1 00:00:06,146 --> 00:00:06,613 >> ALL RIGHT. 2 00:00:06,613 --> 00:00:09,015 GOOD AFTERNOON AND WELCOME TO 3 00:00:09,015 --> 00:00:11,017 TODAY'S CLINICAL CENTER GREAT 4 00:00:11,017 --> 00:00:13,353 TEACHERS GRAND ROUNDS. 5 00:00:13,353 --> 00:00:15,422 THE HOPKINS CLOUD CME ACTIVITY 6 00:00:15,422 --> 00:00:19,092 CODE FOR TODAY'S GRAND ROUNDS IS 7 00:00:19,092 --> 00:00:19,359 57870. 8 00:00:19,359 --> 00:00:21,895 PLEASE TEXT THIS CODE TO THE 9 00:00:21,895 --> 00:00:22,929 JOHNS HOPKINS CME PHONE NUMBER 10 00:00:22,929 --> 00:00:25,232 ON THE SLIDE AND RECEIVE CME 11 00:00:25,232 --> 00:00:27,000 CREDIT FOR THIS LECTURE. 12 00:00:27,000 --> 00:00:28,969 WE GLADLY INVITE YOU TO PROVIDE 13 00:00:28,969 --> 00:00:31,404 FEEDBACK ON TODAY'S LECTURE, BY 14 00:00:31,404 --> 00:00:33,807 SCANNING THE QR CODE SHOWN ON 15 00:00:33,807 --> 00:00:35,609 THE CME SLIDE, FOR THOSE 16 00:00:35,609 --> 00:00:37,310 APPLYING FOR CME, YOU WILL 17 00:00:37,310 --> 00:00:39,880 RECEIVE A FREED BACK SURVEY LINK 18 00:00:39,880 --> 00:00:41,147 VIA E-MAIL, THE SURVEY WILL BE 19 00:00:41,147 --> 00:00:42,883 USED TO PROVIDE US WITH 20 00:00:42,883 --> 00:00:44,084 IMPORTANT FEEDBACK ABOUT THIS 21 00:00:44,084 --> 00:00:46,486 PRESENTATION AND IT ALSO ALLOWS 22 00:00:46,486 --> 00:00:48,688 YOU TO SUBMIT SUGGESTIONS FOR 23 00:00:48,688 --> 00:00:51,391 FUTURE GRAND ROUNDS TOPICS. 24 00:00:51,391 --> 00:00:52,192 FOLLOWING THE PRESENTATION, 25 00:00:52,192 --> 00:00:53,159 QUESTIONS FOR THE SPEAKER WILL 26 00:00:53,159 --> 00:00:56,897 BE TAKEN FROM THE MICROPHONES IN 27 00:00:56,897 --> 00:00:58,498 THE AISLES ADDITIONALLY 28 00:00:58,498 --> 00:00:59,766 VIDEOCAST VIEWERS MAY SUBMIT 29 00:00:59,766 --> 00:01:01,768 QUESTIONS AT ANY TIME DURING THE 30 00:01:01,768 --> 00:01:03,436 LECTURE BY SCROLLING DOWN AND 31 00:01:03,436 --> 00:01:05,538 CLICKING THE LIVE FEEDBACK 32 00:01:05,538 --> 00:01:07,774 BUTTON LOCATED ON THE VIDEOCAST 33 00:01:07,774 --> 00:01:08,008 WEBSITE. 34 00:01:08,008 --> 00:01:09,943 QUESTIONS WILL BE ANSWERED AS 35 00:01:09,943 --> 00:01:12,412 TIME PERMITS AT THE CONCLUSION 36 00:01:12,412 --> 00:01:13,747 OF THIS PRESENTATION. 37 00:01:13,747 --> 00:01:14,915 THE CONTEMPORARY CLINICAL 38 00:01:14,915 --> 00:01:16,216 MEDICINE GREAT TEACHERS GRAND 39 00:01:16,216 --> 00:01:17,817 ROUNDS IS A SPECIAL SERIES 40 00:01:17,817 --> 00:01:21,888 WITHIN THE CLINICAL CENTER GRAND 41 00:01:21,888 --> 00:01:27,527 ROUNDS PROGRAM. 42 00:01:27,527 --> 00:01:29,496 DR. PAUL DISTINGUISHED SCIENTIST 43 00:01:29,496 --> 00:01:31,331 EMERITUS OF NIAMS WHO RECENTLY 44 00:01:31,331 --> 00:01:33,433 PASS ASH WAY, THE PURPOSE WAS TO 45 00:01:33,433 --> 00:01:34,134 INVITE THOUGHT LEADERS IN 46 00:01:34,134 --> 00:01:35,535 MEDICINE AND SCIENCE TO PRESENT 47 00:01:35,535 --> 00:01:39,072 ON TOPICS AND IDEAS ABOUT WHICH 48 00:01:39,072 --> 00:01:40,407 THEY ARE PASSIONATE TO SHARE THE 49 00:01:40,407 --> 00:01:42,142 REFLECTIONS WITH US AND TO 50 00:01:42,142 --> 00:01:43,276 INSPIRE YOUNG INVESTIGATORS IN 51 00:01:43,276 --> 00:01:47,080 THE PURSUIT OF CAREERS IN 52 00:01:47,080 --> 00:01:47,781 BIOMEDICAL RESEARCH. 53 00:01:47,781 --> 00:01:50,850 SO TODAY WE'RE HONORED TO HOST 54 00:01:50,850 --> 00:01:56,122 OUR GREAT TEACHER DR. EERIC 55 00:01:56,122 --> 00:01:58,191 TOPOL, EXECUTIVE VICE PRESIDENT 56 00:01:58,191 --> 00:01:59,359 OF SCRIPTS RESEARCH, WHEN HE WAS 57 00:01:59,359 --> 00:02:01,227 THE DIRECTOR AND FOUNDER AND 58 00:02:01,227 --> 00:02:03,196 DIRECTS THE SCRIPTS 59 00:02:03,196 --> 00:02:05,565 TRANSLATIONAL INSTITUTE AND 60 00:02:05,565 --> 00:02:08,234 PROFESSOR OF MOLECULAR MEDICINE. 61 00:02:08,234 --> 00:02:09,736 DR. TOPOL IS A RENOWNED 62 00:02:09,736 --> 00:02:11,604 SCIENTIST AND THOUGHT LEADER 63 00:02:11,604 --> 00:02:13,006 THIS GENOMIC DIGITAL AND 64 00:02:13,006 --> 00:02:15,976 ARTIFICIAL INTELLIGENCE TOOLS TO 65 00:02:15,976 --> 00:02:19,045 PROMOTE HUMAN HEALTH. 66 00:02:19,045 --> 00:02:21,047 A PIONEERING CARDIOLOGIST 67 00:02:21,047 --> 00:02:23,116 DR. ERIC TOPOL IS 1 OF THE MOST 68 00:02:23,116 --> 00:02:26,086 CITED RESEARCHERS IN MEDICINE 69 00:02:26,086 --> 00:02:26,987 TODAY. 70 00:02:26,987 --> 00:02:30,357 DR. TOTOL EARNED HIS DEGREE WITH 71 00:02:30,357 --> 00:02:31,624 HIGHEST DISTINCTION IN MEDICINE 72 00:02:31,624 --> 00:02:33,793 FROM UNIVERSITY OF VIRGINIA AND 73 00:02:33,793 --> 00:02:34,728 HIS MEDICAL DOCTORATE WITH 74 00:02:34,728 --> 00:02:35,562 HONORS FROM THE ROUGH ATOM 75 00:02:35,562 --> 00:02:37,230 CHESTER SCHOOL OF MEDICINE IN 76 00:02:37,230 --> 00:02:37,797 DENTISTRY. 77 00:02:37,797 --> 00:02:39,165 HE COMPLETED HADIS INTERNSHIP 78 00:02:39,165 --> 00:02:40,700 AND RESIDENCY AT THE UNIVERSITY 79 00:02:40,700 --> 00:02:43,703 OF CALIFORNIA SAN FRANCISCO, AND 80 00:02:43,703 --> 00:02:44,771 HIS FELLOWSHIP IN CARDIOVASCULAR 81 00:02:44,771 --> 00:02:46,172 MEDICINE AT THE YAWNS HOPKINS 82 00:02:46,172 --> 00:02:48,041 HOSPITAL AND THEN A SENIOR 83 00:02:48,041 --> 00:02:49,642 FOLLOW OSHIP IN CORONARY ANGIO 84 00:02:49,642 --> 00:02:51,778 PLAOF THEY AT THE SAN FRANCISCO 85 00:02:51,778 --> 00:02:52,145 HEART INSTITUTE. 86 00:02:52,145 --> 00:02:55,315 IN THE YEARS THAT FOLLOWED, 87 00:02:55,315 --> 00:02:56,683 DR. TOPOL CONTINUED AN 88 00:02:56,683 --> 00:02:59,419 EXTRAORDINARY CAREER AS A 89 00:02:59,419 --> 00:03:00,286 PRACTICING CARDIOLOGIST, 90 00:03:00,286 --> 00:03:01,688 EDUCATOR, MENTOR, RESEARCHER AND 91 00:03:01,688 --> 00:03:04,424 A PROFESSOR IN MEDICINE AT THE 92 00:03:04,424 --> 00:03:05,692 UNIVERSITY OF MICHIGAN, CASE 93 00:03:05,692 --> 00:03:09,696 WESTERN RESERVE AND SCRIPTS 94 00:03:09,696 --> 00:03:10,563 RESEARCH INSTITUTE. 95 00:03:10,563 --> 00:03:14,934 MANY OF HIS OVER 200 FORMER 96 00:03:14,934 --> 00:03:16,569 TRAINEES SERVE IN ACADEMIC 97 00:03:16,569 --> 00:03:19,072 POSITIONS AS LEADERS IN 98 00:03:19,072 --> 00:03:20,373 CARDIOVASCULAR MEDICINE AND 99 00:03:20,373 --> 00:03:22,075 GENOMIC RESEARCH WORLD WIDE. 100 00:03:22,075 --> 00:03:23,710 HE HAS LED CLINICAL TRIALS IN 101 00:03:23,710 --> 00:03:25,178 OVER 40 COUNTRIES AND PIONEERED 102 00:03:25,178 --> 00:03:26,579 THE DEVELOPMENT OF MANY 103 00:03:26,579 --> 00:03:30,283 MEDICATIONS USE INDEED MEDICINE 104 00:03:30,283 --> 00:03:33,520 AND CARDIOLOGY TODAY SUCH AS 105 00:03:33,520 --> 00:03:35,121 TPA, PLAVIX, AND 106 00:03:35,121 --> 00:03:35,722 [INDISCERNIBLE]. 107 00:03:35,722 --> 00:03:37,257 HIS RESEARCH NOW FOCUSES ON THE 108 00:03:37,257 --> 00:03:38,758 BROADER QUESTIONS OF PRECISION 109 00:03:38,758 --> 00:03:43,196 MEDICINE AND ARTIFICIAL 110 00:03:43,196 --> 00:03:43,730 INTELLIGENCE. 111 00:03:43,730 --> 00:03:44,764 DR. TOPOL HAS A LONG ARK 112 00:03:44,764 --> 00:03:48,068 FILIATION WITH THE NIH AND HE 113 00:03:48,068 --> 00:03:49,602 CURRENTLY SERVES AS PRINCIPAL 114 00:03:49,602 --> 00:03:51,938 INVESTIGATOR TO 2 LARGE NIH 115 00:03:51,938 --> 00:03:52,639 GRANT. 116 00:03:52,639 --> 00:03:54,007 ALSO THE ALL OF US RESEARCH 117 00:03:54,007 --> 00:03:55,408 PROGRAM THAT SUPPORTS PRECISION 118 00:03:55,408 --> 00:03:57,377 MEDICINE AND A CLINICAL AND 119 00:03:57,377 --> 00:03:59,079 TRANSLATIONAL SCIENCE AWARD THAT 120 00:03:59,079 --> 00:04:00,380 PROMOTES INNOVATION IN MEDICINE 121 00:04:00,380 --> 00:04:02,382 AND FUTURE MEDICAL RESEARCHERS 122 00:04:02,382 --> 00:04:09,622 EDUCATION AND CAREER TRAINING. 