO BE A LIVING

6591
04:14:12,047 --> 04:14:22,591
DOCUMENT THAT CHANGES AS THINGS

6592
04:14:22,925 --> 04:14:23,092
CHANGE.

6593
04:14:23,092 --> 04:14:25,161
REMINDS ME OF ANONYMOUS SPERM

6594
04:14:25,161 --> 04:14:26,795
AND EGG DONATION, NOBODY THINKS

6595
04:14:26,795 --> 04:14:27,997
THERE'S THE POSSIBILITY OF

6596
04:14:27,997 --> 04:14:29,298
ANONYMITY IN THAT SPACE, IT'S

6597
04:14:29,298 --> 04:14:31,367
NOT A LET'S FIND WAYS TO LAYER

6598
04:14:31,367 --> 04:14:34,103
ANONYMITY OR PROVIDE PEOPLE WITH

6599
04:14:34,103 --> 04:14:35,204
ANONYMITY, IT'S AN UNDERSTANDING

6600
04:14:35,204 --> 04:14:37,840
THERE'S NO SUCH THING AS

6601
04:14:37,840 --> 04:14:38,374
ANONYMITY, THAT CHANGED

6602
04:14:38,374 --> 04:14:40,576
FUNDAMENTALLY IN LIGHT OF

6603
04:14:40,576 --> 04:14:41,577
CHANGES IN TECHNOLOGICAL

6604
04:14:41,577 --> 04:14:41,911
POSSIBILITIES.

6605
04:14:41,911 --> 04:14:43,746
AND SO I THINK WE'RE RAPIDLY

6606
04:14:43,746 --> 04:14:45,614
APPROACHING THAT SPACE HERE AND

6607
04:14:45,614 --> 04:14:47,349
THE QUESTION IS GIVEN THAT HOW

6608
04:14:47,349 --> 04:14:49,985
DO YOU START TO SHIFT GEARS AND

6609
04:14:49,985 --> 04:14:50,886
PROVIDE UPDATED GUIDANCE WITH

6610
04:14:50,886 --> 04:14:57,426
UNDERSTANDING THAT WE'RE IN A

6611
04:14:57,426 --> 04:14:57,793
DIFFERENT ERA.

6612
04:14:57,793 --> 04:15:00,062
>> HOPEFULLY A.I. CAN DO
THAT.

6613
04:15:00,062 --> 04:15:01,897
>> WE'LL ASK ChatGPT TO
WRITE

6614
04:15:01,897 --> 04:15:03,132
THAT FOR US.

6615
04:15:03,132 --> 04:15:04,633
NOT GOOD AT ETHICAL GUIDANCE.

6616
04:15:04,633 --> 04:15:09,138
>> DIDN'T HAVE THE RIGHT

6617
04:15:09,138 --> 04:15:10,472
TRAINING SET YET.

6618
04:15:10,472 --> 04:15:12,508
>> RIGHT MODEL SPEC FOR
BEING

6619
04:15:12,508 --> 04:15:12,741
ETHICAL.

6620
04:15:12,741 --> 04:15:14,076
>> WALTER, INTERESTING
POINT.

6621
04:15:14,076 --> 04:15:16,845
WE DO HAVE OUR NEUROETHICAL NINE

6622
04:15:16,845 --> 04:15:17,980
GUIDING PRINCIPLES, I THINK NINE

6623
04:15:17,980 --> 04:15:18,314
OF THEM.

6624
04:15:18,314 --> 04:15:20,749
THEY ARE A GREAT FRAMEWORK BUT I

6625
04:15:20,749 --> 04:15:22,785
THINK ESPECIALLY GIVEN THE

6626
04:15:22,785 --> 04:15:24,019
DEVELOPMENTS OVER THE LAST EVEN

6627
04:15:24,019 --> 04:15:26,655
YEAR, I THINK IT'S PROBABLY GOOD

6628
04:15:26,655 --> 04:15:29,258
TIME FOR THIS GROUP TO REVISIT

6629
04:15:29,258 --> 04:15:31,126
AND UNDERSTAND WHAT NEEDS

6630
04:15:31,126 --> 04:15:32,795
REFRESHING, WHAT NEEDS FLESHING

6631
04:15:32,795 --> 04:15:34,964
OUT, I THINK THE KINDS OF ISSUES

6632
04:15:34,964 --> 04:15:37,466
WE'RE DISCUSSING TODAY PROBABLY

6633
04:15:37,466 --> 04:15:40,536
WEREN'T REALLY OUT THERE AT THE

6634
04:15:40,536 --> 04:15:41,537
SURFACE IN MOST PEOPLE'S MINDS

6635
04:15:41,537 --> 04:15:42,504
FIVE YEARS AGO.

6636
04:15:42,504 --> 04:15:43,706
A GREAT THING FOR THIS GROUP TO

6637
04:15:43,706 --> 04:15:44,406
LOOK AT.

6638
04:15:44,406 --> 04:15:45,541
POINTS TO CONSIDER WOULD BE

6639
04:15:45,541 --> 04:15:45,908
GREAT.

6640
04:15:45,908 --> 04:15:48,711
WE DID HAVE THIS WORKSHOP, AND

6641
04:15:48,711 --> 04:15:56,518
THERE IS ACTUALLY A MANUSCRIPT

6642
04:15:56,518 --> 04:15:57,586
FLOATING AROUND.

6643
04:15:57,586 --> 04:15:58,721
DATA PRIVACY.

6644
04:15:58,721 --> 04:15:59,788
WE STARTED SCRATCHING THE

6645
04:15:59,788 --> 04:16:01,557
SURFACE OF SOME OF THESE ISSUES

6646
04:16:01,557 --> 04:16:02,891
YOU'RE TALKING ABOUT TODAY.

6647
04:16:02,891 --> 04:16:11,066
A LOT HAS HAPPENED SINCE.

6648
04:16:11,066 --> 04:16:13,702
THIS IS A TOPIC WITH US FOR THE

6649
04:16:13,702 --> 04:16:14,903
DURATION OR FORESEEABLE FUTURE

6650
04:16:14,903 --> 04:16:16,338
AND BEYOND, AND GOING TO GET

6651
04:16:16,338 --> 04:16:17,640
MORE INTERESTING TO SAY THE

6652
04:16:17,640 --> 04:16:17,840
LEAST.

6653
04:16:17,840 --> 04:16:19,708
WE DO NEED TO GET OUR HEADS

6654
04:16:19,708 --> 04:16:20,075
AROUND THIS.

6655
04:16:20,075 --> 04:16:22,144
I THINK THIS IS A GREAT

6656
04:16:22,144 --> 04:16:22,911
DISCUSSION.

6657
04:16:22,911 --> 04:16:26,582
I THINK A COMMENTARY POINT TO

6658
04:16:26,582 --> 04:16:28,183
CONSIDER, SCHOLARLY ARTICLES

6659
04:16:28,183 --> 04:16:29,084
WOULD BE GREAT.

6660
04:16:29,084 --> 04:16:31,153
AT LEAST I WANT TO KNOW MORE.

6661
04:16:31,153 --> 04:16:33,122
WE'LL BE HAVING THIS NEURAL A.I.

6662
04:16:33,122 --> 04:16:36,058
WORKSHOP THAT JOE AND OTHERS,

6663
04:16:36,058 --> 04:16:37,359
GRACE ARE CO-ORGANIZING FOR THIS

6664
04:16:37,359 --> 04:16:37,793
FALL.

6665
04:16:37,793 --> 04:16:39,895
THAT'S GOING TO RAISE SOME

6666
04:16:39,895 --> 04:16:40,663
INTERESTING ISSUES.

6667
04:16:40,663 --> 04:16:42,431
MAYBE WE CAN THINK ABOUT DIVING

6668
04:16:42,431 --> 04:16:44,166
INTO THIS MORE DEEPLY BOTH HERE

6669
04:16:44,166 --> 04:16:46,235
AND PERHAPS AT ANOTHER

6670
04:16:46,235 --> 04:16:48,304
NEWG-SPONSORED WORKSHOP TO DIG

6671
04:16:48,304 --> 04:16:48,470
IN.

6672
04:16:48,470 --> 04:16:49,438
THERE'S A LOT HERE.

6673
04:16:49,438 --> 04:16:52,741
I WANT TO BE ABLE TO TAKE

6674
04:16:52,741 --> 04:16:55,444
LEADERSHIP ON A LOT OF THESE

6675
04:16:55,444 --> 04:16:55,878
ISSUES.

6676
04:16:55,878 --> 04:16:57,613
SOME THINGS ARE GOING TO BE

6677
04:16:57,613 --> 04:17:00,249
PARTICULARLY FOR BRAIN BUT ALSO

6678
04:17:00,249 --> 04:17:02,084
TRANSLATABLE ACROSS THE BOARD

6679
04:17:02,084 --> 04:17:03,752
BECAUSE WE'RE CLOSER TO THE EDGE

6680
04:17:03,752 --> 04:17:06,622
OF, HOW SHOULD WE SAY,

6681
04:17:06,622 --> 04:17:07,456
UNINTENDED AND UNWANTED

6682
04:17:07,456 --> 04:17:07,923
CONSEQUENCES.

6683
04:17:07,923 --> 04:17:09,124
I WILL THROW THAT OUT THERE,

6684
04:17:09,124 --> 04:17:10,626
SOMETHING FOR THIS GROUP TO

6685
04:17:10,626 --> 04:17:11,527
THINK ABOUT.

6686
04:17:11,527 --> 04:17:14,463
ABOUT MAYBE WE COULD THINK ABOUT

6687
04:17:14,463 --> 04:17:16,632
LAYING OUT A YEAR OR TWO

6688
04:17:16,632 --> 04:17:17,966
STRATEGY FOR DIVING IN OR

6689
04:17:17,966 --> 04:17:19,401
GETTING OUR HEADS AROUND THIS.

6690
04:17:19,401 --> 04:17:20,169
>> GREAT POINT.

6691
04:17:20,169 --> 04:17:21,937
THAT'S WHY I THOUGHT WE NEED A

6692
04:17:21,937 --> 04:17:22,504
LIVING DOCUMENT.

6693
04:17:22,504 --> 04:17:23,772
THERE'S SOME THINGS WE KNOW

6694
04:17:23,772 --> 04:17:26,308
ALREADY WHICH ALLOW US TO

6695
04:17:26,308 --> 04:17:31,447
REVISIT AND SAY UPDATED THINGS.

6696
04:17:31,447 --> 04:17:33,549
MY POINT ABOUT THE GAMETE

6697
04:17:33,549 --> 04:17:36,719
ANONYMITY, I THINK WE'RE ALMOST

6698
04:17:36,719 --> 04:17:36,952
THERE.

6699
04:17:36,952 --> 04:17:38,487
>> THERE'S NO ANONYMITY
THAT CAN

6700
04:17:38,487 --> 04:17:41,023
BE GUARANTEED ANYMORE, RIGHT.