123 00:04:09,622 --> 00:04:11,357 DR. TOPOL IS ALSO A MEMBER OF 124 00:04:11,357 --> 00:04:12,659 THE NATIONAL ACADEMY OF MEDICINE 125 00:04:12,659 --> 00:04:14,761 AND RECIPIENT OF NUMEROUS AWARDS 126 00:04:14,761 --> 00:04:16,963 INCLUDING THE ASSOCIATION FOR 127 00:04:16,963 --> 00:04:19,065 BLACK CARDIOLOGIST AWARD FOR 128 00:04:19,065 --> 00:04:20,967 PROMOTING DIVERSITY IN 129 00:04:20,967 --> 00:04:21,768 CARDIOVASCULAR MEDICINE, ARN 130 00:04:21,768 --> 00:04:24,804 OLDER PEOPLE P. GOLD'S 131 00:04:24,804 --> 00:04:26,139 FOUNDATION IN HUMANISM MEDAL AND 132 00:04:26,139 --> 00:04:28,241 THE NATIONAL LIBRARY OF 133 00:04:28,241 --> 00:04:30,443 MEDICINE'S DONALD A. B. 134 00:04:30,443 --> 00:04:33,746 LINDBERGH DISTINGUISHED HEALTH 135 00:04:33,746 --> 00:04:34,314 COMMUNICATIONS AWARD. 136 00:04:34,314 --> 00:04:36,416 DR. TOPOL IS ALSO THE AUTHOR OF 137 00:04:36,416 --> 00:04:39,486 3 BEST SELLERS ON THURE OF 138 00:04:39,486 --> 00:04:41,421 MEDICINE, TO INCLUDE DEEP 139 00:04:41,421 --> 00:04:42,589 MEDICINE HOWARD FICIAL 140 00:04:42,589 --> 00:04:46,526 INTELIENCE CAN MAKE HUMAN 141 00:04:46,526 --> 00:04:46,926 HEALTHCARE AGAIN. 142 00:04:46,926 --> 00:04:47,827 TODAY'S TITLE IS OPPORTUNITIES 143 00:04:47,827 --> 00:04:49,429 AND PERILS OF ARTIFICIAL 144 00:04:49,429 --> 00:04:50,930 INTELLIGENCE IN HEALTH AND 145 00:04:50,930 --> 00:04:51,264 MEDICINE. 146 00:04:51,264 --> 00:04:53,133 PLEASE JOIN ME IN WELCOMING 147 00:04:53,133 --> 00:04:59,472 TODAY'S GRAND ROUND SPEAKER 148 00:04:59,472 --> 00:04:59,839 DR. ERIC TOPOL. 149 00:04:59,839 --> 00:05:03,810 [ APPLAUSE ] 150 00:05:03,810 --> 00:05:04,144 >> THANKS. 151 00:05:04,144 --> 00:05:05,512 >> THANKS, SO MUCH CHRIS FOR 152 00:05:05,512 --> 00:05:06,246 YOUR KIND INTRODUCTION AND IT'S 153 00:05:06,246 --> 00:05:08,114 GOOD TO BE WITH YOU HERE AT THE 154 00:05:08,114 --> 00:05:09,382 CLINICAL CENTER HAVING NOT BEEN 155 00:05:09,382 --> 00:05:10,683 HERE BEFORE OVER ALL THESE YEARS 156 00:05:10,683 --> 00:05:14,087 WITH MANY TRIPS TO NIH CAMPUS 157 00:05:14,087 --> 00:05:16,089 AND I'M GOING TO GET INTO AI AS 158 00:05:16,089 --> 00:05:18,258 IT STANDS TODAY AND AS YOU KNOW 159 00:05:18,258 --> 00:05:20,126 IT'S A VERY DYNAMIC HOT FIELD 160 00:05:20,126 --> 00:05:22,328 AND IN SOME WAYS TROUBLING IN 161 00:05:22,328 --> 00:05:23,930 OUR DAILY LIFE FIELD. 162 00:05:23,930 --> 00:05:26,900 AND I'LL BE TALKING ABOUT 163 00:05:26,900 --> 00:05:29,936 VARIOUS ASPECTS OF HOW WE EXPECT 164 00:05:29,936 --> 00:05:33,473 IT TO TRANSFORM MEDICINE AND 165 00:05:33,473 --> 00:05:34,207 HUMAN HEALTH. 166 00:05:34,207 --> 00:05:35,742 I WOULD JUST MENTION I HAVE NO 167 00:05:35,742 --> 00:05:36,809 CONFLICTS OF INTEREST TO 168 00:05:36,809 --> 00:05:40,713 DISCLOSE AND I WILL NOT BE 169 00:05:40,713 --> 00:05:42,949 DISCUSSING ANY DRUGS. 170 00:05:42,949 --> 00:05:44,450 SO JEFFREY HINTON REGARDED AS 171 00:05:44,450 --> 00:05:48,321 THE FATHER OF DEEP LEARNING AI, 172 00:05:48,321 --> 00:05:49,856 HE RECENTLY LEFT GOOGLE BECAUSE 173 00:05:49,856 --> 00:05:53,092 HE FELT CONFLICTED ABOUT THE 174 00:05:53,092 --> 00:05:53,560 WORRIES OF AI. 175 00:05:53,560 --> 00:05:55,862 BUT WHEN I SPOKE TO HIM WHAT 176 00:05:55,862 --> 00:05:58,498 HEED WAS THIS, AND THIS 177 00:05:58,498 --> 00:06:00,800 IS--ACTUALLY JUST POSTED THIS 178 00:06:00,800 --> 00:06:03,169 MORNING, A PODCAST WITH FRANCIS 179 00:06:03,169 --> 00:06:05,838 COLLINS ON HIS NEW BOOK, THE 180 00:06:05,838 --> 00:06:07,373 ROAD TO KNOWLEDGE, BUT AT THE 181 00:06:07,373 --> 00:06:11,211 TIME WHEN I SWOAK TO 182 00:06:11,211 --> 00:06:12,812 GEOFFREY HINTON, HE SAID I 183 00:06:12,812 --> 00:06:14,681 ALWAYS PIVOT TO MEDICINE AS AN 184 00:06:14,681 --> 00:06:16,416 EXAMPLE OF AI, GOOD IT CAN DO 185 00:06:16,416 --> 00:06:17,784 BECAUSE ALMOST EVERYTHING IT'S 186 00:06:17,784 --> 00:06:19,118 GOING DO THERE IS GOING TO BE 187 00:06:19,118 --> 00:06:19,352 GOOD. 188 00:06:19,352 --> 00:06:20,553 SO THIS IS COMING FROM A PERSON 189 00:06:20,553 --> 00:06:22,589 WHO HAS THE HIGHEST REGARD OF 190 00:06:22,589 --> 00:06:24,390 THE AI COMMUNITY AND SPEAKING 191 00:06:24,390 --> 00:06:25,325 ABOUT THE EXPECTATIONS AND 192 00:06:25,325 --> 00:06:29,329 MEDICINE AND HEALTHCARE. 193 00:06:29,329 --> 00:06:31,698 SO NOW LET ME GET INTO THOSE 194 00:06:31,698 --> 00:06:32,765 BRIGHT PROSPECTS AND FIRSTLY THE 195 00:06:32,765 --> 00:06:34,400 ISSUE OF MEDICAL ERRORS, AND 196 00:06:34,400 --> 00:06:36,369 THAT IN ITSELF IS NOT A BRIGHT 197 00:06:36,369 --> 00:06:38,204 PROSPECT BUT THE FACT THAT WE 198 00:06:38,204 --> 00:06:40,373 CAN TURN THESE AROUND TO SOME 199 00:06:40,373 --> 00:06:41,674 DEGREE. 200 00:06:41,674 --> 00:06:44,978 SO THIS IS A BOOK BY DANIEL 201 00:06:44,978 --> 00:06:46,746 OFFREY, AND MEDICAL ERRORS ARE 202 00:06:46,746 --> 00:06:48,147 NOT DISCUSSED MUCH, THEY'RE KIND 203 00:06:48,147 --> 00:06:50,250 OF KEPT QUIET WHICH HE KIND OF 204 00:06:50,250 --> 00:06:51,384 SOUNDED THE ALARM WITH THIS BOOK 205 00:06:51,384 --> 00:06:52,552 BECAUSE THEY ARE VERY HARMFUL 206 00:06:52,552 --> 00:06:55,388 AND MORE OVER JOHNS HOPKINS 207 00:06:55,388 --> 00:06:57,056 STUDY THAT WAS PUBLISHED LAST 208 00:06:57,056 --> 00:06:59,425 YEAR, POINTED ON UTR THAT THERE 209 00:06:59,425 --> 00:07:03,062 ARE 800,000 AMERICANS 210 00:07:03,062 --> 00:07:04,831 CONSERVATIVE ESTIMATE, FROM 211 00:07:04,831 --> 00:07:07,433 SERIOUS DIAGNOSTIC ERRORS WHO 212 00:07:07,433 --> 00:07:09,269 ARE SIGNIFICANTLY DISABLED OR 213 00:07:09,269 --> 00:07:09,469 DIE. 214 00:07:09,469 --> 00:07:13,906 SO THIS IS A BIG PROBLEM THAT 215 00:07:13,906 --> 00:07:16,442 ISN'T ADEQUATELY ACKNOWLEDGED BY 216 00:07:16,442 --> 00:07:17,010 THE MEDICAL COMMUNITY. 217 00:07:17,010 --> 00:07:19,279 THEN WE HAVE THIS TERM PRECISION 218 00:07:19,279 --> 00:07:20,546 MEDICINE WHICH HAS BEEN OF 219 00:07:20,546 --> 00:07:22,649 COURSE USED QUITE A BIT OVER THE 220 00:07:22,649 --> 00:07:22,849 YEARS. 221 00:07:22,849 --> 00:07:25,184 WE HAD A LITTLE PROBLEM WITH 222 00:07:25,184 --> 00:07:26,686 THIS TERM BECAUSE IF YOU KEEP 223 00:07:26,686 --> 00:07:28,288 MAKING THE SAME MISTAKES OVER 224 00:07:28,288 --> 00:07:31,591 AND OVER AGAIN, IT'S VERY 225 00:07:31,591 --> 00:07:32,025 PREICIZE. 226 00:07:32,025 --> 00:07:33,259 BUT WHAT WE REALLY NEED IS 227 00:07:33,259 --> 00:07:34,627 ACCURACY IN MEDICINE AND THAT'S 228 00:07:34,627 --> 00:07:36,963 HOW WE GET THESE DIAGNOSTIC 229 00:07:36,963 --> 00:07:39,899 ERRORS WHICH ARE VERY TROUBLING 230 00:07:39,899 --> 00:07:40,700 DOWN. 231 00:07:40,700 --> 00:07:43,703 SO I WROTE A PIECE IN SCIENCE 232 00:07:43,703 --> 00:07:47,006 EARLIER THIS YEAR TOWARDS THE 233 00:07:47,006 --> 00:07:50,109 ERADICATION OF MEDICAL ERRORS 234 00:07:50,109 --> 00:07:51,978 BRINGING IN ERA OF LARGE MODEL 235 00:07:51,978 --> 00:07:53,479 AI AND LET ME TAKE YOU THROUGH 236 00:07:53,479 --> 00:07:54,881 WHY THERE'S SO MUCH OPTIMISM 237 00:07:54,881 --> 00:07:59,319 THAT WE CAN BRING DOWN THE ERROR 238 00:07:59,319 --> 00:08:00,119 RATE IN DIAGNOSIS. 239 00:08:00,119 --> 00:08:03,556 SO FIRSTLY BEFORE WE GOT TO THE 240 00:08:03,556 --> 00:08:05,458 CURRENT CHAT GPT ERA IF YOU WILL 241 00:08:05,458 --> 00:08:11,097 FROM LATE 2022, THERE WASUNE 242 00:08:11,097 --> 00:08:12,799 MODAL SUPERVISED AI THAT 243 00:08:12,799 --> 00:08:13,900 REQUIRES GROUND TRUTHS AND 244 00:08:13,900 --> 00:08:15,835 REQUIRES THE EXPERTS TO REVIEW 245 00:08:15,835 --> 00:08:18,604 IMAGES OR WHATEVER THE INPUTS 246 00:08:18,604 --> 00:08:20,139 WERE, TYPICALLY THEY WERE 247 00:08:20,139 --> 00:08:20,373 IMAGES. 248 00:08:20,373 --> 00:08:23,743 AND WE LEARN FROM THAT TIME THAT 249 00:08:23,743 --> 00:08:28,481 EVERY TYPE OF MEDICAL SCAN COULD 250 00:08:28,481 --> 00:08:30,416 GET AN AI INTERPRETATION THAT 251 00:08:30,416 --> 00:08:35,521 WOULD BE AS GOOD AS CLINICIANS, 252 00:08:35,521 --> 00:08:36,789 FOR EXAMPLE, RADIOLOGIST OR 253 00:08:36,789 --> 00:08:40,226 PATHOLOGIST AND THIS IS JUST 1 254 00:08:40,226 --> 00:08:42,929 EXAMPLE FROM NYU AND MILLIONS OF 255 00:08:42,929 --> 00:08:44,864 CHEST X-RAYS WHEREBY THE 256 00:08:44,864 --> 00:08:46,499 RADIOLOGIST MISSED A NODULE AND 257 00:08:46,499 --> 00:08:48,067 IT WAS PICKED UP BY THE AI AND 258 00:08:48,067 --> 00:08:52,171 IT TURNED OUT TO BE A CANCEROUS 259 00:08:52,171 --> 00:08:52,505 NODULE. 260 00:08:52,505 --> 00:08:56,275 SO THIS IS PRETTY PROTOTYPIC AND 261 00:08:56,275 --> 00:08:58,444 THE LARGEST STUDY THAT'S BEEN 262 00:08:58,444 --> 00:09:02,248 DONE TO DATE WAS IN SWEDEN WHERE 263 00:09:02,248 --> 00:09:03,282 THEY RANDOMLY ASSIGN 80,000 264 00:09:03,282 --> 00:09:05,585 WOMEN UNDER GOING MAMMOGRAPHY TO 265 00:09:05,585 --> 00:09:06,285 EITHER HAVE TRADITIONAL 266 00:09:06,285 --> 00:09:08,621 INTERPRETATION BY A TRAINED 267 00:09:08,621 --> 00:09:10,490 RADIOLOGIST OR AI ASSISTED AND 268 00:09:10,490 --> 00:09:11,391 THIS PROVED WITHOUT QUESTION 269 00:09:11,391 --> 00:09:13,960 THAT THE PICK UP RATE FOR CANCER 270 00:09:13,960 --> 00:09:17,063 WAS SUPERIOR WHEN THE 2 WERE 271 00:09:17,063 --> 00:09:21,768 COMBINED RATHER THAN THE 272 00:09:21,768 --> 00:09:22,769 RADIOLOGIST ALONE. 273 00:09:22,769 --> 00:09:25,271 AND ALSO IN DOING THAT THE 274 00:09:25,271 --> 00:09:26,606 STREAMLINING OF THE PROCESS WAS 275 00:09:26,606 --> 00:09:26,873 IMPROVED. 276 00:09:26,873 --> 00:09:28,808 NOW THIS IS THE AREA THAT'S HAD 277 00:09:28,808 --> 00:09:32,779 THE MOST STUDY, SO FAR, 278 00:09:32,779 --> 00:09:35,348 INTERESTINGLY AND THAT'S IN 279 00:09:35,348 --> 00:09:35,648 COLONOSCOPY. 280 00:09:35,648 --> 00:09:38,785 AND THERE'S NOW BEEN 40 281 00:09:38,785 --> 00:09:42,388 RANDOMIZED TRIALS OF AI PLUS THE 282 00:09:42,388 --> 00:09:45,024 MACHINE VISION WITH REALTIME, 283 00:09:45,024 --> 00:09:47,460 WITH THE GASTRONEUROLOGYIST OR 284 00:09:47,460 --> 00:09:48,494 JUST A 285 00:09:48,494 --> 00:09:49,128 GASTROENTHUSIASMERROLOGYIST AND 286 00:09:49,128 --> 00:09:50,263 WHAT TURNS OUT TO BE THE CASE IS 287 00:09:50,263 --> 00:09:53,399 THAT THE PICK UP RATE FOR POLYPS 288 00:09:53,399 --> 00:09:56,135 IS SUBSTANTIALLY HIGHER WHEN THE 289 00:09:56,135 --> 00:09:56,602 AI IS USED. 290 00:09:56,602 --> 00:09:58,604 AND SO THIS OF COURSE IS JUST 1 291 00:09:58,604 --> 00:10:01,040 EXAMPLE, AND NOT ONLY IS IT PICK 292 00:10:01,040 --> 00:10:02,875 UP THE POLYPS BUT ALSO IN 293 00:10:02,875 --> 00:10:03,976 REALTIME GIVES A GRADATION OF 294 00:10:03,976 --> 00:10:07,046 HOW LIKELY IT IS TO BE CANCEROUS 295 00:10:07,046 --> 00:10:09,315 AND REQUIRING A BIOPSY. 296 00:10:09,315 --> 00:10:12,685 SO THIS IS JUST A SUMMARY, WHEN 297 00:10:12,685 --> 00:10:14,720 THERE WERE 33 RANDOMIZED TRIALS, 298 00:10:14,720 --> 00:10:16,823 ALMOST EVERY WEEK OR 2 THERE'S 299 00:10:16,823 --> 00:10:19,358 ANOTHER 1 ADDED AND IT SHOWS HOW 300 00:10:19,358 --> 00:10:21,461 THERE'S MARKED IMPROVEMENT OF 301 00:10:21,461 --> 00:10:24,096 PICKING UP POLYPS AND ADENOMAS, 302 00:10:24,096 --> 00:10:25,531 AND INTERESTINGLY STUDY VS SHOWN 303 00:10:25,531 --> 00:10:27,099 THAT IT ESPECIALLY WORKS WELL 304 00:10:27,099 --> 00:10:30,603 TOWARDS THE END OF THE DAY, WHEN 305 00:10:30,603 --> 00:10:32,171 THE DOCTOR OR GASTRONEUROLOGYIST 306 00:10:32,171 --> 00:10:34,740 HAS TIRED EYES BUT THE AI NEVER 307 00:10:34,740 --> 00:10:35,575 GETS TIRED. 308 00:10:35,575 --> 00:10:37,610 NOW WE RECENTLY REVIEWED ALL THE 309 00:10:37,610 --> 00:10:38,644 RANDOMIZED TRIALS, NOW THERE'S 310 00:10:38,644 --> 00:10:40,146 JUST OVER A HUNDRED, BUT AT THE 311 00:10:40,146 --> 00:10:43,983 TIME THERE WERE ABOUT 90 AND 312 00:10:43,983 --> 00:10:45,284 THEIR LARGEST PROPORTION OR 313 00:10:45,284 --> 00:10:46,452 INTERESTINGLY IN GI, LIMITED 314 00:10:46,452 --> 00:10:53,826 NUMBER IN OTHER AREAS SUCH AS 315 00:10:53,826 --> 00:10:54,327 CARDIOLOGY, RADIOLOGY. 316 00:10:54,327 --> 00:10:58,531 SO THIS IS AN EXAMPLE OF AI NOW 317 00:10:58,531 --> 00:10:59,465 THAT HAS BEEN IMPLEMENTED AND 318 00:10:59,465 --> 00:11:01,100 THIS IS A BIG DEAL BECAUSE NOT 319 00:11:01,100 --> 00:11:05,037 MUCH OF IT HAS BEEN IMPLEMENTED. 320 00:11:05,037 --> 00:11:06,372 LOTS OF PUBLICATIONS, THOUSANDS 321 00:11:06,372 --> 00:11:07,473 OF PUBLICATIONS BUT WHEN YOU 322 00:11:07,473 --> 00:11:09,242 LOOK AT WHAT IS DONE TO CHANGE 323 00:11:09,242 --> 00:11:10,743 IMMEDIATE, THERE'S NOT THAT 324 00:11:10,743 --> 00:11:11,177 MUCH. 325 00:11:11,177 --> 00:11:14,247 THERE'S ALMOST A THOUSAND AI 326 00:11:14,247 --> 00:11:15,214 ALGORITHMS THAT ARE AUTHORIZED 327 00:11:15,214 --> 00:11:16,415 BY THE FOOTWORK DECKERA BUT MOST 328 00:11:16,415 --> 00:11:19,252 OF THEM ARE NOT IN COMMON USE 329 00:11:19,252 --> 00:11:20,586 AND MOST OF THEM DON'T HAVE 330 00:11:20,586 --> 00:11:21,387 GREAT DATA. 331 00:11:21,387 --> 00:11:24,757 SO THIS IS THE MAYO CLINIC, THEY 332 00:11:24,757 --> 00:11:27,226 DID A RANDOMIZED TRIAL WITH AI 333 00:11:27,226 --> 00:11:29,295 ASSISTED ECG READING AND SHOWED 334 00:11:29,295 --> 00:11:31,397 IT WAS SUPERIOR FOR THE ACCURATE 335 00:11:31,397 --> 00:11:33,533 DIAGNOSIS AS COMPARED TO JUST 336 00:11:33,533 --> 00:11:35,735 THE PHYSICIAN AND NOW THEY'VE 337 00:11:35,735 --> 00:11:37,570 IMPLEMENTED THIS AND THEY ARE 338 00:11:37,570 --> 00:11:39,305 PICKING UP PEOPLE WHO HAVE POOR 339 00:11:39,305 --> 00:11:41,140 EJECTION FRACTION AND VARIOUS 340 00:11:41,140 --> 00:11:41,974 OTHER DIAGNOSIS THAT WOULD 341 00:11:41,974 --> 00:11:44,043 OTHERWISE NOT GET PICKED UP AS 342 00:11:44,043 --> 00:11:46,245 FREQUENTLY BY THE PHYSICIAN. 343 00:11:46,245 --> 00:11:49,248 SO THAT GETS US TO THIS TOPIC OF 344 00:11:49,248 --> 00:11:51,551 MEDICINE EYES OR DIGITAL EYES, 345 00:11:51,551 --> 00:11:55,421 AND WHAT I MEAN BY THAT IS, THE 346 00:11:55,421 --> 00:11:56,822 MACHINE WILL SEE THINGS THAT 347 00:11:56,822 --> 00:11:59,625 HUMANS WILL NEVER SEE. 348 00:11:59,625 --> 00:12:00,226 IT'S ACTUALLY QUITE 349 00:12:00,226 --> 00:12:01,861 EXTRAORDINARY AND THIS IS WHY 350 00:12:01,861 --> 00:12:04,397 THE HOPE FOR IMPROVING ACCURACY 351 00:12:04,397 --> 00:12:07,700 IS SO RICH. 352 00:12:07,700 --> 00:12:09,569 SO THE RETINAL LOCATIONINNA IS A 353 00:12:09,569 --> 00:12:10,403 PROTOTYPIC EXAMPLE. 354 00:12:10,403 --> 00:12:12,605 WHO WOULD HAVE EVER THOUGHT THAT 355 00:12:12,605 --> 00:12:16,609 THROUGH THE RETINA GATEWAY TO 356 00:12:16,609 --> 00:12:18,344 THE HUMAN BODY WE WOULD BE ABLE 357 00:12:18,344 --> 00:12:20,413 TO DETECT ALL THESE THINGS SUCH 358 00:12:20,413 --> 00:12:22,014 AS THE BLOOD PRESSURE AND 359 00:12:22,014 --> 00:12:24,951 DIABETES CONTROL OR AS YOU SEE 360 00:12:24,951 --> 00:12:26,352 HERE, THE KIDNEY DISEASE BEFORE 361 00:12:26,352 --> 00:12:28,421 THERE'S EVER BEEN ANY KIDNEY 362 00:12:28,421 --> 00:12:30,623 DISEASE, THE LIKELIHOOD, LIVER 363 00:12:30,623 --> 00:12:33,459 AND GALLBLADDER DISEASE FROM THE 364 00:12:33,459 --> 00:12:36,495 RETINA, THE HEART CALCIUM SCORE 365 00:12:36,495 --> 00:12:40,166 EQUIVALENT TO A CT OF THE HEART, 366 00:12:40,166 --> 00:12:43,502 OF THE CHEST, BY THE VESSELS IN 367 00:12:43,502 --> 00:12:45,771 THE RETINA, PREDICTING 368 00:12:45,771 --> 00:12:46,372 ALZHEIMER'S DISEASE, 7 YEARS 369 00:12:46,372 --> 00:12:50,543 BEFORE THERE ARE ANY SYMPTOMS, 370 00:12:50,543 --> 00:12:52,612 THE LIKELIHOOD AND THE SAME FOR 371 00:12:52,612 --> 00:12:55,581 PREDICTING STROKE AND HEART 372 00:12:55,581 --> 00:12:58,784 ATTACK RISK, LIPIDEMIA, HYPER 373 00:12:58,784 --> 00:12:59,619 LIPIDEMIA AND PARKINSON'S 374 00:12:59,619 --> 00:13:02,355 DISEASE MORE THAN 5 YEARS BEFORE 375 00:13:02,355 --> 00:13:05,591 ANY SYMPTOMS HAVE BEEN MANIFEST, 376 00:13:05,591 --> 00:13:07,493 THIS IS QUITE EXTRAORDINARY AND 377 00:13:07,493 --> 00:13:10,129 IT'S NOT THE ONLY EXAMPLE. 