6701
04:17:41,023 --> 04:17:41,323
>> RIGHT.

6702
04:17:41,323 --> 04:17:42,958
GIVEN WE CAN ACCEPT THAT FACT

6703
04:17:42,958 --> 04:17:47,229
THEN THE QUESTION IS WHAT IS THE

6704
04:17:47,229 --> 04:17:48,530
UPDATE CONSENT AND GUIDANCE THAT

6705
04:17:48,530 --> 04:17:50,632
NEEDS TO BE PROVIDED.

6706
04:17:50,632 --> 04:17:52,034
SOME THINGS WE CAN ALREADY

6707
04:17:52,034 --> 04:17:53,135
APPROACH WITH UNDERSTANDING OF

6708
04:17:53,135 --> 04:17:55,037
HOW RAPIDLY THINGS HAVE CHANGED.

6709
04:17:55,037 --> 04:17:56,972
AND SOME WE NEED TO LEARN MORE.

6710
04:17:56,972 --> 04:17:58,941
THERE'S A LOT OF EFFORTS

6711
04:17:58,941 --> 04:18:01,577
HAPPENING ACROSS THE WORLD IN

6712
04:18:01,577 --> 04:18:03,445
THESE CASES AS CONVERGENCE.

6713
04:18:03,445 --> 04:18:05,147
CHRISTINE, PLEASE JUMP IN.

6714
04:18:05,147 --> 04:18:07,549
>> I WAS GOING TO -- YOU
WERE

6715
04:18:07,549 --> 04:18:10,052
JOKING ABOUT THE TRAINING DATA

6716
04:18:10,052 --> 04:18:10,853
FOR ETHICAL GUIDANCE FOR

6717
04:18:10,853 --> 04:18:12,688
ChatGPT WOULDN'T BE SO GOOD.

6718
04:18:12,688 --> 04:18:14,022
I THINK THAT'S RIGHT BECAUSE SO

6719
04:18:14,022 --> 04:18:19,361
MANY OF THE THINGS THAT WE HAVE

6720
04:18:19,361 --> 04:18:22,331
FOCUSED ON IN ETHICS LIKE

6721
04:18:22,331 --> 04:18:23,532
INFORMED CONSENT, IT'S A

6722
04:18:23,532 --> 04:18:25,067
QUESTION MARK ABOUT HOW IT

6723
04:18:25,067 --> 04:18:26,935
SHOULD WORK IN THIS SETTING.

6724
04:18:26,935 --> 04:18:28,470
I MEAN, WHAT ARE WE TALKING

6725
04:18:28,470 --> 04:18:31,306
ABOUT WHEN TALKING ABOUT THAT?

6726
04:18:31,306 --> 04:18:32,408
OR DECREASING RISK BY

6727
04:18:32,408 --> 04:18:33,509
DE-IDENTIFYING THINGS.

6728
04:18:33,509 --> 04:18:35,911
THE KINDS OF THINGS THAT HAVE

6729
04:18:35,911 --> 04:18:39,081
BEEN OUT THERE NEED TO BE

6730
04:18:39,081 --> 04:18:39,415
REVISITED.

6731
04:18:39,415 --> 04:18:42,050
AND I WAS -- I LIKE THE CONCEPT

6732
04:18:42,050 --> 04:18:51,326
THAT WAS RAISED EARLIER TODAY

6733
04:18:51,326 --> 04:18:52,327
ABOUT INSCRUTABILITY, AN

6734
04:18:52,327 --> 04:18:54,630
IMPORTANT CONCEPT IN TERMS OF

6735
04:18:54,630 --> 04:18:56,565
HOW THIS -- HOW THESE

6736
04:18:56,565 --> 04:18:59,868
TECHNOLOGIES GIVE US INFORMATION

6737
04:18:59,868 --> 04:19:06,442
THAT CAN BE USEFUL. 

6738
04:19:06,442 --> 04:19:06,875
>> AGREED.

6739
04:19:06,875 --> 04:19:08,844
WE HAVE SOME WORK CUT OUT FOR

6740
04:19:08,844 --> 04:19:11,146
US, THINGS WE CAN START.

6741
04:19:11,146 --> 04:19:12,781
JOHN, SOUNDS LIKE SOME COULD BE

6742
04:19:12,781 --> 04:19:14,082
FLESHED OUT FOR ADDITIONAL

6743
04:19:14,082 --> 04:19:15,751
WORKSHOPS BUT YOU PUT ON THE

6744
04:19:15,751 --> 04:19:18,020
TABLE POSSIBILITY OF US HERE AT

6745
04:19:18,020 --> 04:19:19,755
NEWG DOING ADDITIONAL WORKSHOP

6746
04:19:19,755 --> 04:19:20,889
THAT MIGHT FOCUS OUR EFFORTS

6747
04:19:20,889 --> 04:19:23,358
MORE SPECIFICALLY AND IN A MORE

6748
04:19:23,358 --> 04:19:24,693
TARGETED WAY.

6749
04:19:24,693 --> 04:19:26,128
LET ME INVITE OTHER NEWG MEMBERS

6750
04:19:26,128 --> 04:19:29,398
TO SAY WHAT DO YOU THINK ABOUT

6751
04:19:29,398 --> 04:19:33,035
THAT, ABOUT ADDITIONAL  WORKSHOP

6752
04:19:33,035 --> 04:19:39,575
TO TAKE A DEEPER DIVE.

6753
04:19:39,575 --> 04:19:41,443
SASKIA HAS A LOOK OF, YOU KNOW,

6754
04:19:41,443 --> 04:19:43,612
DISCOMFORT AND UNEASE.

6755
04:19:43,612 --> 04:19:46,615
BY ALL MEANS, SPEAK UP IF THAT'S

6756
04:19:46,615 --> 04:19:48,784
MORE WORK THAN TRANSLATION TO

6757
04:19:48,784 --> 04:19:52,621
BENEFIT AT THIS POINT.

6758
04:19:52,621 --> 04:19:53,689
LET'S GET PEOPLE TALK BECOME

6759
04:19:53,689 --> 04:19:55,557
WHAT WE NEED TO GET TO THE NEXT

6760
04:19:55,557 --> 04:19:58,093
STEP TOGETHER OF BEING ABLE TO

6761
04:19:58,093 --> 04:19:59,194
AT LEAST REVISIT AND UPDATE

6762
04:19:59,194 --> 04:20:00,696
GUIDANCE OF THE THINGS WE KNOW

6763
04:20:00,696 --> 04:20:04,433
NOW THAT COULD BE USEFUL.

6764
04:20:04,433 --> 04:20:05,667
I KNOW IT'S LATE IN THE DAY.

6765
04:20:05,667 --> 04:20:07,536
I KNOW IT'S THE START OF THE

6766
04:20:07,536 --> 04:20:09,271
SEMESTER FOR A LOT OF PEOPLE.

6767
04:20:09,271 --> 04:20:10,706
BUT WE'RE INVITING MORE WORK ON

6768
04:20:10,706 --> 04:20:17,179
YOUR PLATE SO NOW IS A GOOD TIME

6769
04:20:17,179 --> 04:20:19,348
TO SPEAK UP.

6770
04:20:19,348 --> 04:20:21,416
>> BEFORE WE AGREE TO BURY

6771
04:20:21,416 --> 04:20:22,784
OURSELVES IN MORE WORK, YOU

6772
04:20:22,784 --> 04:20:26,488
MENTIONED AS WELL IN THE CHAT,

6773
04:20:26,488 --> 04:20:28,190
THERE ARE ONGOING SYMPOSIA,

6774
04:20:28,190 --> 04:20:31,293
WORKSHOPS, CONFERENCES ON THE

6775
04:20:31,293 --> 04:20:32,060
AREA.

6776
04:20:32,060 --> 04:20:34,329
AGAIN, LET'S BE INTENTIONAL

6777
04:20:34,329 --> 04:20:36,331
ABOUT FINDING SOMETHING AROUND

6778
04:20:36,331 --> 04:20:38,400
THAT IS BOTH IMPORTANT FOR US

6779
04:20:38,400 --> 04:20:40,402
AND EVEN UNIQUE BUT NOT BEING

6780
04:20:40,402 --> 04:20:45,741
COVERED BY SOMEBODY ELSE SO WE

6781
04:20:45,741 --> 04:20:47,743
DON'T JUST EXPEND EXTRA ENERGY

6782
04:20:47,743 --> 04:20:48,844
TO NOT GREAT EFFECT.

6783
04:20:48,844 --> 04:20:50,245
THERE'S A LOT TO DO.

6784
04:20:50,245 --> 04:20:51,547
I'M SURE WE CAN FIND SOMETHING

6785
04:20:51,547 --> 04:20:53,081
PARTICULAR TO US TO ADD TO THE

6786
04:20:53,081 --> 04:20:59,421
CONVERSATION AND NOT REPEAT

6787
04:20:59,421 --> 04:20:59,788
OTHERS.

6788
04:20:59,788 --> 04:21:03,058
>> I'D LOVE TO HEAR FROM
OTHER,

6789
04:21:03,058 --> 04:21:10,265
INCLUDING SASKIA, BUT NOT ONLY

6790
04:21:10,265 --> 04:21:16,104
SASKIA, OTHER MEMBERS OF THE

6791
04:21:16,104 --> 04:21:18,807
NEWG, WRITING UPDATES, A

6792
04:21:18,807 --> 04:21:20,242
WORKSHOP, INVENTORY, WHICH ONES

6793
04:21:20,242 --> 04:21:28,016
ARE YOU INTERESTED IN IS MY

6794
04:21:28,016 --> 04:21:28,884
QUESTION.

6795
04:21:28,884 --> 04:21:33,922
>> SASKIA, YOU CAN CLARIFY
THAT

6796
04:21:33,922 --> 04:21:34,823
LOOK OR --

6797
04:21:34,823 --> 04:21:37,426
>> WELL, SO, I'M CURIOUS
WHAT

6798
04:21:37,426 --> 04:21:40,395
OTHERS ON THIS CALL THINK ABOUT

6799
04:21:40,395 --> 04:21:45,400
IT FEET LIKE THERE'S A LOT OF

6800
04:21:45,400 --> 04:21:46,501
DIFFERENT -- THIS IS A BROAD

6801
04:21:46,501 --> 04:21:47,869
SPACE, IN TERMS OF ETHICS

6802
04:21:47,869 --> 04:21:49,905
THERE'S A LOT OF TOPICS THAT

6803
04:21:49,905 --> 04:21:51,573
CAME UP.

6804
04:21:51,573 --> 04:21:55,043
I WONDER ABOUT PRIORITY SETTING,

6805
04:21:55,043 --> 04:21:56,478
AMONG ALL THE DIFFERENT

6806
04:21:56,478 --> 04:21:57,245
QUESTIONS THAT HAVE ARISEN

6807
04:21:57,245 --> 04:21:59,214
BECAUSE I LOVE THE IDEA OF A

6808
04:21:59,214 --> 04:22:01,750
WORKSHOP OR PAPER BUT I THINK IF

6809
04:22:01,750 --> 04:22:06,555
WE COULD THINK ABOUT RIGHT NOW

6810
04:22:06,555 --> 04:22:10,158
WHAT ARE THE MOST IMPORTANT

6811
04:22:10,158 --> 04:22:10,826
UNRESOLVED QUESTIONS, ESPECIALLY

6812
04:22:10,826 --> 04:22:13,128
AS THEY RELATE TO BRAIN IN THIS

6813
04:22:13,128 --> 04:22:15,564
SPACE, I THINK THAT WOULD BE

6814
04:22:15,564 --> 04:22:17,733
REALLY HELPFUL AND FOCUSING OUR

6815
04:22:17,733 --> 04:22:18,400
NEXT EFFORTS.

6816
04:22:18,400 --> 04:22:21,036
SO I DON'T HAVE ANSWERS BUT THAT

6817
04:22:21,036 --> 04:22:25,307
WAS WHAT I WAS WONDERING ABOUT.

6818
04:22:25,307 --> 04:22:28,443
>> IS THERE A WORKSHOP,
THE

6819
04:22:28,443 --> 04:22:29,444
POST-ANONYMITY WORLD, RIGHT?

6820
04:22:29,444 --> 04:22:35,017
I MEAN -- WE'RE GOING TO START

6821
04:22:35,017 --> 04:22:36,351
CALLING ON NEWG MEMBERS TO GET

6822
04:22:36,351 --> 04:22:38,754
YOU TO SPEAK UP AT THIS POINT.

6823
04:22:38,754 --> 04:22:41,156
WE'RE COMING SOON FOR THE

6824
04:22:41,156 --> 04:22:42,457
ROUNDTABLE BUT IF YOU'VE NOT

6825
04:22:42,457 --> 04:22:44,526
SPOKEN, WEIGH IN AND TELL US

6826
04:22:44,526 --> 04:22:45,627
WHAT YOU THINK.

6827
04:22:45,627 --> 04:22:47,162
IS THERE A PARTICULAR SLICE THAT

6828
04:22:47,162 --> 04:22:48,263
YOU'RE FOCUSING ON THAT YOU

6829
04:22:48,263 --> 04:22:52,200
THINK WE SHOULD BE FOCUSING ON.

6830
04:22:52,200 --> 04:23:02,711
I SEE JEN PUT IN THE CHAT BEST

6831
04:23:03,311 --> 04:23:05,580
PRACTICES, FROM INDUSTRY.

6832
04:23:05,580 --> 04:23:10,285
I AGREE IT WOULD BE HELPFUL,

6833
04:23:10,285 --> 04:23:11,787
OUTPUT DELIVERABLES, THAT'S ONE

6834
04:23:11,787 --> 04:23:15,824
OF THE DIFFERENCES BETWEEN

6835
04:23:15,824 --> 04:23:17,693
UPCOMING WORKSHOP TO PRESENT

6836
04:23:17,693 --> 04:23:22,364
MORE SCHOLARLY, FOCUS ON DIVE

6837
04:23:22,364 --> 04:23:23,699
INTO SCHOLARLY DETAILS VERSUS A

6838
04:23:23,699 --> 04:23:25,233
SPECIFIC GROUP OF DETAILS TO PUT

6839
04:23:25,233 --> 04:23:32,674
OUT, INFORMED BY THAT WORK, BUT

6840
04:23:32,674 --> 04:23:34,209
TRANSLATIONAL WORK.

6841
04:23:34,209 --> 04:23:34,643
AMY?

6842
04:23:34,643 --> 04:23:37,646
>> TRYING TO ANSWER
SASKIA'S

6843
04:23:37,646 --> 04:23:38,814
QUESTIONS, THERE'S A LOT OF

6844
04:23:38,814 --> 04:23:40,515
WORK, I WAS GOING TO SAY MY

6845
04:23:40,515 --> 04:23:43,285
FIRST REACTION ONE OF THE MOST

6846
04:23:43,285 --> 04:23:45,954
PRESSING ISSUES IS AROUND SORT

6847
04:23:45,954 --> 04:23:47,289
OF HEALTH EQUITY AND DISPARATE

6848
04:23:47,289 --> 04:23:48,056
SORT OF BURDEN.

6849
04:23:48,056 --> 04:23:49,591
BUT I ALSO THINK THERE'S A LOT

6850
04:23:49,591 --> 04:23:51,226
OF WORK BEING DONE IN THAT

6851
04:23:51,226 --> 04:23:51,560
SPACE.

6852
04:23:51,560 --> 04:23:53,528
ONE OF THE THINGS WE TALKED

6853
04:23:53,528 --> 04:23:54,930
ABOUT TODAY THAT I THINK MAYBE

6854
04:23:54,930 --> 04:23:56,598
THERE'S LESS WORK BEING DONE IN

6855
04:23:56,598 --> 04:23:58,867
IS AROUND LIKE DATA GOVERNANCE

6856
04:23:58,867 --> 04:23:59,668
AND RETURN OF RESULTS, AND, YOU

6857
04:23:59,668 --> 04:24:02,504
KNOW, SOME OF THE QUESTIONS

6858
04:24:02,504 --> 04:24:04,473
RAISED ABOUT, YOU KNOW, DUTY TO

6859
04:24:04,473 --> 04:24:05,774
REPORT, DUTY TO WARN, LIKE WHAT

6860
04:24:05,774 --> 04:24:08,176
WE DO WITH ALL OF THE DATA THAT

6861
04:24:08,176 --> 04:24:10,312
WE'RE COLLECTING, AND SO THAT'S

6862
04:24:10,312 --> 04:24:11,613
ONE SUGGESTION, I'M NOT -- MAYBE

6863
04:24:11,613 --> 04:24:14,116
THERE ARE PEOPLE OUT THERE WHO

6864
04:24:14,116 --> 04:24:16,051
ARE WORKING VERY DILIGENTLY ON

6865
04:24:16,051 --> 04:24:17,085
THOSE PROBLEMS BUT I'VE SEEN

6866
04:24:17,085 --> 04:24:19,688
LESS OF IT THAN I HAVE IN SORT

6867
04:24:19,688 --> 04:24:22,190
OF THE EQUITY AND

6868
04:24:22,190 --> 04:24:32,067
REPRESENTATIVENESS, BIAS SPACE. 

6869
04:24:32,067 --> 04:24:33,568
>> WALTER PLEASE JUMP IN.

6870
04:24:33,568 --> 04:24:37,506
>> YOU'RE MUTED, WALTER.

6871
04:24:37,506 --> 04:24:40,008
>> MY BACKGROUND, I WOULD
BE

6872
04:24:40,008 --> 04:24:41,643
MOST WORRIED THAT THE CONFUSION

6873
04:24:41,643 --> 04:24:45,147
AROUND THE CONSENT FORM IS GOING

6874
04:24:45,147 --> 04:24:46,248
TO EVENTUALLY STALL EVERYTHING

6875
04:24:46,248 --> 04:24:48,116
TO A HALT UNTIL THAT CAN BE

6876
04:24:48,116 --> 04:24:52,821
SORTED OUT.

6877
04:24:52,821 --> 04:24:57,826
SO I WOULD SAY IF WE COULD JUST

6878
04:24:57,826 --> 04:24:59,361
FOCUS ON GUIDELINES FOR CONSENT,

6879
04:24:59,361 --> 04:25:01,897
THAT COULD BE SEEN AS A

6880
04:25:01,897 --> 04:25:03,064
CONSENSUS AND PUBLISHES, THAT

6881
04:25:03,064 --> 04:25:04,866
IRBs COULD LOOK AT AND SAY,

6882
04:25:04,866 --> 04:25:07,035
OKAY, I CAN AGREE WITH THIS,

6883
04:25:07,035 --> 04:25:12,073
OTHERWISE I THINK THE IRBs ARE

6884
04:25:12,073 --> 04:25:13,275
GOING TO BE COMPLETELY CONFUSE

6885
04:25:13,275 --> 04:25:17,012
AND THROW UP THEIR HANDS.

6886
04:25:17,012 --> 04:25:19,514
>> COULD I ASK WHEN YOU
SAY THAT

6887
04:25:19,514 --> 04:25:22,450
YOU MEAN CONSENT FOR THE USE OF

6888
04:25:22,450 --> 04:25:26,288
A.I. WITH DATA, WITH NEURODATA?

6889
04:25:26,288 --> 04:25:26,621
>> CORRECT.

6890
04:25:26,621 --> 04:25:27,989
YEAH, WE'VE TOUCHED ON THAT IN

6891
04:25:27,989 --> 04:25:29,791
THE PAST BUT I THINK THAT'S

6892
04:25:29,791 --> 04:25:35,497
PROBABLY THE THING THAT I CAN

6893
04:25:35,497 --> 04:25:45,874
SEE HITTING US FIRST.

6894
04:25:46,908 --> 04:25:47,175
 

6895
04:25:47,175 --> 04:25:47,843
>> WELL, SEEING
CONVERSATION

6896
04:25:47,843 --> 04:25:52,214
SLOWING, IT MAY BE TIME TO

6897
04:25:52,214 --> 04:25:55,050
TRANSITION TO DISCUSSION OF

6898
04:25:55,050 --> 04:25:55,617
UPDATES.

6899
04:25:55,617 --> 04:25:57,052
HOPEFULLY THAT WILL GET EVERYONE

6900
04:25:57,052 --> 04:25:58,787
TALKING ABOUT WHAT'S ON THEIR

6901
04:25:58,787 --> 04:26:00,989
PLATES, WHAT WE SHOULD BE AWARE

6902
04:26:00,989 --> 04:26:02,290
OF, AND SOME SHARING.

6903
04:26:02,290 --> 04:26:03,925
DOES THAT MAKE SENSE?

6904
04:26:03,925 --> 04:26:04,693
ANYBODY HAVE ADDITIONAL THOUGHTS

6905
04:26:04,693 --> 04:26:06,461
BEFORE WE DO THAT?

6906
04:26:06,461 --> 04:26:06,661
OKAY.

6907
04:26:06,661 --> 04:26:08,296
LET'S TRANSITION TO THIS PART OF

6908
04:26:08,296 --> 04:26:12,667
THE CONVERSATION WHICH IS TO DO

6909
04:26:12,667 --> 04:26:14,035
ROUNDTABLE UPDATES.

6910
04:26:14,035 --> 04:26:15,837
>> SHOULD WE MODERATE SO
YOU CAN

6911
04:26:15,837 --> 04:26:16,037
SHARE?

6912
04:26:16,037 --> 04:26:18,373
>> I WAS GOING TO HAND IT
OVER

6913
04:26:18,373 --> 04:26:19,908
TO CHRISTINE TO MODERATE.

6914
04:26:19,908 --> 04:26:21,443
OR JOHN, YOU CAN MODERATE,

6915
04:26:21,443 --> 04:26:22,844
WHOEVER WANTS TO MODERATE.