378 00:13:10,129 --> 00:13:11,864 THE ELECTROCARDIOGRAM WITH 379 00:13:11,864 --> 00:13:12,565 MACHINE EYES PROVIDES 380 00:13:12,565 --> 00:13:14,200 EXTRAORDINARY DATA AS SHEEN HERE 381 00:13:14,200 --> 00:13:16,502 SUMMARIZED, YOU WILL SEE THAT IT 382 00:13:16,502 --> 00:13:19,572 CAN NOT JUST GET THE AGE AND SEX 383 00:13:19,572 --> 00:13:23,075 OF THE PATIENT HAVING READ 384 00:13:23,075 --> 00:13:24,510 CARDIA GRAMS I WOULD NEVER GUESS 385 00:13:24,510 --> 00:13:25,911 WHAT THAT IS AND HERE IT IS 386 00:13:25,911 --> 00:13:27,747 QUITE ACCURATE BUT YAWBD THAT 387 00:13:27,747 --> 00:13:31,917 THE EJECTION FRACTION OF THE 388 00:13:31,917 --> 00:13:32,852 PERSON'S HEART, MANY DIFFICULT 389 00:13:32,852 --> 00:13:36,155 DIAGNOSIS THAT MIGHT BE MISSED, 390 00:13:36,155 --> 00:13:38,691 THE HEMODPLOABIN TO THE DECIMAL 391 00:13:38,691 --> 00:13:40,526 POINT FROM THE CARDIO GRAM. 392 00:13:40,526 --> 00:13:43,396 AND THEN ADD TO THAT IN SOMEONE 393 00:13:43,396 --> 00:13:45,331 WITH NORMAL 69US RHYTHM 394 00:13:45,331 --> 00:13:48,034 PREDICTING WHETHER THEY WILL 395 00:13:48,034 --> 00:13:50,803 EVER HAVE ATRI OLDER PEOPLE 396 00:13:50,803 --> 00:13:53,139 FIBRILLATION OR STROKE FROM A 12 397 00:13:53,139 --> 00:13:53,873 WEEK CARDIOGRAM. 398 00:13:53,873 --> 00:13:54,774 DIABETES NEVER WOULD HAVE 399 00:13:54,774 --> 00:13:57,576 GUESSED THAT 1 OR PREDIABETES 400 00:13:57,576 --> 00:13:59,078 DIAGNOSED FROM THE 401 00:13:59,078 --> 00:14:00,212 ELECTROCARDIOGRAM, THE FILLS 402 00:14:00,212 --> 00:14:03,282 PRESSURE OF THE HEART, 403 00:14:03,282 --> 00:14:04,450 HYPOTHYROIDISM, KIDNEY DISEASE 404 00:14:04,450 --> 00:14:07,853 LIKE THE RETINA, AND 405 00:14:07,853 --> 00:14:08,621 PERIPARADIGM UMKC CARDIOMYOPATHY 406 00:14:08,621 --> 00:14:09,855 LIKELIHOOD EMPLOY SO IT'S JUST 407 00:14:09,855 --> 00:14:11,791 EXTRAORDINARY WHAT CAN BE 408 00:14:11,791 --> 00:14:12,625 GLEANED THROUGH DIGITAL EYES 409 00:14:12,625 --> 00:14:13,926 THAT WE WOULD NOT HAVE 410 00:14:13,926 --> 00:14:15,428 ANTICIPATED AND FOR 411 00:14:15,428 --> 00:14:16,829 PATHOLOGISTS, THIS IS REALLY 412 00:14:16,829 --> 00:14:18,431 ALSO QUITE NOTE WORTHY BECAUSE 413 00:14:18,431 --> 00:14:21,367 IT CAN PICK UP THE MUTATIONS 414 00:14:21,367 --> 00:14:26,539 THAT ARE DRIVING THE PATIENT'S 415 00:14:26,539 --> 00:14:28,841 CANCER FROM THE PATTERN 416 00:14:28,841 --> 00:14:29,975 RECOGNITION HAVING THEM TRAIN 417 00:14:29,975 --> 00:14:31,944 WIDE HUNDREDS OF THOUSANDS OF 418 00:14:31,944 --> 00:14:34,613 PATH SLIDES, WHOLE SLIDE IMAGES, 419 00:14:34,613 --> 00:14:35,881 MALIGNANT VERSUS BENIGN, THE 420 00:14:35,881 --> 00:14:40,119 ORIGIN OF THE TUMOR, THE 421 00:14:40,119 --> 00:14:43,189 STRUCTURAL CNV GENOMIC 422 00:14:43,189 --> 00:14:44,790 UNDERPINNINGS, AND RESPONSE TO 423 00:14:44,790 --> 00:14:45,958 THERAPY ALONG WITH PROGNOSIS, 424 00:14:45,958 --> 00:14:50,296 ALL FROM THE SLIDE, WHO WOULD 425 00:14:50,296 --> 00:14:50,629 HAVE GUESSED. 426 00:14:50,629 --> 00:14:52,598 OKAY, SO THIS IS A RECENT STUDY 427 00:14:52,598 --> 00:14:55,901 THAT LOOKED AT THE CHEST X-RAY, 428 00:14:55,901 --> 00:14:57,269 AND THE CHEST X-RAY ALSO IS 429 00:14:57,269 --> 00:14:57,937 QUITE EXTRAORDINARY BECAUSE WHAT 430 00:14:57,937 --> 00:15:00,940 YOU CAN SEE IS IT PROVIED BETTER 431 00:15:00,940 --> 00:15:04,376 RISK SCORE FOR A BURN DEVELOPING 432 00:15:04,376 --> 00:15:05,277 CARDIOVASCULAR DISEASE THAN RISK 433 00:15:05,277 --> 00:15:09,381 SCORES THAT WE USE TODAY FROM 434 00:15:09,381 --> 00:15:10,583 THE CHEST X-RAY. 435 00:15:10,583 --> 00:15:11,650 SO THIS IS REALLY QUITE 436 00:15:11,650 --> 00:15:14,019 EXTRAORDINARY AND I THINK IT 437 00:15:14,019 --> 00:15:15,554 REALLY CHANGES THE WHOLE WAY WE 438 00:15:15,554 --> 00:15:21,060 MIGHT BE ADDING WAYS TO ASSESSA 439 00:15:21,060 --> 00:15:21,660 PERSON'S CARDIOVASCULAR RISK 440 00:15:21,660 --> 00:15:24,797 EMPLOY THIS FIELD IS CALLED 441 00:15:24,797 --> 00:15:25,765 OPPORTUNISTIC AI AND BASICALLY 442 00:15:25,765 --> 00:15:26,565 WHAT WE'RE TALKING ABOUT HERE IS 443 00:15:26,565 --> 00:15:27,733 THAT YOU GET A SCAN FOR SOME 444 00:15:27,733 --> 00:15:31,937 REASON AND YOU GET ALL THESE 445 00:15:31,937 --> 00:15:34,306 OTHER EXTRA ADDED OUTPUTS THAT 446 00:15:34,306 --> 00:15:35,975 MAYBE USEFUL WHICH IS WHAT WE'RE 447 00:15:35,975 --> 00:15:37,676 LEARNING FROM DIGITAL EYES. 448 00:15:37,676 --> 00:15:38,377 IT'S PRETTY EXCITING. 449 00:15:38,377 --> 00:15:40,646 OKAY, SO NOW LET ME MOVE TO THE 450 00:15:40,646 --> 00:15:42,281 CURRENT ERA, BECAUSE THAT'S AN 451 00:15:42,281 --> 00:15:45,918 OUTGROWTH OF DEEP LEARNING, THAT 452 00:15:45,918 --> 00:15:47,953 THE MULTIMODAL AI IS WHERE WE 453 00:15:47,953 --> 00:15:49,588 ARE NOW WITH SELF-SUPER VISED 454 00:15:49,588 --> 00:15:50,990 THE DATA IS BASICALLY LEARNING 455 00:15:50,990 --> 00:15:52,525 FROM ITSELF RATHER THAN 456 00:15:52,525 --> 00:15:54,593 REQUIRING ANNOTATIONS BY EXPERTS 457 00:15:54,593 --> 00:15:56,629 WHICH IS WELL ALINED IN MEDICINE 458 00:15:56,629 --> 00:16:01,400 SINCE WE DON'T HAVE EXPERTS WITH 459 00:16:01,400 --> 00:16:03,936 TIME TO ANNOTATE HUNDREDS OF 460 00:16:03,936 --> 00:16:04,670 THOUSANDS OF IMAGES. 461 00:16:04,670 --> 00:16:06,105 SO YOU MAY KNOW THAT THIS WHOLE 462 00:16:06,105 --> 00:16:11,010 ERA OF DEEP LEARNING AND NOW, 463 00:16:11,010 --> 00:16:12,244 TRANSFORMER MODELS ONLY BECAME 464 00:16:12,244 --> 00:16:13,779 POSSIBLE BECAUSE OF VIDEO GAMES. 465 00:16:13,779 --> 00:16:17,416 I MEAN IS THAT AMAZING. 466 00:16:17,416 --> 00:16:21,153 IT WAS THE NVIDIA CHIPS THAT 467 00:16:21,153 --> 00:16:23,656 BACK AT THE TURN OF THE CENTURY 468 00:16:23,656 --> 00:16:26,091 THAT MADE THE COMPUTING POSSIBLE 469 00:16:26,091 --> 00:16:26,992 TO CHANGE IMMEDIATE AND 470 00:16:26,992 --> 00:16:28,427 EVERYTHING IN OUR DAILY LIVES 471 00:16:28,427 --> 00:16:35,734 AND NOW WE'RE AT A POINT WITH 472 00:16:35,734 --> 00:16:37,102 THE LATEST NVIDIA CHIP THAT IT 473 00:16:37,102 --> 00:16:39,405 HAS OVER 2 BILLION TRANSISTORS 474 00:16:39,405 --> 00:16:41,540 IN IT, AND THAT'S PRETTY 475 00:16:41,540 --> 00:16:42,708 STRIKING, SO WHAT WE'RE TALKING 476 00:16:42,708 --> 00:16:46,278 ABOUT HERE IS GOOD OLD DEEP 477 00:16:46,278 --> 00:16:48,881 NEURAL NETWORKS THAT WERE NOW, 478 00:16:48,881 --> 00:16:50,583 7, 8, 9, YEARS REALLY CHANGING 479 00:16:50,583 --> 00:16:52,284 THE LOOKS OF A LOT OF STUDIES I 480 00:16:52,284 --> 00:16:55,321 WAS MENTIONING TO NOW, SWITCHING 481 00:16:55,321 --> 00:16:59,024 TO TRANSFORMER MODELS THAT ARE 482 00:16:59,024 --> 00:16:59,525 BASICALLY MULTIATTENTION 483 00:16:59,525 --> 00:17:04,096 MODELINGS THAT CAN BRING IN THE 484 00:17:04,096 --> 00:17:05,297 CONTEXT IN THE DATA IN THE FORM 485 00:17:05,297 --> 00:17:07,299 OF TOKENS AND THIS IS REALLY 486 00:17:07,299 --> 00:17:10,870 CHANGING, THAT'S WHAT REALLY SET 487 00:17:10,870 --> 00:17:14,039 THE WHOLE WORLD INTO MOTION WITH 488 00:17:14,039 --> 00:17:17,142 EXCITEMENT OF CHAT BOTS AND LIKE 489 00:17:17,142 --> 00:17:20,179 GPT AND SUBSEQUENT 1S. 490 00:17:20,179 --> 00:17:22,681 DPT 4 UP UNTIL SEVERAL MONTHS 491 00:17:22,681 --> 00:17:25,150 AGO WAS REGARDED AS THE MOST 492 00:17:25,150 --> 00:17:26,652 ADVANCED LARGE LANGUAGE MODEL. 493 00:17:26,652 --> 00:17:31,423 AND WHAT'S INTERESTING ABOUT 494 00:17:31,423 --> 00:17:34,193 THAT IS THAT IT HAS OVER 495 00:17:34,193 --> 00:17:35,895 1 TRILLION PARAMETERS OR 496 00:17:35,895 --> 00:17:37,496 CONNECTIONS AND WHAT'S ALSO 497 00:17:37,496 --> 00:17:40,933 INTERESTING THAT IT REQUIRED 498 00:17:40,933 --> 00:17:43,469 24,000 GRAPHIC PROCESSING UNITS. 499 00:17:43,469 --> 00:17:45,704 NOW DO HAVE YOU ANY IDEA HOW 500 00:17:45,704 --> 00:17:46,639 MANY GRAPHIC PROCESSING UNITS 501 00:17:46,639 --> 00:17:51,543 ARE ON THE NIH CAMPUS? 502 00:17:51,543 --> 00:17:51,877 ANYBODY? 503 00:17:51,877 --> 00:17:53,979 MIGHT BE A COUPLE HUNDRED, 504 00:17:53,979 --> 00:17:54,213 MAYBE. 505 00:17:54,213 --> 00:17:55,481 THERE'S NOT THAT MANY AT 506 00:17:55,481 --> 00:17:57,216 STANFORD OR M. I.T. AND THIS IS 507 00:17:57,216 --> 00:18:00,319 WHAT IS NOW THE ARMS RACE OF 508 00:18:00,319 --> 00:18:03,422 GPUs TO DO THESE BASE LARGE 509 00:18:03,422 --> 00:18:04,123 LANGUAGE MODELS. 510 00:18:04,123 --> 00:18:07,993 AND THE AMOUNT OF TRAINING COSTS 511 00:18:07,993 --> 00:18:09,461 INVOLVED IN SETTING UP THESE 512 00:18:09,461 --> 00:18:10,696 MODELS BECAUSE OF THE MASSIVE 513 00:18:10,696 --> 00:18:13,699 DATA THAT NEEDS TO BE PROCESSED 514 00:18:13,699 --> 00:18:15,134 IS REALLY AMAZING, IT'S WORTH 515 00:18:15,134 --> 00:18:16,402 TALKING ABOUT MILLIONS TO A 516 00:18:16,402 --> 00:18:18,437 COUPLE HUNDRED MILLION FOR SOME 517 00:18:18,437 --> 00:18:20,205 OF THE RECENT TRANSFORMER 518 00:18:20,205 --> 00:18:20,773 MODELS. 519 00:18:20,773 --> 00:18:22,308 IN ANY CASE, DESPITE THESE 520 00:18:22,308 --> 00:18:23,809 LIMITATIONS, THIS IS SETTING UP 521 00:18:23,809 --> 00:18:27,246 A LOT OF NEW APPLICATIONS IN 522 00:18:27,246 --> 00:18:29,715 HEALTHCARE, IN MEDICINE, AND 523 00:18:29,715 --> 00:18:31,183 THAT WAS A SECOND PIECE I WROTE 524 00:18:31,183 --> 00:18:33,686 FOR SCIENCE LAST YEAR. 525 00:18:33,686 --> 00:18:37,756 SO ALL THESE LAYERS OF THE HUMAN 526 00:18:37,756 --> 00:18:40,960 THAT MAKE US UNIQUE, EVERYTHING 527 00:18:40,960 --> 00:18:43,028 FROM THE ANATOMY, THE 528 00:18:43,028 --> 00:18:47,399 PHYSIOLOGY, ALL THE DIFFERENT 529 00:18:47,399 --> 00:18:52,237 BIOLOGY OMICs, DNA, RNA, 530 00:18:52,237 --> 00:18:53,172 EPIGENOMICS, MICROBIOME, METRICS 531 00:18:53,172 --> 00:18:55,874 TAB O LOAM, AND THE EXPOETICSOME 532 00:18:55,874 --> 00:18:58,277 ARE EXPOSURES, COLLECTIVELY AND 533 00:18:58,277 --> 00:19:01,013 THE IMMUNE OME MAKE US UNIQUE 534 00:19:01,013 --> 00:19:02,448 AND INTERESTINGLY IT ISN'T JUST 535 00:19:02,448 --> 00:19:04,917 A 1-OFF, MAYBE A DNA SEQUENCE IS 536 00:19:04,917 --> 00:19:06,218 A 1-OFF BUT MANY OTHERS ARE 537 00:19:06,218 --> 00:19:08,053 DYNAMIC AND THIS IS OF COURSE 538 00:19:08,053 --> 00:19:12,458 THE IDEA THAT WE HAVE TO 539 00:19:12,458 --> 00:19:17,363 SERIALLY ASSESS THESE DATA OVER 540 00:19:17,363 --> 00:19:18,163 A PERSON'S LIFE. 541 00:19:18,163 --> 00:19:20,632 SO RECENTLY AS OF LAST YEAR 542 00:19:20,632 --> 00:19:24,403 PIERCE KEEN AT MOOREFIELD'S--THE 543 00:19:24,403 --> 00:19:25,838 MOST PREOF THE IEJOUS EYE 544 00:19:25,838 --> 00:19:28,107 INSTITUTE IN THE WORLD--WE HAD A 545 00:19:28,107 --> 00:19:29,174 CHANCE TO COLLABORATE WITH HIS 546 00:19:29,174 --> 00:19:33,379 GROUP BECAUSE THEY HAD 547 00:19:33,379 --> 00:19:35,280 1.6 MILLION RETINAL IMAGES AND 548 00:19:35,280 --> 00:19:37,616 THEY DID A SO CALLED FOUNDATION 549 00:19:37,616 --> 00:19:37,850 MODEL. 550 00:19:37,850 --> 00:19:39,651 SO REMEMBER THAT RETINA THAT I 551 00:19:39,651 --> 00:19:40,786 SHOWED 8 DIFFERENT THINGS IT 552 00:19:40,786 --> 00:19:42,254 COULD DO, THEY DID IT ALL IN 1 553 00:19:42,254 --> 00:19:44,690 MODEL WITH THE MASSIVE RETINAL 554 00:19:44,690 --> 00:19:46,725 IMAGES WITH THE CLINICAL 555 00:19:46,725 --> 00:19:47,860 FEATURES PREDICTING PARKINSON'S 556 00:19:47,860 --> 00:19:48,794 DISEASE, HEART DISEASE, KIDNEY 557 00:19:48,794 --> 00:19:50,362 DISEASE AND EVERYTHING ELSE IT 558 00:19:50,362 --> 00:19:52,264 WAS PREVIOUSLY MENTIONED. 559 00:19:52,264 --> 00:19:57,102 SO THAT'S WHAT'S REALLY SEEN AS 560 00:19:57,102 --> 00:20:00,339 THE FOUNDATION MODEL, THIS 561 00:20:00,339 --> 00:20:01,373 MULTIMODAL, MULTITASK CAPABILITY 562 00:20:01,373 --> 00:20:03,776 WHICH HAS NOW BEEN BROUGHT IN BY 563 00:20:03,776 --> 00:20:05,110 THESE NEW MODELS, THESE LARGE 564 00:20:05,110 --> 00:20:08,147 LANGUAGE MODELS. 565 00:20:08,147 --> 00:20:11,417 SO, THIS WEEK, THERE WAS A PAPER 566 00:20:11,417 --> 00:20:12,785 IN NATURE. 567 00:20:12,785 --> 00:20:14,953 THIS IS ON CANCER PATHOLOGY AND 568 00:20:14,953 --> 00:20:19,425 THIS IS THE MOST ADVANCED CANCER 569 00:20:19,425 --> 00:20:20,559 PATHOLOGY FOUNDATION MODEL YET 570 00:20:20,559 --> 00:20:23,262 AND IT BASICALLY IS PICKING UP 571 00:20:23,262 --> 00:20:26,031 NOT JUST A DIAGNOSIS, BUT AS I 572 00:20:26,031 --> 00:20:29,001 SHOWED YOU THE DIFFERENT DRIVING 573 00:20:29,001 --> 00:20:29,735 MUTATIONS, THE PROGNOSIS OF THE 574 00:20:29,735 --> 00:20:34,139 PATIENT AND THESE CIRCLES HERE, 575 00:20:34,139 --> 00:20:35,974 ARE BASICALLY SHOWING ITS 576 00:20:35,974 --> 00:20:37,176 PERFORMANCE AGAINST PRIOR 577 00:20:37,176 --> 00:20:38,243 PATHOLOGY AI MODELS, IT JUST 578 00:20:38,243 --> 00:20:39,445 KEEPING GET BEING BETTER AND 579 00:20:39,445 --> 00:20:40,946 BETTER FOR DOING ALL THESE 580 00:20:40,946 --> 00:20:41,713 THINGS. 581 00:20:41,713 --> 00:20:44,349 NOW THE NEXT THING IT'S CAPABLE 582 00:20:44,349 --> 00:20:50,522 OF DOING WHICH IS EXCITING IS A 583 00:20:50,522 --> 00:20:50,956 DIFFICULT DIAGNOSIS. 584 00:20:50,956 --> 00:20:53,492 WRASSE PATIENTS COME TO THE NIH 585 00:20:53,492 --> 00:20:55,027 CLINICAL CENTER WITH IDIO PATHIC 586 00:20:55,027 --> 00:20:58,530 CONDITIONS NOW WE HAVE THE 587 00:20:58,530 --> 00:21:01,600 ADVANTAGE OF SAYING TO GPT4 OR 588 00:21:01,600 --> 00:21:03,135 GEMINA ULTRA OR WHATEVER, WHAT 589 00:21:03,135 --> 00:21:06,872 IS THE DIAGNOSIS OF THE PATIENT? 590 00:21:06,872 --> 00:21:10,008 HERE'S AN EXAMPLE, INTERESTINGLY 591 00:21:10,008 --> 00:21:12,077 THIS BOY'S CASE WAS FIRST 592 00:21:12,077 --> 00:21:16,181 PRESENTED ON THE TODAY SHOW OF 593 00:21:16,181 --> 00:21:19,084 ALL PLACES, BUT IT'S A TAC 594 00:21:19,084 --> 00:21:20,786 FASCINATING CASE BECAUSE HE HAD 595 00:21:20,786 --> 00:21:23,288 3 YEARS OF RELENTLESS WORSENING 596 00:21:23,288 --> 00:21:25,190 PAIN, DIFFICULTY WALKING, 597 00:21:25,190 --> 00:21:27,893 HEADACHES, AND FAILURE TO GROW. 598 00:21:27,893 --> 00:21:31,330 AND HIS MOTHER AND FATHER IN 599 00:21:31,330 --> 00:21:32,865 MICHIGAN TOOK HIM TO 17 DOCTORS, 600 00:21:32,865 --> 00:21:35,000 NOT EVEN JUST IN MICHIGAN TO GET 601 00:21:35,000 --> 00:21:35,901 HIM EVALUATED AND THEY COULDN'T 602 00:21:35,901 --> 00:21:38,570 COME UP WITH A DIAGNOSIS, SO 603 00:21:38,570 --> 00:21:40,706 ULTIMATELY OUT OF FRUSTRATION 604 00:21:40,706 --> 00:21:42,808 SHE ENTERED ALL HIS DATA INTO 605 00:21:42,808 --> 00:21:43,909 CHAT GPT AND WITHIN SECONDS IT 606 00:21:43,909 --> 00:21:50,749 CAME OUT WITH THE DIAGNOSIS OF A 607 00:21:50,749 --> 00:21:52,417 SPINA BIFIDA AND SHE WENT TO A 608 00:21:52,417 --> 00:21:54,086 NEUROSURGEON AND INDEED THAT'S 609 00:21:54,086 --> 00:21:55,521 WHAT HE HAD AND HE HAD SURGERY 610 00:21:55,521 --> 00:21:57,589 TO CORRECT THIS AND HE'S DONE 611 00:21:57,589 --> 00:21:58,790 EXTREMELY WELL. 612 00:21:58,790 --> 00:22:01,126 THIS IS 1 OF MANY ANECDOTES OF 613 00:22:01,126 --> 00:22:03,328 DIFFICULT CASES THAT WERE 614 00:22:03,328 --> 00:22:05,264 STUMPING VARIOUS PHYSICIANS, 1 I 615 00:22:05,264 --> 00:22:07,466 WAS INVOLVED IN WAS THIS PATIENT 616 00:22:07,466 --> 00:22:10,602 WHOSE SISTER WAS TOLD SHE HAD 617 00:22:10,602 --> 00:22:13,005 LONG COVID WHICH HAS YOU KNOW 618 00:22:13,005 --> 00:22:14,373 HAS NO VALIDATED TREATMENT. 