6916
04:26:22,844 --> 04:26:24,813
YES, BY ALL MEANS, SOMEBODY TAKE

6917
04:26:24,813 --> 04:26:27,349
OVER. 

6918
04:26:27,349 --> 04:26:28,650
>> I WANTED TO HEAR FROM

6919
04:26:28,650 --> 04:26:30,185
CHRISTINE TOO.

6920
04:26:30,185 --> 04:26:32,053
IS THAT OKAY?

6921
04:26:32,053 --> 04:26:33,588
>> SURE, ABSOLUTELY.

6922
04:26:33,588 --> 04:26:36,424
>> TWO, FOUR, SIX, EIGHT,
TEN

6923
04:26:36,424 --> 04:26:36,725
FOLKS.

6924
04:26:36,725 --> 04:26:37,726
IS WINSTON HERE?

6925
04:26:37,726 --> 04:26:39,261
WE'RE MOSTLY STILL HERE.

6926
04:26:39,261 --> 04:26:40,795
I'M GOING TO GO DOWN THE LIST,

6927
04:26:40,795 --> 04:26:42,530
I'LL FINISH WITH THE CO-CHAIRS,

6928
04:26:42,530 --> 04:26:44,199
HOW ABOUT THAT?

6929
04:26:44,199 --> 04:26:45,066
>> SOUNDS GREAT.

6930
04:26:45,066 --> 04:26:48,236
>> WINSTON, I DON'T SEE
YOU

6931
04:26:48,236 --> 04:26:48,570
HERE.

6932
04:26:48,570 --> 04:26:48,904
>> I'M HERE.

6933
04:26:48,904 --> 04:26:50,205
>> THERE YOU ARE.

6934
04:26:50,205 --> 04:26:53,708
YEAH, YEAH, ARE YOU READY?

6935
04:26:53,708 --> 04:26:58,313
>> COME BACK TO ME.

6936
04:26:58,313 --> 04:26:59,080
>> THERE'S LIKE
EIGHT-PLUS,

6937
04:26:59,080 --> 04:27:02,784
MAYBE TWO MINUTES EACH, THAT

6938
04:27:02,784 --> 04:27:04,552
WOULD BE GREAT.

6939
04:27:04,552 --> 04:27:10,225
JEN, DO YOU HAVE A FEW THINGS TO

6940
04:27:10,225 --> 04:27:13,528
SHARE WITH US? 

6941
04:27:13,528 --> 04:27:15,163
>> MAYBE NOT.

6942
04:27:15,163 --> 04:27:16,031
>> I DO, YES.

6943
04:27:16,031 --> 04:27:17,999
SORRY ABOUT THAT.

6944
04:27:17,999 --> 04:27:19,434
SORRY FOR THE DELAY.

6945
04:27:19,434 --> 04:27:21,603
YES, ACTUALLY IF YOU DON'T MIND,

6946
04:27:21,603 --> 04:27:25,674
WE'D LOVE TO GIVE A QUICK

6947
04:27:25,674 --> 04:27:27,976
UPDATE. 

6948
04:27:27,976 --> 04:27:30,178
WE LAUNCHED THE IMPLANTABLE

6949
04:27:30,178 --> 04:27:30,679
BRAIN-COMPUTER INTERFACE

6950
04:27:30,679 --> 04:27:32,147
COLLABORATIVE COMMUNITY IN

6951
04:27:32,147 --> 04:27:33,481
MARCH, WE'RE UP TO 200 MEMBERS

6952
04:27:33,481 --> 04:27:33,648
NOW.

6953
04:27:33,648 --> 04:27:36,017
MANY OF THOSE ON THE CALL HERE

6954
04:27:36,017 --> 04:27:38,186
HAVE JOINED SOME OF OUR WORKING

6955
04:27:38,186 --> 04:27:38,420
GROUPS.

6956
04:27:38,420 --> 04:27:40,455
WE ENCOURAGE MORE TO JOIN.

6957
04:27:40,455 --> 04:27:43,425
WE DO HAVE A WORK GROUP

6958
04:27:43,425 --> 04:27:44,759
DEDICATED TO ETHICS, NEURAL

6959
04:27:44,759 --> 04:27:47,395
DATA, NEURAL SECURITY.

6960
04:27:47,395 --> 04:27:49,431
I KNOW THAT ETHICS CAN GO INTO

6961
04:27:49,431 --> 04:27:52,067
AREAS OF BRAIN AND COMPUTER

6962
04:27:52,067 --> 04:27:53,935
INTERFACE, BUT I THINK THE

6963
04:27:53,935 --> 04:27:55,503
LOW-HANGING FRUIT WHERE WE ARE

6964
04:27:55,503 --> 04:27:58,006
RIGHT NOW IS IN TERMS OF DATA

6965
04:27:58,006 --> 04:28:00,842
AND SECURITY AT THIS POINT.

6966
04:28:00,842 --> 04:28:02,944
IT'S OPEN TO THE PUBLIC TO JOIN.

6967
04:28:02,944 --> 04:28:04,145
WE'RE LOOKING FOR A LOT OF

6968
04:28:04,145 --> 04:28:06,448
DIVERSE MEMBERS TO JOIN OUR

6969
04:28:06,448 --> 04:28:08,516
CONVERSATION AND REALLY HAVING

6970
04:28:08,516 --> 04:28:10,752
OUTPUTS. 

6971
04:28:10,752 --> 04:28:13,121
KEEP IN MIND WE DEFINE IMPLANT

6972
04:28:13,121 --> 04:28:14,189
AND INTERFACES MORE NARROWLY

6973
04:28:14,189 --> 04:28:16,091
THAN A LOT OF PEOPLE DO.

6974
04:28:16,091 --> 04:28:17,192
WE WERE JUST LOOKING AT THOSE

6975
04:28:17,192 --> 04:28:19,361
SENSING FROM THE BRAIN AT THIS

6976
04:28:19,361 --> 04:28:20,462
POINT, EVENTUALLY WE'LL BE ABLE

6977
04:28:20,462 --> 04:28:22,430
TO BROADEN THAT OUT BUT RIGHT

6978
04:28:22,430 --> 04:28:25,367
NOW IT'S JUST FOCUSED ON THAT.

6979
04:28:25,367 --> 04:28:27,469
FOR THOSE THAT MIGHT NOT BE

6980
04:28:27,469 --> 04:28:29,104
PRESENTED WITH THE COLLABORATIVE

6981
04:28:29,104 --> 04:28:32,073
COMMUNITY, IT IS AN EFFORT TO

6982
04:28:32,073 --> 04:28:33,608
BRING TOGETHER DIVERSE

6983
04:28:33,608 --> 04:28:35,910
STAKEHOLDERS TO SOLVE SYSTEMIC

6984
04:28:35,910 --> 04:28:36,144
ISSUES.

6985
04:28:36,144 --> 04:28:42,584
WE WENT THROUGH A YEAR PROCESS

6986
04:28:42,584 --> 04:28:44,219
OF DEVELOPING THE COMMUNITY AND

6987
04:28:44,219 --> 04:28:45,854
LIAISON TO BRING IN FEDERAL

6988
04:28:45,854 --> 04:28:47,389
PARTNERS TO SPEAK ACTIVELY AND

6989
04:28:47,389 --> 04:28:49,791
WORK ACTIVELY WITHIN OUR WORK

6990
04:28:49,791 --> 04:28:52,193
GROUPS ALONG WITH THOSE FROM

6991
04:28:52,193 --> 04:28:56,264
INDUSTRY, RESEARCH, PATIENT

6992
04:28:56,264 --> 04:28:57,332
ADVOCACY ORGANIZATION, AND

6993
04:28:57,332 --> 04:28:57,565
OTHERS.

6994
04:28:57,565 --> 04:28:59,634
>> YOUR KICKOFF WORKSHOP
WILL BE

6995
04:28:59,634 --> 04:29:01,703
THIS SEPTEMBER 18, CORRECT?

6996
04:29:01,703 --> 04:29:02,270
CHEVY CHASE?

6997
04:29:02,270 --> 04:29:03,571
>> THAT'S CORRECT.

6998
04:29:03,571 --> 04:29:06,307
IT WILL BE IN SILVER SPRING.

6999
04:29:06,307 --> 04:29:07,942
>> I'M SORRY, SILVER
SPRING.

7000
04:29:07,942 --> 04:29:08,209
>> YEP.

7001
04:29:08,209 --> 04:29:09,144
AGAIN, THAT'S OPEN TO THE

7002
04:29:09,144 --> 04:29:11,012
PUBLIC, SO I'M GOING TO PUT OUR

7003
04:29:11,012 --> 04:29:12,881
WEBSITE INTO THE CHAT WHERE YOU

7004
04:29:12,881 --> 04:29:15,683
CAN LEARN MORE AND BE ABLE TO

7005
04:29:15,683 --> 04:29:19,888
REGISTER INURE EVENT.  EVENT --T

7006
04:29:19,888 --> 04:29:20,288
EVENT.

7007
04:29:20,288 --> 04:29:21,623
WE'RE REQUIRED TO HAVE PUBLIC

7008
04:29:21,623 --> 04:29:22,824
MEETINGS SO WE'RE WELCOMING YOU

7009
04:29:22,824 --> 04:29:25,560
ALL TO JOIN, AND JOIN OUR

7010
04:29:25,560 --> 04:29:27,462
CONVERSATIONS, AND REALLY WE'RE

7011
04:29:27,462 --> 04:29:30,732
FOCUSED NOT ON JUST DISCUSSIONS

7012
04:29:30,732 --> 04:29:35,537
BUT BE ABLE TO HAVE OUTPUTS THAT

7013
04:29:35,537 --> 04:29:38,173
CAN REALLY BRING IMPLANTED

7014
04:29:38,173 --> 04:29:40,742
BRAIN-COMPUTER INTERFACES

7015
04:29:40,742 --> 04:29:43,545
AVAILABLE IN A SAFE AND

7016
04:29:43,545 --> 04:29:44,079
EFFICACIOUS WAY.

7017
04:29:44,079 --> 04:29:51,086
>> LOOKING FORWARD TO
THAT.

7018
04:29:51,086 --> 04:29:51,853
OKAY, AMY?

7019
04:29:51,853 --> 04:29:54,022
>> I DON'T HAVE SPECIFIC

7020
04:29:54,022 --> 04:29:55,990
UPDATES, I DON'T THINK.

7021
04:29:55,990 --> 04:30:02,464
SORRY, ARE YOU CALLING ON ME?

7022
04:30:02,464 --> 04:30:03,765
YES.

7023
04:30:03,765 --> 04:30:04,866
>> YES, SORRY, AMY
MCGUIRE.

7024
04:30:04,866 --> 04:30:06,668
>> I DON'T HAVE SPECIFIC
UPDATES

7025
04:30:06,668 --> 04:30:08,036
TO SHARE WITH THE GROUP.

7026
04:30:08,036 --> 04:30:09,437
>> OKAY, GREAT.