619 00:22:14,373 --> 00:22:17,709 SO SHE PUT IN ALL OR DATA TO 620 00:22:17,709 --> 00:22:19,778 CHAT GPT AND IT SAID SHE DOESN'T 621 00:22:19,778 --> 00:22:24,416 HAVE LONG COVID, SHE HAS EMILY 622 00:22:24,416 --> 00:22:25,851 CIBOL ENCEPHALITIS WHICH IS 623 00:22:25,851 --> 00:22:26,552 IMMINENTLY TREATABLE, SHE WAS 624 00:22:26,552 --> 00:22:32,424 TREATED AND SHE FULLY RECOVERED. 625 00:22:32,424 --> 00:22:35,127 SO THERE'S THE MASS GENERAL 626 00:22:35,127 --> 00:22:36,995 HOSPITAL CASE CPC'S CLINICAL 627 00:22:36,995 --> 00:22:38,397 PATHOLOGIC CONFERENCE OF THE 628 00:22:38,397 --> 00:22:40,365 STUMP THE EXPERT, JUST A MASTER 629 00:22:40,365 --> 00:22:42,067 CLINICIAN OF MAN'S GREATEST 630 00:22:42,067 --> 00:22:43,569 HOSPITAL AND THIS HAS BEEN GOING 631 00:22:43,569 --> 00:22:46,238 ON FOR A HUNDRED YEARS, RIGHT? 632 00:22:46,238 --> 00:22:49,608 WELL, TURNS OUT GUESS WHAT? 633 00:22:49,608 --> 00:22:51,310 WHEN YOU PUT A LARGE LANGUAGE 634 00:22:51,310 --> 00:22:53,512 MODEL UP AGAINST THE MASTER 635 00:22:53,512 --> 00:22:54,513 CLINICIANS, IT DOES EXTREMELY 636 00:22:54,513 --> 00:22:58,217 WELL, IT DOES AS WELL AS THE 637 00:22:58,217 --> 00:22:59,885 MASTER CLINICIANS, THESE ARE 638 00:22:59,885 --> 00:23:01,253 CASES, ACTUAL CASES. 639 00:23:01,253 --> 00:23:02,921 300 NEW ENGLAND JOURNAL CPCs, 640 00:23:02,921 --> 00:23:05,123 THEY DID A RANDOMIZED TRIAL AND 641 00:23:05,123 --> 00:23:06,892 YOU SEE THE LARGE LANGUAGE MODEL 642 00:23:06,892 --> 00:23:10,495 DID BETTER THAN THE MASTER 643 00:23:10,495 --> 00:23:13,198 CLINICIANS AND HERE IS ANOTHER 644 00:23:13,198 --> 00:23:13,332 1. 645 00:23:13,332 --> 00:23:16,068 THIS IS A GOOGLE DEEP MIND LARGE 646 00:23:16,068 --> 00:23:19,037 LANGUAGE MODEL, IT LOOKED AT 26 647 00:23:19,037 --> 00:23:20,606 DIFFERENT OUTCOMES, NOT JUST 648 00:23:20,606 --> 00:23:21,873 DIAGNOSTIC ACCURACY BUT WE'RE 649 00:23:21,873 --> 00:23:25,777 TALKING ABOUT EXAM SKILLS, 650 00:23:25,777 --> 00:23:26,445 PATIENT-CENTERED COMMUNICATION, 651 00:23:26,445 --> 00:23:29,081 EMPATHY, I WILL COME BACK TO 652 00:23:29,081 --> 00:23:30,415 THAT, WOULD HAVE THOUGHT? 653 00:23:30,415 --> 00:23:31,883 DIAGNOSIS AND MANAGE AM. 654 00:23:31,883 --> 00:23:33,485 OUT OF 26 DIFFERENT TASKS THE 655 00:23:33,485 --> 00:23:34,920 LARGE LANGUAGE MODEL DID BETTER 656 00:23:34,920 --> 00:23:36,688 THAN THE PHYSICIAN. 657 00:23:36,688 --> 00:23:39,091 SO, THIS WAS THE NATURE COVERAGE 658 00:23:39,091 --> 00:23:41,627 OF THAT PAPER, GOOGLE AI HAS 659 00:23:41,627 --> 00:23:44,296 BETTER BEDSIDE MANNER THAT IS 660 00:23:44,296 --> 00:23:46,431 HUMAN DOCTORS, THAT DOESN'T 661 00:23:46,431 --> 00:23:51,003 SOUND VERY UNCOURAGING DOES IT? 662 00:23:51,003 --> 00:23:53,171 OKAY, THIS WAS ANOTHER STUDY 663 00:23:53,171 --> 00:23:55,007 PUBLISHED MORE RECENTLY IN 664 00:23:55,007 --> 00:23:58,076 NATURE MEDICINE SHOWING HOW 665 00:23:58,076 --> 00:24:01,346 COMPARING THE GPT 4 TO PHYSICIAN 666 00:24:01,346 --> 00:24:04,549 FOR THE NOTES, COMPLETENESS, 667 00:24:04,549 --> 00:24:07,386 CORRECTNESS AND CONCISENESS AND 668 00:24:07,386 --> 00:24:10,122 YOU CAN SEE GPT4 WAS SUPERIOR 669 00:24:10,122 --> 00:24:12,858 AND THEY LOOKED AT VARIOUS OTHER 670 00:24:12,858 --> 00:24:13,825 LARGE LANGUAGE MODELS AS WELL 671 00:24:13,825 --> 00:24:15,961 AND THIS IS JUST AN EXAMPLE OF 672 00:24:15,961 --> 00:24:18,597 HOW THIS WAS PROCESSED BY THE 673 00:24:18,597 --> 00:24:20,999 LARGE LANGUAGE MODEL, ALL THIS 674 00:24:20,999 --> 00:24:21,800 DIFFERENT DATA PROCESSED 675 00:24:21,800 --> 00:24:26,405 CORRECTLY BY THE LARGE LANGUAGE 676 00:24:26,405 --> 00:24:27,739 MODEL AND INCORRECTLY, THE 677 00:24:27,739 --> 00:24:28,774 EXPERT, THE MEDICAL EXPERT 678 00:24:28,774 --> 00:24:31,777 MISSED THE THINGS IN GREEN, SO 679 00:24:31,777 --> 00:24:32,978 TOGETHER, THE ACCURACY IS 680 00:24:32,978 --> 00:24:34,446 GREATLY FOSTERED AND THE 681 00:24:34,446 --> 00:24:36,882 CONCLUSION OF THAT STUDY WAS 682 00:24:36,882 --> 00:24:39,351 INCORPORATING LARGE LANGUAGE 683 00:24:39,351 --> 00:24:41,753 MODELS WILL MAKE THESE SUMMARIES 684 00:24:41,753 --> 00:24:43,388 OF DOCUMENTATION SYNTHESIS OF A 685 00:24:43,388 --> 00:24:46,358 PATIENT'S RECORD MUCH BETTER, 686 00:24:46,358 --> 00:24:47,759 LESS STRAIN, TIME FOR THE 687 00:24:47,759 --> 00:24:51,797 CLINICIAN AND IMPROVING PATIENT 688 00:24:51,797 --> 00:24:51,997 CARE. 689 00:24:51,997 --> 00:24:55,567 SO NOW WE HAVE THESE MULTIMODAL 690 00:24:55,567 --> 00:24:56,968 MODELS THAT ARE FLOURISHING AND 691 00:24:56,968 --> 00:24:59,371 BEING PUBLISHED ON A FREQUENT 692 00:24:59,371 --> 00:25:00,272 BASIS. 693 00:25:00,272 --> 00:25:04,176 THIS WAS FROM LAST MONTH, BIOMED 694 00:25:04,176 --> 00:25:05,677 GPT TAKING IN ALL THESE 695 00:25:05,677 --> 00:25:06,878 DIFFERENT MODES OF DAILY BASIS 696 00:25:06,878 --> 00:25:08,980 THEA, LAYERS OF DATA FROM THE 697 00:25:08,980 --> 00:25:10,482 MEDICAL LITERATURE, FROM THE 698 00:25:10,482 --> 00:25:11,650 ELECTRONIC HELT RECORD, EVERY 699 00:25:11,650 --> 00:25:14,686 TYPE OF SCAN, REALLY QUITE 700 00:25:14,686 --> 00:25:14,986 IMPRESSIVE. 701 00:25:14,986 --> 00:25:18,757 THIS IS ANOTHER NOTE WORTHY 1 702 00:25:18,757 --> 00:25:20,959 BECAUSE AS YOU KNOW DIABETIC 703 00:25:20,959 --> 00:25:21,793 RETINAL LOCATION NUMBER OF 704 00:25:21,793 --> 00:25:22,728 PATIENTSATHY IS FREQUENTLY 705 00:25:22,728 --> 00:25:23,962 MISSED BECAUSE THERE'S NO SCAN 706 00:25:23,962 --> 00:25:28,066 OF THE RETINA IN HALF OF 707 00:25:28,066 --> 00:25:31,002 DIABETICS AND IT'S AN IMMINENTLY 708 00:25:31,002 --> 00:25:32,804 PREVENTIBLE FORM OF PLIENDNESS 709 00:25:32,804 --> 00:25:33,839 AMONG DIABETICS. 710 00:25:33,839 --> 00:25:37,376 WELL, THIS WAS A DEEP DR, 711 00:25:37,376 --> 00:25:38,877 DIABETIC RETINAL LOCATION NUMBER 712 00:25:38,877 --> 00:25:40,545 OF PATIENTSATHY NLM, AND IT 713 00:25:40,545 --> 00:25:41,480 SHOWED 800S PATIENTS WERE 714 00:25:41,480 --> 00:25:42,814 ASSESSED AND WHEN THE PRIMARY 715 00:25:42,814 --> 00:25:46,184 CARE PHYSICIAN WAS GIVEN THIS 716 00:25:46,184 --> 00:25:47,986 TOOL THE ACCURACY WAS IMPROVED 717 00:25:47,986 --> 00:25:49,855 SUBSTANTIALLY AND THEN IT WAS 718 00:25:49,855 --> 00:25:50,555 RAIDED BY ENDOCRINOLOGIST AS 719 00:25:50,555 --> 00:25:55,060 WELL AS BY PATIENTS AND YOU CAN 720 00:25:55,060 --> 00:25:57,129 SEE BY PATIENTS THEY WERE RATING 721 00:25:57,129 --> 00:25:59,097 THE AI TO MORE EMPATHETIC THAN 722 00:25:59,097 --> 00:25:59,898 THE PHYSICIAN. 723 00:25:59,898 --> 00:26:03,935 AND THEN THE ENDOCRINOLOGIST, 724 00:26:03,935 --> 00:26:08,173 THAT'S AGAIN RATING THIS, THEY 725 00:26:08,173 --> 00:26:09,908 ALSO RATED THE COMBINATION OF 726 00:26:09,908 --> 00:26:11,143 THE PC P, PRIMARY CARE PHYSICIAN 727 00:26:11,143 --> 00:26:15,647 AND THE AI TO BE SUPERIOR TO 728 00:26:15,647 --> 00:26:16,014 PHYSICIANS. 729 00:26:16,014 --> 00:26:17,849 THIS WASN'T ANTICIPATED BUT IT'S 730 00:26:17,849 --> 00:26:19,017 CONSISTENTLY SHOWN NOW IN 10 731 00:26:19,017 --> 00:26:21,920 DIFFERENT STUDIES THAT YOU CAN 732 00:26:21,920 --> 00:26:23,789 BOLSTER EMPATHY BY HAVING A 733 00:26:23,789 --> 00:26:27,426 LARGE LANGUAGE MODEL ASSIST A 734 00:26:27,426 --> 00:26:28,059 PHYSICIAN, PASQUANTINOINATEING. 735 00:26:28,059 --> 00:26:31,663 SO WE WROTE A REVIEW NOW A 736 00:26:31,663 --> 00:26:33,498 COUPLE YEARS AGO, IT'S ALREADY 737 00:26:33,498 --> 00:26:34,299 WELL SURPASSED BUT MANY 738 00:26:34,299 --> 00:26:36,501 DIFFERENT WAYS WE CAN USE 739 00:26:36,501 --> 00:26:38,236 MULTIMODAL AI IN THE FUTURE. 740 00:26:38,236 --> 00:26:40,505 ONE OF THOSE IS HOSPITAL AT 741 00:26:40,505 --> 00:26:43,308 HOME, REVIEW ARTICLE WE 742 00:26:43,308 --> 00:26:44,242 PUBLISHED BECAUSE WHEN HAVE YOU 743 00:26:44,242 --> 00:26:46,678 A PERSON WHO CAN HAVE SENSORS 744 00:26:46,678 --> 00:26:48,580 ON, CAN HAVE EXQUISITE 745 00:26:48,580 --> 00:26:50,415 MONITORING, WHY WOULD THEY BE 746 00:26:50,415 --> 00:26:51,283 ADMITTED TO THE HOSPITAL WHEN 747 00:26:51,283 --> 00:26:53,285 THEY CAN BE IN THE CONVENIENCE 748 00:26:53,285 --> 00:26:57,722 OF THEIR OWN HOME WHICH IS A LOW 749 00:26:57,722 --> 00:26:59,157 EXPENSE AND HIGH COMFORT AND BE 750 00:26:59,157 --> 00:27:00,725 WITH THEIR FAMILIES AND THIS IS 751 00:27:00,725 --> 00:27:02,661 THE FUTURE, NOT FOR PEOPLE IN AN 752 00:27:02,661 --> 00:27:04,830 INTENSIVE CARE UNIT BUT PER 753 00:27:04,830 --> 00:27:06,198 THOSE WHO WOULD OTHERWISE BE IN 754 00:27:06,198 --> 00:27:07,265 A REGULAR HOSPITAL ROOM. 755 00:27:07,265 --> 00:27:14,105 NOW TO GET TO THIS NEXT SECTIO, 756 00:27:14,105 --> 00:27:15,040 ESPECIALLY TAKING A STEP FORWARD 757 00:27:15,040 --> 00:27:16,775 BUT 1 I THINK WILL HAVE QUITE A 758 00:27:16,775 --> 00:27:21,346 LARGE IMPACT IN THE TIMES AHEAD 759 00:27:21,346 --> 00:27:24,816 AND THAT'S FORECASTING. 760 00:27:24,816 --> 00:27:27,819 SO YOU PROBABLY KNOW ALTHOUGH IT 761 00:27:27,819 --> 00:27:30,188 HASPT BEEN UNLEASHED ON THE 762 00:27:30,188 --> 00:27:32,090 WORLD, BUT WEATHER FORECASTING 763 00:27:32,090 --> 00:27:34,493 WITH LARGE LANGUAGE MODELS THE 1 764 00:27:34,493 --> 00:27:36,795 PROTOTYPE IS CALLED GRAPH CAST, 765 00:27:36,795 --> 00:27:38,930 IT HAS WEATHER FORECASTING 766 00:27:38,930 --> 00:27:40,398 ACCURATE TO 99.7% WHICH IS A BIG 767 00:27:40,398 --> 00:27:44,436 JUMP FROM WHAT WE HAVE TODAY 768 00:27:44,436 --> 00:27:46,404 MUCH WE OFTEN MAKE JOKES ABOUT 769 00:27:46,404 --> 00:27:47,539 THE WEATHER PREDICTIONS, THAT'S 770 00:27:47,539 --> 00:27:48,874 PROBABLY GOING TO BECOME A THING 771 00:27:48,874 --> 00:27:49,541 OF THE PAST. 772 00:27:49,541 --> 00:27:51,943 WHAT IF WE HAVE MEDICAL 773 00:27:51,943 --> 00:27:54,045 FORECASTING THAT'S AS GOOD AS 774 00:27:54,045 --> 00:27:55,714 THE WEATHER FORECASTING WE'RE 775 00:27:55,714 --> 00:27:56,648 SEEING WITH LARGE LANGUAGE 776 00:27:56,648 --> 00:27:59,518 MODELS AND HERE IS THE 777 00:27:59,518 --> 00:28:02,320 OPPORTUNITY OUTLINED BECAUSE 778 00:28:02,320 --> 00:28:04,222 AGAIN, ANOTHER SCIENCE INVITED 779 00:28:04,222 --> 00:28:09,861 PIECE ABOUT THIS POTENTIAL. 780 00:28:09,861 --> 00:28:11,029 WE HAVE THESE 5 DOMAINS THAT ARE 781 00:28:11,029 --> 00:28:12,097 COME TOGETHER NOW THAT ARE 782 00:28:12,097 --> 00:28:14,366 GIVING US ABILITY TO PREDICT AND 783 00:28:14,366 --> 00:28:15,767 FORECAST THINGS IN MEDICINE AT 784 00:28:15,767 --> 00:28:20,138 THE INDIVIDUAL LEVEL THAT WE 785 00:28:20,138 --> 00:28:20,805 NEVER HAD BEFORE. 786 00:28:20,805 --> 00:28:22,807 WHETHER IT'S THE VARIOUS 787 00:28:22,807 --> 00:28:24,075 OMICs, THE CELLS THAT WE HAVE 788 00:28:24,075 --> 00:28:25,710 SO MUCH THE NEWER DRUGS AND 789 00:28:25,710 --> 00:28:27,779 VACCINES THAT ARE AVAILABLE AND 790 00:28:27,779 --> 00:28:28,847 THE LIFESTYLE + AS A CALL IT 791 00:28:28,847 --> 00:28:30,248 BECAUSE WE'RE TALKING ABOUT 792 00:28:30,248 --> 00:28:31,983 EXPOSURES, WE'RE TALKING ABOUT A 793 00:28:31,983 --> 00:28:36,955 LOT MORE THAN JUST SLEEP, 794 00:28:36,955 --> 00:28:37,756 EXERCISE AND DIET. 795 00:28:37,756 --> 00:28:43,128 SO WITH THAT, THE ABILITY TO 796 00:28:43,128 --> 00:28:44,262 PREDICT A PERSON'S POLYGENIC 797 00:28:44,262 --> 00:28:46,031 RISK SCORE THAT,'S ONLY 1 LAYER 798 00:28:46,031 --> 00:28:47,799 OF DATA AND IT'S STILL BEING 799 00:28:47,799 --> 00:28:49,668 DEBATED HOW MUCH EVALUATED AS 800 00:28:49,668 --> 00:28:55,907 WELL AS 1 LAYER WITH ORTHOGONAL 801 00:28:55,907 --> 00:28:57,576 LAYERS TO COMPLEMENT IT, IT'S 802 00:28:57,576 --> 00:28:58,143 EXTREMELY VALUABLE. 803 00:28:58,143 --> 00:29:00,111 WE SHOULD BE USING POLYGENIC 804 00:29:00,111 --> 00:29:01,746 RISK SCORES IN ALL OUR PATIENT 805 00:29:01,746 --> 00:29:03,348 BUT WE'RE NOT AND IT'S 806 00:29:03,348 --> 00:29:04,783 UNFORTUNATE BECAUSE IT'S AN 807 00:29:04,783 --> 00:29:06,818 INEXPENSIVE WAY TO GET AN 808 00:29:06,818 --> 00:29:08,653 ASSESSMENT OF COMMON DISEASE 809 00:29:08,653 --> 00:29:09,621 ANDS TO COMPLEMENT OTHER TOOLS 810 00:29:09,621 --> 00:29:10,221 WE HAVE. 811 00:29:10,221 --> 00:29:12,290 THIS IS A REVIEW WE RECENTLY 812 00:29:12,290 --> 00:29:13,792 WROTE ON CARDIO METABOLIC RISK 813 00:29:13,792 --> 00:29:15,427 THAT YOU COULD PREDICT AT THE 814 00:29:15,427 --> 00:29:20,165 INDIVIDUAL LEVEL USING MULTIPLE 815 00:29:20,165 --> 00:29:21,099 LAYERS AND REBOOTING SCREENING 816 00:29:21,099 --> 00:29:26,071 FOR CANCER WHICH IS OF 817 00:29:26,071 --> 00:29:28,640 CONSIDERABLE WASTE, USES THE 818 00:29:28,640 --> 00:29:30,609 DUMB-DOWN CRITERIA OF AGE, SOLE 819 00:29:30,609 --> 00:29:31,876 CRITERIA FOR CANCER SCREENING 820 00:29:31,876 --> 00:29:33,311 FOR MOST OF THE COMMON CANCERS, 821 00:29:33,311 --> 00:29:34,446 WE MUST DO BETTER THAN THAT 822 00:29:34,446 --> 00:29:36,281 BECAUSE MOST OF THE PEOPLE FOR 823 00:29:36,281 --> 00:29:37,682 EXAMPLE, WOMEN WITH BREAST 824 00:29:37,682 --> 00:29:39,618 CANCER, 88% OF THE WOMEN WILL 825 00:29:39,618 --> 00:29:41,786 NEVER DEVELOP BREAST CANCER IN 826 00:29:41,786 --> 00:29:43,488 THEIR LIFE, BUT ALL ARE SUPPOSED 827 00:29:43,488 --> 00:29:45,323 TO BE HAVING FREQUENT 828 00:29:45,323 --> 00:29:46,391 MAMMOGRAMSA THE A CERTAIN AGE 829 00:29:46,391 --> 00:29:47,592 AND EVEN THAT'S DEBATED. 830 00:29:47,592 --> 00:29:49,561 SO WE CAN DO BETTER THAN THE WAY 831 00:29:49,561 --> 00:29:51,730 WE CANCER SCREEN AND PREDICT THE 832 00:29:51,730 --> 00:29:53,932 RISK OF A PERSON SPECIFIC CANCER 833 00:29:53,932 --> 00:30:00,839 IN THE YEARS AHEAD. 834 00:30:00,839 --> 00:30:03,742 SO LET ME ALSO MENTION PAN 835 00:30:03,742 --> 00:30:06,911 KRISTATTIC CANCER, 1 OF THE MOST 836 00:30:06,911 --> 00:30:08,079 DIFFICULT TO DIAGNOSE, ALMOST 837 00:30:08,079 --> 00:30:11,516 ALSO STAGE 3 OR 4 WHEN IT IS 838 00:30:11,516 --> 00:30:12,083 FINALLY DIAGNOSED. 839 00:30:12,083 --> 00:30:13,818 WE'VE SEEN HOW AI COULD PICK UP 840 00:30:13,818 --> 00:30:15,120 THINGS IN THE ELECTRONIC HEALTH 841 00:30:15,120 --> 00:30:16,354 RECORD THAT WOULD OTHERWISE BE 842 00:30:16,354 --> 00:30:19,424 MISS INDEED A LARGE NATIONWIDE 843 00:30:19,424 --> 00:30:22,260 DENMARK STUDY AS WELL AS THE VA 844 00:30:22,260 --> 00:30:24,329 SYSTEM TO PICK UP PANCREATIC 845 00:30:24,329 --> 00:30:26,231 CANCER YEARS BEFORE IT WAS 846 00:30:26,231 --> 00:30:26,631 MANIFEST. 