7027
04:30:09,437 --> 04:30:19,881
APPRECIATE YOU BEING HERE.

7028
04:30:20,248 --> 04:30:21,149
CAROLINE MONTOJO.

7029
04:30:21,149 --> 04:30:22,917
>> WE'VE HAD A FEW UPDATES
THEY

7030
04:30:22,917 --> 04:30:24,419
ASKED ME TO SHARE SINCE IT'S

7031
04:30:24,419 --> 04:30:27,722
RELEVANT TO OUR WORK AT THE DANA

7032
04:30:27,722 --> 04:30:28,056
FOUNDATION.

7033
04:30:28,056 --> 04:30:30,358
I'LL GO THROUGH THESE

7034
04:30:30,358 --> 04:30:31,893
EFFICIENTLY BUT ALSO SHARING

7035
04:30:31,893 --> 04:30:34,629
SOME LINKS IN THE CHAT SO PEOPLE

7036
04:30:34,629 --> 04:30:36,297
CAN HAVE MORE CONTEXT.

7037
04:30:36,297 --> 04:30:41,002
FIRST IS THAT A NEW EPISODE OF

7038
04:30:41,002 --> 04:30:45,940
NEUROSOCIETY STORIES WAS

7039
04:30:45,940 --> 04:30:47,375
RELEASED YET, FRANCIS CHEN,

7040
04:30:47,375 --> 04:30:53,581
PRODUCED BY DANA FOUNDATION,

7041
04:30:53,581 --> 04:30:56,317
HIGHLIGHTS CONNECTION THROUGH

7042
04:30:56,317 --> 04:30:57,852
GRANTEES, PARTNERS, OUTSIDE THE

7043
04:30:57,852 --> 04:30:58,853
DANA FOUNDATION NETWORK

7044
04:30:58,853 --> 04:31:00,688
DEDICATED TO CONNECTING BRAIN

7045
04:31:00,688 --> 04:31:02,991
SCIENCE TO SOCIETY.

7046
04:31:02,991 --> 04:31:06,728
I'LL DROP IN THE LINK TO THAT

7047
04:31:06,728 --> 04:31:07,028
HERE.

7048
04:31:07,028 --> 04:31:09,898
SO FYI A FEW LINKS COMING.

7049
04:31:09,898 --> 04:31:12,300
SECOND UPDATE IS ABOUT THE DANA

7050
04:31:12,300 --> 04:31:14,869
CAREER NETWORK IN NEUROSCIENCE

7051
04:31:14,869 --> 04:31:16,137
AND SOCIETY, THIS IS

7052
04:31:16,137 --> 04:31:23,244
INTERESTINGLY ALSO LED BY

7053
04:31:23,244 --> 04:31:24,345
FRANCIS CHEN, ORGANIZING A

7054
04:31:24,345 --> 04:31:27,315
SECOND CAREER FAIR THAT'S TAKING

7055
04:31:27,315 --> 04:31:30,485
PLACE SEPTEMBER 9-12,  SO IF YOU

7056
04:31:30,485 --> 04:31:39,761
HAVE FOLKS IN YOUR LAB, TRAINEES

7057
04:31:39,761 --> 04:31:41,529
THAT MIGHT BE ATTENDED, THERE

7058
04:31:41,529 --> 04:31:43,198
YOU GO.

7059
04:31:43,198 --> 04:31:44,299
IT'S COMING.

7060
04:31:44,299 --> 04:31:48,136
THAT'S FOR THE CAREER FAIR.

7061
04:31:48,136 --> 04:31:50,972
AND THEN ANOTHER SHOUT OUT FOR A

7062
04:31:50,972 --> 04:31:52,407
HASTINGS CENTER SERIES OF

7063
04:31:52,407 --> 04:31:57,979
ARTICLES AND ESSAYS ON ETHICAL,

7064
04:31:57,979 --> 04:32:00,815
LEGAL, SOCIAL ISSUES PRESENTED

7065
04:32:00,815 --> 04:32:01,382
BY IMAGING NEUROSCIENCE,

7066
04:32:01,382 --> 04:32:04,652
PUBLISHED A FEW ARTICLES IN THAT

7067
04:32:04,652 --> 04:32:08,590
SERIES THROUGH THE HASTING

7068
04:32:08,590 --> 04:32:09,791
CENTER REPORT, A SHOUT OUT TO

7069
04:32:09,791 --> 04:32:14,729
THOSE ON THE STEERING COMMITTEE,

7070
04:32:14,729 --> 04:32:16,364
JENNIFER, WINSTON, JOE, SARAH,

7071
04:32:16,364 --> 04:32:17,665
JONATHAN AND OLIVER FOR

7072
04:32:17,665 --> 04:32:19,734
PARTICIPATING IN PLANNING OF

7073
04:32:19,734 --> 04:32:20,068
THAT.

7074
04:32:20,068 --> 04:32:23,771
AND THEN THE OTHER PIECE IS THAT

7075
04:32:23,771 --> 04:32:26,741
THE U.N. GENERAL ASSEMBLY THIS

7076
04:32:26,741 --> 04:32:29,110
COMING SEPTEMBER THE THEME OF

7077
04:32:29,110 --> 04:32:33,014
THAT ASSEMBLY EVENT WILL BE ON

7078
04:32:33,014 --> 04:32:34,549
NEUROSCIENCE IN SOCIETY, A SIDE

7079
04:32:34,549 --> 04:32:37,652
OF THE U.N. GENERAL ASSEMBLY, ON

7080
04:32:37,652 --> 04:32:39,687
SEPTEMBER 20th THE DANA

7081
04:32:39,687 --> 04:32:42,090
FOUNDATION WILL HOST A SPECIAL

7082
04:32:42,090 --> 04:32:43,091
PANEL ON APPLYING NEUROSCIENCE

7083
04:32:43,091 --> 04:32:44,726
IN THE COURTROOM, SO WE'RE

7084
04:32:44,726 --> 04:32:48,363
FORTUNATE TO HAVE A FEW

7085
04:32:48,363 --> 04:32:49,230
PARTICIPANTS, DEBRA DENNO FROM

7086
04:32:49,230 --> 04:32:55,703
FORDHAM UNIVERSITY, DIEGO

7087
04:32:55,703 --> 04:32:56,871
RODRIGUEZ FROM COLOMBIA, OLIVER

7088
04:32:56,871 --> 04:33:00,408
ROLLINS FROM M.I.T. AND JUDGE

7089
04:33:00,408 --> 04:33:02,510
GLORIA TAN FROM MASSACHUSETTS

7090
04:33:02,510 --> 04:33:03,678
JUVENILE COURT.

7091
04:33:03,678 --> 04:33:06,748
AND FOR THIS EVENT IT'S OPEN

7092
04:33:06,748 --> 04:33:09,050
BOTH IN PERSON AS WELL AS

7093
04:33:09,050 --> 04:33:09,717
VIRTUAL.

7094
04:33:09,717 --> 04:33:12,220
SO IF ANYBODY IS INTERESTED BUT

7095
04:33:12,220 --> 04:33:13,721
CAN'T ATTEND IN PERSON IN NEW

7096
04:33:13,721 --> 04:33:15,189
YORK CITY THERE WILL BE A

7097
04:33:15,189 --> 04:33:17,125
VIRTUAL OPTION.

7098
04:33:17,125 --> 04:33:21,963
THE LAST PIECE IS THAT SOCIETY

7099
04:33:21,963 --> 04:33:23,331
FOR NEUROSCIENCE THIS COMING

7100
04:33:23,331 --> 04:33:26,768
OCTOBER THE DANA FOUNDATION WILL

7101
04:33:26,768 --> 04:33:28,503
BE SPONSORING THE DIALOGUE

7102
04:33:28,503 --> 04:33:29,704
BETWEEN NEUROSCIENCE AND SOCIETY

7103
04:33:29,704 --> 04:33:32,240
SO THIS IS AN EVENT THAT HAS

7104
04:33:32,240 --> 04:33:34,409
TAKEN PLACE PREVIOUSLY AT FSN,

7105
04:33:34,409 --> 04:33:36,844
AND THEY WENT TO A HIATUS

7106
04:33:36,844 --> 04:33:39,013
PERIOD, NOW RESTARTING AGAIN

7107
04:33:39,013 --> 04:33:39,681
THIS COMING OCTOBER.

7108
04:33:39,681 --> 04:33:41,849
SO IT WILL BE A PLENARY SESSION

7109
04:33:41,849 --> 04:33:44,485
SO HOPEFULLY YOU CAN MAKE IT TO

7110
04:33:44,485 --> 04:33:45,486
THAT.

7111
04:33:45,486 --> 04:33:46,054
THAT'S IT.

7112
04:33:46,054 --> 04:33:47,021
>> THANKS, CAROLINE.

7113
04:33:47,021 --> 04:33:49,023
YOU FOLKS HAVE BEEN PRODUCTIVE

7114
04:33:49,023 --> 04:33:50,425
AND BUSY.

7115
04:33:50,425 --> 04:33:52,527
THAT'S GREAT.

7116
04:33:52,527 --> 04:33:53,861
THANKS SO MUCH.

7117
04:33:53,861 --> 04:33:59,100
SAMEER, ARE YOU STILL WITH US? 

7118
04:33:59,100 --> 04:34:01,202
MAYBE NOT.

7119
04:34:01,202 --> 04:34:03,404
WE'LL COME BACK.

7120
04:34:03,404 --> 04:34:03,638
JEN?

7121
04:34:03,638 --> 04:34:06,040
WHAT'S NEW ON YOUR END?

7122
04:34:06,040 --> 04:34:08,543
>> THANKS, JOHN.

7123
04:34:08,543 --> 04:34:11,512
THERE'S NO NEUROETHICS UPDATE ON

7124
04:34:11,512 --> 04:34:13,247
MY PART.

7125
04:34:13,247 --> 04:34:13,514
>> OKAY.

7126
04:34:13,514 --> 04:34:17,618
WE CAN HAVE OTHER DISCUSSIONS AT

7127
04:34:17,618 --> 04:34:19,287
ANOTHER TIME.

7128
04:34:19,287 --> 04:34:21,255
SID, WHAT'S NEW?

7129
04:34:21,255 --> 04:34:23,324
>> WELL, IT'S BEEN A LONG
TIME

7130
04:34:23,324 --> 04:34:23,991
SINCE WE'VE MET SO I DON'T

7131
04:34:23,991 --> 04:34:25,893
REMEMBER WHAT I TOLD YOU ABOUT

7132
04:34:25,893 --> 04:34:26,494
LAST TIME.

7133
04:34:26,494 --> 04:34:30,131
BUT I HAVE A LITTLE BOOK ON

7134
04:34:30,131 --> 04:34:31,699
PHILOSOPHICAL MEDICAL AND LEGAL

7135
04:34:31,699 --> 04:34:35,069
CONTROVERSIES ABOUT BRAIN DEATH

7136
04:34:35,069 --> 04:34:37,605
THAT CAME OUT LAST WINTER.

7137
04:34:37,605 --> 04:34:41,743
ANOTHER ONE COMING OUT VERY SOON

7138
04:34:41,743 --> 04:34:43,544
ON RESEARCH WITH NON-HUMAN

7139
04:34:43,544 --> 04:34:43,945
PRIMATES.

7140
04:34:43,945 --> 04:34:49,183
I HAVE A PAPER IN THE JOURNAL OF

7141
04:34:49,183 --> 04:34:49,851
COGNITIVE NEUROSCIENCE WHICH

7142
04:34:49,851 --> 04:34:52,487
DEVELOPED OUT OF A TALK I GAVE

7143
04:34:52,487 --> 04:34:54,789
AT AN NIH FRONTIERS OF

7144
04:34:54,789 --> 04:34:57,625
CONSCIOUSNESS WORKSHOP FROM LAST

7145
04:34:57,625 --> 04:35:00,695
SUMMER, AND THAT'S ON ENTITIES,

7146
04:35:00,695 --> 04:35:02,663
UNCERTAINTIES, BEHAVIORAL

7147
04:35:02,663 --> 04:35:04,298
INDICATIONS OF CONSCIOUSNESS, IN

7148
04:35:04,298 --> 04:35:10,071
HUMANS, ANIMALS, AND A.I.

7149
04:35:10,071 --> 04:35:12,840
AND, OH, I'VE BEEN GIVING

7150
04:35:12,840 --> 04:35:14,242
INTERVIEWS ABOUT NEURAL LINK

7151
04:35:14,242 --> 04:35:17,245
EVERY TIME THEY IMPLANT A NEW

7152
04:35:17,245 --> 04:35:21,349
PATIENT, AND THERE'S A LOT OF

7153
04:35:21,349 --> 04:35:22,683
HYPE COMING OUT ABOUT THEIR

7154
04:35:22,683 --> 04:35:25,420
BRAIN IMPLANT SO I'VE BEEN

7155
04:35:25,420 --> 04:35:28,689
TALKING ABOUT PATIENT SAFETY

7156
04:35:28,689 --> 04:35:28,923
ISSUES.

7157
04:35:28,923 --> 04:35:30,458
VERY MUCH INFORMED BY THINGS

7158
04:35:30,458 --> 04:35:35,496
THAT I HAVE LEARNED IN NEWG

7159
04:35:35,496 --> 04:35:37,365
WORKSHOPS AND MEETINGS ON BRAIN

7160
04:35:37,365 --> 04:35:38,466
IMPLANTS SO HOPEFULLY ABLE TO

7161
04:35:38,466 --> 04:35:41,636
SPREAD SOME OF THAT INFORMATION.

7162
04:35:41,636 --> 04:35:42,036
>> TERRIFIC.

7163
04:35:42,036 --> 04:35:43,938
PLEASE LET US KNOW WHEN --

7164
04:35:43,938 --> 04:35:45,139
CONGRATULATIONS ON THE SECOND

7165
04:35:45,139 --> 04:35:47,909
BOOK, I THINK LAST TIME WE MET

7166
04:35:47,909 --> 04:35:49,744
THE PREVIOUS BOOK HAD JUST BEEN

7167
04:35:49,744 --> 04:35:52,046
RELEASED BUT LET US KNOW WHEN

7168
04:35:52,046 --> 04:35:54,015
THE SECOND BOOK COMES OUT.

7169
04:35:54,015 --> 04:35:55,650
WE'D LOVE TO LEARN MORE ABOUT

7170
04:35:55,650 --> 04:35:56,751
THAT.

7171
04:35:56,751 --> 04:36:02,223
>> WILL DO.

7172
04:36:02,223 --> 04:36:03,524
>> TERRIFIC.

7173
04:36:03,524 --> 04:36:03,858
KAREN?

7174
04:36:03,858 --> 04:36:04,659
HI.

7175
04:36:04,659 --> 04:36:07,161
YES, I WAS TRYING TO REMEMBER

7176
04:36:07,161 --> 04:36:08,896
WHEN WE LAST MET, LIKE SYD.

7177
04:36:08,896 --> 04:36:10,097
I THINK IT WAS FEBRUARY.

7178
04:36:10,097 --> 04:36:13,835
I WANTED TO SHARE ONE PAST EVENT

7179
04:36:13,835 --> 04:36:14,902
FROM JUNE, ACTUALLY.

7180
04:36:14,902 --> 04:36:20,842
JOHN, I SAW YOU THERE.

7181
04:36:20,842 --> 04:36:24,879
WE HOSTED THE INAUGURAL

7182
04:36:24,879 --> 04:36:26,214
NEUROETHICS HACK-A-THON,

7183
04:36:26,214 --> 04:36:27,315
PROBABLY THE FIRST, OFFICIALLY

7184
04:36:27,315 --> 04:36:29,383
DUBBED AT SUCH.

7185
04:36:29,383 --> 04:36:33,955
THE FOCUS ON NEURO-A.I. WHERE WE

7186
04:36:33,955 --> 04:36:36,157
INVITED -- WE CAPPED AT 80 FOR

7187
04:36:36,157 --> 04:36:40,194
REGISTRATION TO COME IN AND WORK

7188
04:36:40,194 --> 04:36:46,000
ON NEAR TERM FICTITIOUS SLIGHTLY

7189
04:36:46,000 --> 04:36:50,805
FUTURE-LOOKING NEURO-A.I.

7190
04:36:50,805 --> 04:36:52,907
SCENARIOS, WORKING OUT

7191
04:36:52,907 --> 04:36:53,174
SOLUTIONS.

7192
04:36:53,174 --> 04:36:53,975
TECHNICAL HACK-A-THONS LOOKING

7193
04:36:53,975 --> 04:36:55,510
TO SOLVE A TECHNICAL PROBLEM,

7194
04:36:55,510 --> 04:36:57,712
HERE ALSO ASKING THEM TO DESIGN

7195
04:36:57,712 --> 04:37:00,648
SOCIAL SOLUTIONS AND THESE COULD

7196
04:37:00,648 --> 04:37:03,150
BE POLICY PIECES, COULD BE

7197
04:37:03,150 --> 04:37:03,417
WHATEVER.

7198
04:37:03,417 --> 04:37:05,486
WE HAD THIS METHODOLOGY WE

7199
04:37:05,486 --> 04:37:09,223
UTILIZE AND PILOTED THERE.

7200
04:37:09,223 --> 04:37:11,359
ANYWAY, A PUBLICATION WILL BE

7201
04:37:11,359 --> 04:37:12,560
FORTHCOMING THIS FALL.

7202
04:37:12,560 --> 04:37:14,762
I BRING THAT UP WE FOUND IN

7203
04:37:14,762 --> 04:37:19,367
REVIEWS OF THE EVENT PEOPLE,

7204
04:37:19,367 --> 04:37:21,836
THIS IS LARGELY POSTDOCS,

7205
04:37:21,836 --> 04:37:23,738
GRADUATE STUDENTS, MAYBE A FEW

7206
04:37:23,738 --> 04:37:26,240
FACULTY SPRINKLED IN, LARGELY

7207
04:37:26,240 --> 04:37:27,575
ACADEMIC COMMUNITY, BUT THEY

7208
04:37:27,575 --> 04:37:29,010
REALLY ENJOYED THE EXERCISE OF

7209
04:37:29,010 --> 04:37:34,148
WORKING THROUGH A REAL CASE, AND

7210
04:37:34,148 --> 04:37:43,491
IT MADE ETHICS TANGIBLE.

7211
04:37:43,491 --> 04:37:44,792
IF WE WANTED TO DO SOMETHING

7212
04:37:44,792 --> 04:37:46,894
WITH YOUR NEURAL A.I. EFFORT THE

7213
04:37:46,894 --> 04:37:48,996
CASE STUDIES WORKS AS A WAY TO

7214
04:37:48,996 --> 04:37:51,432
COME UP WITH TANGIBLE EXAMPLES

7215
04:37:51,432 --> 04:37:56,604
AND I LOVE ACTUALLY SYD'S POINT

7216
04:37:56,604 --> 04:37:58,005
ABOUT US UTILIZING TOOLS.

7217
04:37:58,005 --> 04:38:01,275
WE HAVEN'T DONE MUCH WITH THE

7218
04:38:01,275 --> 04:38:02,009
NEUROETHICS ROAD MAP.

7219
04:38:02,009 --> 04:38:03,578
THANKS TO JIM FOR HIS STRONG

7220
04:38:03,578 --> 04:38:08,282
LEADERSHIP ON THAT. 

7221
04:38:08,282 --> 04:38:11,252
WE HAVE SOME PIECES, WE COULD

7222
04:38:11,252 --> 04:38:13,220
ADD THINGS, ADD CASE STUDIES TO

7223
04:38:13,220 --> 04:38:14,755
BUILD TANGIBLE SOLUTIONS.

7224
04:38:14,755 --> 04:38:16,624
TACKLING CONSENT IS ONE THING

7225
04:38:16,624 --> 04:38:17,491
BUT THERE'S SOME REALLY DAILY

7226
04:38:17,491 --> 04:38:20,561
THINGS THAT COULD BE APPROACHED

7227
04:38:20,561 --> 04:38:26,067
IN THE RESEARCH PROCESS. 

7228
04:38:26,067 --> 04:38:32,707
AND SECOND THING, CAR OLYN PUT

7229
04:38:32,707 --> 04:38:34,675
THE LINK IN, BRAIN HEALTH,

7230
04:38:34,675 --> 04:38:37,178
SHARING A PANEL, INSTITUTE OF

7231
04:38:37,178 --> 04:38:39,714
THE NEUROETHICS WITH OUR THINK

7232
04:38:39,714 --> 04:38:41,449
AND DO TANK PARTNERING WITH

7233
04:38:41,449 --> 04:38:44,185
INTERNATIONAL CENTER FOR FUTURE

7234
04:38:44,185 --> 04:38:45,286
GENERATIONS AND WE'VE BEEN

7235
04:38:45,286 --> 04:38:48,155
DEVELOPING STRATEGY FOR THE E.U.

7236
04:38:48,155 --> 04:38:49,890
ON PARTICIPATORY GOVERNANCE IN

7237
04:38:49,890 --> 04:38:53,728
NEUROTECHNOLOGY, THERE'S A LOT

7238
04:38:53,728 --> 04:38:55,363
OF NEUROTECHNOLOGY GOVERNANCE

7239
04:38:55,363 --> 04:38:56,430
CONVERSATIONS, EXCELLENT ONES,

7240
04:38:56,430 --> 04:38:58,899
LARGELY DEVOID OF PUBLIC

7241
04:38:58,899 --> 04:38:59,533
ENGAGEMENT, MEANINGFUL PUBLIC

7242
04:38:59,533 --> 04:39:00,968
ENGAGEMENT OPPORTUNITIES SO

7243
04:39:00,968 --> 04:39:02,269
WE'RE TRYING TO PILOT A FEW

7244
04:39:02,269 --> 04:39:04,438
METHODS TO SHARE AND PUT FORWARD

7245
04:39:04,438 --> 04:39:05,973
A STRATEGY LOOKING FORWARD TO

7246
04:39:05,973 --> 04:39:08,609
THE NEXT FIVE-YEAR AGENDA FOR

7247
04:39:08,609 --> 04:39:10,778
THE E.U. BUT HOPE THESE ARE

7248
04:39:10,778 --> 04:39:13,014
HISTORICAL MODELS TO BE USED

7249
04:39:13,014 --> 04:39:14,515
GLOBALLY.

7250
04:39:14,515 --> 04:39:17,685
>> THANKS VERY MUCH,
KAREN.

7251
04:39:17,685 --> 04:39:19,687
GREAT ACTIVITIES THERE.

7252
04:39:19,687 --> 04:39:21,422
WINSTON, ARE YOU BACK?

7253
04:39:21,422 --> 04:39:22,089
>> YES.

7254
04:39:22,089 --> 04:39:22,623
>> COOL.

7255
04:39:22,623 --> 04:39:28,529
>> SORRY, I HAD TO POP OUT
WITH

7256
04:39:28,529 --> 04:39:29,764
A CONFLICTING MEETING.

7257
04:39:29,764 --> 04:39:32,166
ONE THING WE'RE ALMOST READY TO

7258
04:39:32,166 --> 04:39:34,368
FULLY ANNOUNCE, I CAN GIVE A

7259
04:39:34,368 --> 04:39:36,771
PREVIEW HERE, WITHIN THE

7260
04:39:36,771 --> 04:39:37,304
INTERNATIONAL NEUROETHICS

7261
04:39:37,304 --> 04:39:39,507
SOCIETY WE'VE BEEN INTERESTED IN

7262
04:39:39,507 --> 04:39:41,375
BUILDING SOME BRIDGES BETWEEN

7263
04:39:41,375 --> 04:39:42,476
NEUROETHICS AND CLINICAL

7264
04:39:42,476 --> 04:39:50,151
COMMUNITY SO FOLKS IN NEUROLOGY,

7265
04:39:50,151 --> 04:39:50,718
PSYCHIATRY, NEUROPSYCHOLOGY,

7266
04:39:50,718 --> 04:39:51,652
NEUROSURGERY, ADDRESSING THE

7267
04:39:51,652 --> 04:39:53,988
BRAIN AND NERVOUS SYSTEM, SO

7268
04:39:53,988 --> 04:39:55,923
WE'RE BUILDING A CLINICAL

7269
04:39:55,923 --> 04:39:57,258
NEUROETHICS AFFINITY GROUP,

7270
04:39:57,258 --> 04:39:58,426
PLANNING AN EVENT THAT WILL BE

7271
04:39:58,426 --> 04:40:00,861
RELATED TO CAREER AROUND FUNDING

7272
04:40:00,861 --> 04:40:02,396
FOR PEOPLE FROM CLINICAL

7273
04:40:02,396 --> 04:40:04,165
BACKGROUNDS INTERESTED IN THE

7274
04:40:04,165 --> 04:40:08,202
NEUROETHICS AND ETHICS SPACE.

7275
04:40:08,202 --> 04:40:12,707
I'LL LINK UP WITH YOU ABOUT THE

7276
04:40:12,707 --> 04:40:13,574
CAREER NETWORK, LOOKING AT

7277
04:40:13,574 --> 04:40:15,309
EVENTS IN THE FUTURE SOME

7278
04:40:15,309 --> 04:40:16,977
RELATED TO COMMON PEDIATRIC

7279
04:40:16,977 --> 04:40:20,147
ISSUES, YOU KNOW, ACROSS THESE

7280
04:40:20,147 --> 04:40:23,317
BRAIN-BASED SPECIALTIES AND

7281
04:40:23,317 --> 04:40:24,952
OTHER INTERNATIONAL ISSUES AS

7282
04:40:24,952 --> 04:40:25,119
OLD.

7283
04:40:25,119 --> 04:40:30,958
HOPE TO HAVE MORE TO SAY IN THE

7284
04:40:30,958 --> 04:40:31,726
COMING WEEKS.

7285
04:40:31,726 --> 04:40:32,927
>> GREAT TO SEE NEWG
MEMBERS

7286
04:40:32,927 --> 04:40:35,463
BEING ACTIVITY IN THIS SPACE,

7287
04:40:35,463 --> 04:40:41,702
GREAT MOMENTUM. 

7288
04:40:41,702 --> 04:40:43,237
LET'S SEE.

7289
04:40:43,237 --> 04:40:43,771
CHRISTINE?

7290
04:40:43,771 --> 04:40:44,105
>> SORRY.

7291
04:40:44,105 --> 04:40:46,273
TWO SHORT THINGS FOR ME.

7292
04:40:46,273 --> 04:40:49,910
I THINK MOST OF YOU PROBABLY SAW

7293
04:40:49,910 --> 04:40:51,445
BUT SASKIA AND I HAD BEEN

7294
04:40:51,445 --> 04:40:53,314
INVITED TO CONTRIBUTE TO A

7295
04:40:53,314 --> 04:40:55,049
SERIES, IN THE NEW ENGLAND

7296
04:40:55,049 --> 04:40:56,817
JOURNAL, FUNDAMENTALS OF MEDICAL

7297
04:40:56,817 --> 04:40:57,051
ETHICS.

7298
04:40:57,051 --> 04:40:59,987
WE WROTE A PAPER ON ETHICS,

7299
04:40:59,987 --> 04:41:01,956
HIGHLY INNOVATIVE RESEARCH ON

7300
04:41:01,956 --> 04:41:03,157
BRAIN DISEASES.

7301
04:41:03,157 --> 04:41:05,126
COMMENTARY, IT'S A COMMENTARY,

7302
04:41:05,126 --> 04:41:08,395
WITH CASES, KAREN.

7303
04:41:08,395 --> 04:41:08,629
CASES.

7304
04:41:08,629 --> 04:41:09,597
SO JUST THAT.

7305
04:41:09,597 --> 04:41:12,433
THE OTHER THING I WANTED TO

7306
04:41:12,433 --> 04:41:15,503
MENTION IS THAT WHEN NEUROETHICS

7307
04:41:15,503 --> 04:41:17,371
GROUP USUALLY MEETS IN JANUARY,

7308
04:41:17,371 --> 04:41:18,472
FEBRUARY, AND THEN AUGUST, BUT

7309
04:41:18,472 --> 04:41:21,942
WE'RE NOT GOING TO MEET IN

7310
04:41:21,942 --> 04:41:23,177
JANUARY/ FEBRUARY OF 2025.

7311
04:41:23,177 --> 04:41:27,548
WE'LL HAVE OUR NEXT MEETING IN

7312
04:41:27,548 --> 04:41:28,849
MAY OF 2025.

7313
04:41:28,849 --> 04:41:30,384
YOU'LL GET MORE INFORMATION

7314
04:41:30,384 --> 04:41:31,752
ABOUT THIS SOON BUT I WANTED TO

7315
04:41:31,752 --> 04:41:32,953
LET PEOPLE KNOW THAT WAS GOING

7316
04:41:32,953 --> 04:41:37,825
TO HAPPEN.

7317
04:41:37,825 --> 04:41:38,492
>> RIGHT.

7318
04:41:38,492 --> 04:41:39,794
THANKS FOR THE REMINDER.

7319
04:41:39,794 --> 04:41:43,063
WE'LL SEND OUT INFORMATION ON

7320
04:41:43,063 --> 04:41:44,498
THAT SHORTLY.

7321
04:41:44,498 --> 04:41:44,698
NITA?

7322
04:41:44,698 --> 04:41:45,699
>> THANKS, JOHN.

7323
04:41:45,699 --> 04:41:48,002
A BUNCH OF THINGS, LET ME GO

7324
04:41:48,002 --> 04:41:49,804
THROUGH THEM QUICKLY.

7325
04:41:49,804 --> 04:41:51,605
FIRST, NEXT WEEK I'LL BE IN

7326
04:41:51,605 --> 04:41:52,173
PARIS AGAIN.

7327
04:41:52,173 --> 04:41:57,178
WE HAVE THE SECOND MEETING OF

7328
04:41:57,178 --> 04:42:00,080
THE UNESCO GROUP ON

7329
04:42:00,080 --> 04:42:01,882
NEUROTECHNOLOGY, GOING INTO IN A

7330
04:42:01,882 --> 04:42:05,820
WE HAD THE FIRST DRAFT WHICH WAS

7331
04:42:05,820 --> 04:42:09,023
RELEASED TO THE BROADER PUBLIC.

7332
04:42:09,023 --> 04:42:11,792
WE RECEIVED OVER 7,000

7333
04:42:11,792 --> 04:42:13,427
INDIVIDUAL COMMENTS.

7334
04:42:13,427 --> 04:42:15,262
THERE WERE OVER 25 GLOBAL

7335
04:42:15,262 --> 04:42:17,531
CONSULTATIONS RUN AS WELL.

7336
04:42:17,531 --> 04:42:19,733
MANY OF YOU PARTICIPATED BOTH

7337
04:42:19,733 --> 04:42:20,601
THROUGH THE CONSULTATION

7338
04:42:20,601 --> 04:42:22,803
PROCESS, ALSO THROUGH COMMENTS.

7339
04:42:22,803 --> 04:42:24,338
AND WE HAVE BEEN OVER THE PAST

7340
04:42:24,338 --> 04:42:25,840
FEW WEEKS INCORPORATING MANY OF

7341
04:42:25,840 --> 04:42:29,043
THOSE COMMENTS INTO AN UPDATED

7342
04:42:29,043 --> 04:42:29,243
DRAFT.

7343
04:42:29,243 --> 04:42:32,012
THAT DRAFT HAS TO BE COMPLETE BY

7344
04:42:32,012 --> 04:42:34,415
STATUTORY REQUIREMENTS BY THE

7345
04:42:34,415 --> 04:42:36,917
END OF NEXT WEEK FOR IT TO BE

7346
04:42:36,917 --> 04:42:38,352
TRANSLATED INTO SIX LANGUAGES

7347
04:42:38,352 --> 04:42:40,221
THAT THEN GOES THROUGH A

7348
04:42:40,221 --> 04:42:41,188
YEAR-LONG POLITICAL PROCESS.

7349
04:42:41,188 --> 04:42:42,623
LOOK FOR THE UPDATED DRAFT

7350
04:42:42,623 --> 04:42:44,158
COMING OUT SOON.

7351
04:42:44,158 --> 04:42:46,994
SECOND IS THE UNIFORM LAWS

7352
04:42:46,994 --> 04:42:51,065
COMMISSION HAS JUST LAUNCHED A

7353
04:42:51,065 --> 04:42:53,133
STUDY COMMITTEE ON MENTAL

7354
04:42:53,133 --> 04:42:56,403
PRIVACY, AND IN PARTICULAR

7355
04:42:56,403 --> 04:42:56,904
ADDRESSING UPDATED LEGAL

7356
04:42:56,904 --> 04:43:02,576
MOVEMENTS IN A  -- THAT IS HAVE

7357
04:43:02,576 --> 04:43:04,745
HAPPENED IN THE UNITED STATES,

7358
04:43:04,745 --> 04:43:06,413
PROTECT OF NEURAL AND COGNITIVE

7359
04:43:06,413 --> 04:43:07,147
DATA.

7360
04:43:07,147 --> 04:43:12,419
IF YOU'RE INTERESTING IN SERVING

7361
04:43:12,419 --> 04:43:14,822
AS BE A OBSERVER, REACH OUT TO

7362
04:43:14,822 --> 04:43:15,155
ME.

7363
04:43:15,155 --> 04:43:15,756
WE'RE INTERVIEWING POTENTIAL

7364
04:43:15,756 --> 04:43:17,858
REPORTERS TO SERVE ON THE

7365
04:43:17,858 --> 04:43:18,425
COMMITTEE.

7366
04:43:18,425 --> 04:43:20,728
THAT WILL -- IF ALL GOES

7367
04:43:20,728 --> 04:43:21,795
ACCORDING TO THE WAY THE

7368
04:43:21,795 --> 04:43:24,098
PROJECTS GO IT WILL BE A

7369
04:43:24,098 --> 04:43:25,633
YEAR-LONG STUDY COMMITTEE, GOING

7370
04:43:25,633 --> 04:43:28,002
INTO A DRAFTING COMMITTEE,

7371
04:43:28,002 --> 04:43:29,570
PROPOSE DRAFT LEGISLATION FOR

7372
04:43:29,570 --> 04:43:29,803
STATES.

7373
04:43:29,803 --> 04:43:32,406
THIRD, THERE'S AN AMERICAN LAW

7374
04:43:32,406 --> 04:43:34,608
INSTITUTE AND EUROPEAN LAW

7375
04:43:34,608 --> 04:43:35,709
INSTITUTE PROJECT, JOINT

7376
04:43:35,709 --> 04:43:37,544
PROJECT, LAUNCHING ON DEVELOPING

7377
04:43:37,544 --> 04:43:40,281
PRINCIPLES OF BIOMETRICS, MOIST

7378
04:43:40,281 --> 04:43:41,081
ON COGNITIVE BIOMETRICS,

7379
04:43:41,081 --> 04:43:42,716
INFERENCES DRAWN ABOUT BRAIN AND

7380
04:43:42,716 --> 04:43:44,151
MENTAL STATES.

7381
04:43:44,151 --> 04:43:49,523
AGAIN, IF YOU'RE INTERESTED IN

7382
04:43:49,523 --> 04:43:50,858
SEARCH SERVING AS OBSERVER LET

7383
04:43:50,858 --> 04:43:55,696
ME KNOW.

7384
04:43:55,696 --> 04:43:57,631
WE CONCLUDED APPLIED ETHICS PLUS

7385
04:43:57,631 --> 04:43:58,732
PROGRAM PARTNERING WITH REAL

7386
04:43:58,732 --> 04:44:04,204
WORLD ORGANIZATIONS LIKE THIS

7387
04:44:04,204 --> 04:44:07,808
SUMMER UNESCO OECD AND OPEN A.I.

7388
04:44:07,808 --> 04:44:08,575
AND OTHER PROBLEMS, BRINGING

7389
04:44:08,575 --> 04:44:13,080
REAL WORLD PROJECTS TO --

7390
04:44:13,080 --> 04:44:14,982
PROBLEMS TO STUDENTS, A NUMBER

7391
04:44:14,982 --> 04:44:15,749
WERE NEUROETHICS RELATED.

7392
04:44:15,749 --> 04:44:18,519
IF YOU HAVE A PROBLEM YOU'RE

7393
04:44:18,519 --> 04:44:21,989
INTERESTED TO BRING TO TALENTED

7394
04:44:21,989 --> 04:44:24,925
AND INTERESTING DUKE STUDENTS TO

7395
04:44:24,925 --> 04:44:25,993
WORK ON COLLABORATIVELY, REACH

7396
04:44:25,993 --> 04:44:27,661
OUT AND LET ME KNOW.

7397
04:44:27,661 --> 04:44:31,832
THE RESULTS OF A LOT OF THOSE

7398
04:44:31,832 --> 04:44:32,833
REAL WORLD PROBLEM-SOLVING

7399
04:44:32,833 --> 04:44:36,203
PRACTICES ARE UP ON OUR WEBSITE.

7400
04:44:36,203 --> 04:44:38,739
AND THE LAST IS THAT I HAVE A

7401
04:44:38,739 --> 04:44:42,109
PAPER COMING OUT WITH MARCELLO

7402
04:44:42,109 --> 04:44:43,877
AND PATRICK IN NEURON NEXT

7403
04:44:43,877 --> 04:44:47,581
MONTH, ON NEURAL DATA AND

7404
04:44:47,581 --> 04:44:48,482
COGNITIVE BIOMETRICS, HOW WE

7405
04:44:48,482 --> 04:44:50,017
OUGHT TO THINK ABOUT EXPANDING

7406
04:44:50,017 --> 04:44:51,986
THE CATEGORIES AND WHAT THE

7407
04:44:51,986 --> 04:44:54,488
LEGAL APPROACH, IT HAS A

7408
04:44:54,488 --> 04:44:57,658
TREMENDOUS AMOUNT OF USEFUL

7409
04:44:57,658 --> 04:44:58,859
TABLES AND DATA THAT PATRICK

7410
04:44:58,859 --> 04:45:01,061
CREATED THAT I HOPE WILL BE A

7411
04:45:01,061 --> 04:45:04,131
USEFUL RESOURCE FOR THE

7412
04:45:04,131 --> 04:45:05,332
COMMUNITY.

7413
04:45:05,332 --> 04:45:06,533
THAT'S IT.

7414
04:45:06,533 --> 04:45:08,168
>> THANKS, NITA.

7415
04:45:08,168 --> 04:45:10,904
DID I MISS ANYBODY?

7416
04:45:10,904 --> 04:45:11,105
GREAT.

7417
04:45:11,105 --> 04:45:12,773
AGAIN, IT SHOWS HOW MUCH

7418
04:45:12,773 --> 04:45:13,974
MOMENTUM IS BEING BUILD AROUND

7419
04:45:13,974 --> 04:45:15,075
THIS AREA.

7420
04:45:15,075 --> 04:45:16,910
THERE ARE MANY THINGS WE DO

7421
04:45:16,910 --> 04:45:22,483
UNIQUE TO BRAIN,  UNIQUE FOR NO,

7422
04:45:22,483 --> 04:45:23,817
EVENTUALLY TRANSLATABLE ACROSS

7423
04:45:23,817 --> 04:45:24,985
THE FEET IN NEUROSCIENCE AND

7424
04:45:24,985 --> 04:45:25,786
BIOMEDICAL SCIENCE.

7425
04:45:25,786 --> 04:45:27,554
THANK YOU FOR BEING HERE, IN

7426
04:45:27,554 --> 04:45:29,857
PARTICULAR TO THE FOLKS WHO GAVE

7427
04:45:29,857 --> 04:45:30,824
PRESENTATIONS AND PARTICIPATED

7428
04:45:30,824 --> 04:45:31,592
IN DISCUSSION.

7429
04:45:31,592 --> 04:45:33,794
AS ALWAYS, IF YOU'RE LIKE ME YOU

7430
04:45:33,794 --> 04:45:38,565
LEARNED A LOT TODAY WITH FOOD

7431
04:45:38,565 --> 04:45:38,866
FOR THOUGHT.

7432
04:45:38,866 --> 04:45:39,900
THANKS TO THE FOLKS WHO MADE

7433
04:45:39,900 --> 04:45:41,335
THIS HAPPEN.

7434
04:45:41,335 --> 04:45:42,536
WE MENTIONED THEM EARLIER.

7435
04:45:42,536 --> 04:45:44,405
WITH THAT I WILL TURN IT OVER TO

7436
04:45:44,405 --> 04:45:48,342
ANDREA TO CLOSE THE MEETING OUT.

7437
04:45:48,342 --> 04:45:49,443
>> THANKS, JOHN.

7438
04:45:49,443 --> 04:45:52,079
ADDING MY THANKS TO ALL.

7439
04:45:52,079 --> 04:45:55,115
AND JUST A COUPLE REMINDERS.

7440
04:45:55,115 --> 04:45:56,650
AS WAS JUST MENTIONED, MEETING

7441
04:45:56,650 --> 04:45:58,285
WILL BE MOVED TO MAY SO THERE

7442
04:45:58,285 --> 04:46:01,288
WON'T BE THAT JANUARY MEETING,

7443
04:46:01,288 --> 04:46:03,223
YOU'LL BE BETTING INFORMATION.

7444
04:46:03,223 --> 04:46:04,758
WE'RE SHARE INFORMATION ON THE

7445
04:46:04,758 --> 04:46:12,733
NEURAL A.I.  WORKSHOP JOE

7446
04:46:12,733 --> 04:46:13,600
MENTIONED, REGISTRATION LIVE IN

7447
04:46:13,600 --> 04:46:15,702
THE NEXT WEEK OR SO.

7448
04:46:15,702 --> 04:46:17,337
FINAL REMINDER THAT THIS

7449
04:46:17,337 --> 04:46:18,405
VIDEOCAST FOR THOSE THAT WEREN'T

7450
04:46:18,405 --> 04:46:20,707
HERE ALL DAY AND WANTED TO CATCH

7451
04:46:20,707 --> 04:46:24,778
UP OR MAY HAVE JOINED LATE

7452
04:46:24,778 --> 04:46:25,746
VIDEOCAST WILL BE AVAILABLE,

7453
04:46:25,746 --> 04:46:27,681
POSTED SHORTLY ON THE NIH

7454
04:46:27,681 --> 04:46:30,250
VIDEOCAST SITE AND THE BRAIN

7455
04:46:30,250 --> 04:46:30,484
WEBSITE.

7456
04:46:30,484 --> 04:46:33,087
WITH THAT, WE CAN CLOSE OUT THIS

7457
04:46:33,087 --> 04:46:33,320
MEETING.

7458
04:46:33,320 --> 04:46:35,823
AND WE HOPE TO SEE YOU AT

7459
04:46:35,823 --> 04:46:36,690
VARIOUS OTHER MEETINGS BUT

7460
04:46:36,690 --> 04:46:39,193
CERTAINLY WHEN WE MEET AGAIN AS

7461
04:46:39,193 --> 04:46:41,161
THE NEWG IN MAY.

7462
04:46:41,161 --> 04:46:41,562
THANKS, EVERYONE.

7463
04:46:41,562 --> 04:55:43,518
[END OF PROGRAM]

7464
04:55:43,518 --> 04:55:49,558
 

7465
04:55:49,558 --> 04:55:59,601