847 00:30:26,631 --> 00:30:28,433 SO HOPEFULLY WE WILL USE THESE 848 00:30:28,433 --> 00:30:30,969 EARLY DETECTIONS OF CANCER, 849 00:30:30,969 --> 00:30:32,804 PETITIONING OF RISK IN THE TIMES 850 00:30:32,804 --> 00:30:36,074 AHEAD EMPLOY NOW THIS WILL 851 00:30:36,074 --> 00:30:37,509 BECOMING OUT LATER THIS MONTH, 852 00:30:37,509 --> 00:30:39,978 ALSO IN SCIENCE, IT'S ABOUT THIS 853 00:30:39,978 --> 00:30:42,480 REVOLUTION GOING ON IN 854 00:30:42,480 --> 00:30:43,882 PROTEOMICS, HIGH THROUGH PUT 855 00:30:43,882 --> 00:30:45,850 PROTEOMICS AND AI AND I JUST 856 00:30:45,850 --> 00:30:46,918 SUMMARIZED 10 RECENT STUDIES, 857 00:30:46,918 --> 00:30:50,855 MOST OF THEM ARE ERROR RECENT, 858 00:30:50,855 --> 00:30:53,458 AND WE'RE SEEING ORGAN CLOCKS, 859 00:30:53,458 --> 00:30:54,659 BODY WIDE PROTEOMIC CLOCKS THAT 860 00:30:54,659 --> 00:30:56,061 GIVE US HIGH ACCURACY AND 861 00:30:56,061 --> 00:30:58,129 SPECIFICITY ABOUT A PERSON'S 862 00:30:58,129 --> 00:31:00,932 RISK AND DOWN TO THE ORGAN LEVEL 863 00:31:00,932 --> 00:31:03,435 FOR BRAIN, FOR HEART, FOR 864 00:31:03,435 --> 00:31:04,936 KIDNEY, WHATEVER THE ORGAN, 865 00:31:04,936 --> 00:31:06,237 INCLUDING THE IMMUNE SYSTEM, SO 866 00:31:06,237 --> 00:31:07,439 I WON'T GO THROUGH THAT BUT I 867 00:31:07,439 --> 00:31:09,374 JUST WANT TO TAKE YOU THROUGH 868 00:31:09,374 --> 00:31:13,578 THE 3 MOST IMPORTANT AGE RELATED 869 00:31:13,578 --> 00:31:15,346 DISEASES AND HOW WE CAN REVAMP 870 00:31:15,346 --> 00:31:17,415 OUR APPROACH WITH AI, TAKING ALL 871 00:31:17,415 --> 00:31:21,386 THESE DIFFERENT LAYERS OF DATA, 872 00:31:21,386 --> 00:31:23,021 SOME OF THEM I'VE BEEN 873 00:31:23,021 --> 00:31:24,823 MENTIONING PREDICTING HEART 874 00:31:24,823 --> 00:31:26,257 DISEASE, PUTTING PEOPLE UNDER 875 00:31:26,257 --> 00:31:28,927 ACTIVE SURVEILLANCE NOT JUST 876 00:31:28,927 --> 00:31:30,628 THAT THEY WOULD DEVELOP HEART 877 00:31:30,628 --> 00:31:31,996 DISEASE BUT IF THEY'RE 878 00:31:31,996 --> 00:31:34,466 DEVELOPING HEART DISEASE AT AGE 879 00:31:34,466 --> 00:31:37,335 100 NOT NEARLY AS MUCH OF A 880 00:31:37,335 --> 00:31:39,304 CONCERN AS DEVELOPING IT AT AGE 881 00:31:39,304 --> 00:31:43,374 60 EMPLOY AND WE CAN USE THIS TO 882 00:31:43,374 --> 00:31:45,110 MANIFEST THIS AND ALL OF THESE 883 00:31:45,110 --> 00:31:45,977 OTHERS TAKE 2 DECADES TO 884 00:31:45,977 --> 00:31:47,612 DEVELOP, AND THEN GO INTO HIGH 885 00:31:47,612 --> 00:31:49,247 GEAR PREVENTION SO THAL RICK OF 886 00:31:49,247 --> 00:31:53,184 A PERSON'S DEC NEVER IMETS 887 00:31:53,184 --> 00:31:53,451 MANIFEST. 888 00:31:53,451 --> 00:31:56,387 THE SAME IS TRUE FOR CANCER, 889 00:31:56,387 --> 00:31:58,323 SOMEWHAT DIFFERENT LAYERINGS OF 890 00:31:58,323 --> 00:31:59,691 DATA AND SOMEWHAT DIFFERENT 891 00:31:59,691 --> 00:32:02,227 SURVEILLANCE BUT THE SAME THEME 892 00:32:02,227 --> 00:32:03,294 OF PREVENTINGA THAT CANCER NOT 893 00:32:03,294 --> 00:32:05,230 JUST BY EARLY DIAGNOSIS BUT BY 894 00:32:05,230 --> 00:32:08,399 KNOWING THE PERSON'S RISK IN THE 895 00:32:08,399 --> 00:32:10,301 FIRST PLACE AND THEN USING THE 896 00:32:10,301 --> 00:32:11,302 THINGS LIKE MULTICANCKER EARLY 897 00:32:11,302 --> 00:32:13,171 DETECTION TEST WHICH IS NOT THE 898 00:32:13,171 --> 00:32:15,240 WAY THEY'RE BEING USED TODAY. 899 00:32:15,240 --> 00:32:17,742 THEY'RE BEING USED AGE 50 OR 900 00:32:17,742 --> 00:32:20,044 OLDER, THE SAME VERY DUMB-DOWN 901 00:32:20,044 --> 00:32:21,212 APPROACH AND ESPECIALLY NOW WE 902 00:32:21,212 --> 00:32:22,947 KNOW THAT MANY CANCERS ARE 903 00:32:22,947 --> 00:32:25,183 DEVELOPING IN YOUNG PEOPLE'S AGE 904 00:32:25,183 --> 00:32:27,852 AND IN FACT IN 20 ANDS 30INGS 905 00:32:27,852 --> 00:32:31,723 FOR COLON CANCER AND BREAST 906 00:32:31,723 --> 00:32:32,056 CANCER. 907 00:32:32,056 --> 00:32:33,658 AND THEN FINALLY FOR ALSZ 908 00:32:33,658 --> 00:32:35,860 JEIMERS FOR EXAMPLE OF THE 3 909 00:32:35,860 --> 00:32:37,428 ANAL RELATED DECS, THE SAME 910 00:32:37,428 --> 00:32:38,863 THEME HERE USING DIFFERENT 911 00:32:38,863 --> 00:32:41,833 LAYERS OF DATA, DIFFERENT TOOLS, 912 00:32:41,833 --> 00:32:43,434 AI CAN TELL US WHEN THIS IS 913 00:32:43,434 --> 00:32:45,770 LIKELY TO SHOW UP, AND ALSO SET 914 00:32:45,770 --> 00:32:47,739 UP THE INTERVENTION TO PREVENT 915 00:32:47,739 --> 00:32:50,341 THIS FROM EVER HAPPENING. 916 00:32:50,341 --> 00:32:51,409 REALLY EXCITING TIME IN 917 00:32:51,409 --> 00:32:54,579 MEDICINE, WE'VE NEVER HAD THIS 918 00:32:54,579 --> 00:32:55,013 OPPORTUNITY BEFORE. 919 00:32:55,013 --> 00:32:56,080 NOW THE LAST THING I WANT TO 920 00:32:56,080 --> 00:32:57,382 TOUCH ON BEFORE OPENING UP TO 921 00:32:57,382 --> 00:33:01,553 QUESTION SYSTEM ABOUT THE 922 00:33:01,553 --> 00:33:02,320 PATIENT DOCTOR RELATIONSHIP. 923 00:33:02,320 --> 00:33:04,856 YOU KNOW CHRIS MENTIONED THAT I 924 00:33:04,856 --> 00:33:07,158 HAD WRITTEN A BOOK 5 YEARS AGO 925 00:33:07,158 --> 00:33:09,761 ABOUT DEEP MEDICINE AND HOW THIS 926 00:33:09,761 --> 00:33:10,728 IS THE OVERARCHING GOAL. 927 00:33:10,728 --> 00:33:12,897 IF WE CAN IMPROVE THAT 928 00:33:12,897 --> 00:33:15,967 RELATIONSHIP, WHICH IS ERODED 929 00:33:15,967 --> 00:33:19,504 TERRIBLY OVER THE DECADES SINCE 930 00:33:19,504 --> 00:33:20,572 THE 70S ESPECIALLY, THAT WOULD 931 00:33:20,572 --> 00:33:23,441 BE A BIG CRITICAL OBJECTIVE. 932 00:33:23,441 --> 00:33:29,080 SO TODAY, THIS IS ALL TYPES OF 933 00:33:29,080 --> 00:33:30,782 CLINICIANS SEEING A DOCTOR, THEY 934 00:33:30,782 --> 00:33:33,718 DON'T REALLY SEE THE DOCTOR 935 00:33:33,718 --> 00:33:35,987 BECAUSE THEY'RE KEYBOARD, DATA 936 00:33:35,987 --> 00:33:36,921 CLERK KEYBOARD SLAVES AND SO 937 00:33:36,921 --> 00:33:39,090 THIS IS NOT THE RIGHT TYPE OF 938 00:33:39,090 --> 00:33:40,592 MEDICINE THAT WE WANT TO 939 00:33:40,592 --> 00:33:41,125 PRACTICE. 940 00:33:41,125 --> 00:33:43,862 WE WANT TO HAVE PRESENCE, AND 941 00:33:43,862 --> 00:33:48,900 TRUST, AND A BOND THAT'S 942 00:33:48,900 --> 00:33:50,835 AMELIORATED AND BUILT UPON 943 00:33:50,835 --> 00:33:53,905 DURING A VOICE NOLESS BEYOND THE 944 00:33:53,905 --> 00:33:55,506 VISIT AND DANNY [INDISCERNIBLE] 945 00:33:55,506 --> 00:33:57,709 WHO RECENTLY PASSED HE WROTE A 946 00:33:57,709 --> 00:33:59,310 PHENOMENAL BOOK THAT ACTUALLY 947 00:33:59,310 --> 00:34:00,378 ALSO APPLIES TO MEDICINE. 948 00:34:00,378 --> 00:34:02,614 BECAUSE THE WAY MEDICINE IS 949 00:34:02,614 --> 00:34:05,083 TODAY IS FAST AND SHALLOW. 950 00:34:05,083 --> 00:34:10,722 THE AFRAMMING APPOINTMENT IS AT 951 00:34:10,722 --> 00:34:10,989 7 MINUTES. 952 00:34:10,989 --> 00:34:11,990 HOW DO YOU HAVE A RELATIONSHIP 953 00:34:11,990 --> 00:34:13,191 WITH THE PATIENT WHEN THE 954 00:34:13,191 --> 00:34:15,159 PATIENT IS INTERRUPTED AFTER THE 955 00:34:15,159 --> 00:34:16,761 FIRST 82ndS OF TALKING, NEVER 956 00:34:16,761 --> 00:34:19,063 ABLE TO TELL THEIR STORY, NOLESS 957 00:34:19,063 --> 00:34:21,799 NOT HAVING AN ADEQUATE PHYSICAL 958 00:34:21,799 --> 00:34:22,467 EXAM. 959 00:34:22,467 --> 00:34:24,269 SO, WE NEED TIME AND WHAT'S 960 00:34:24,269 --> 00:34:25,937 HAPPENING IS REALLY EXCITING 961 00:34:25,937 --> 00:34:29,340 RIGHT NOW AND THAT'S WITH THIS 962 00:34:29,340 --> 00:34:32,010 ERA OF AI AUTOMATED NOTES FROM 963 00:34:32,010 --> 00:34:34,012 THAT CONVERSATION, A SYNTHETIC 964 00:34:34,012 --> 00:34:35,046 NOTE IS GENERATED WHICH IS 965 00:34:35,046 --> 00:34:38,116 BETTER THAN THE NOTES THAT EXIST 966 00:34:38,116 --> 00:34:40,418 TODAY IN EPIC CENTER AND ALL OF 967 00:34:40,418 --> 00:34:43,588 OUR DIFFERENT HEALTH IT SYSTEMS. 968 00:34:43,588 --> 00:34:45,123 THAT NOTE IS NOT JUST ABOUT THE 969 00:34:45,123 --> 00:34:46,758 NOTE THAT'S GIVEN TO THE PATIENT 970 00:34:46,758 --> 00:34:49,127 WITH AUDIO LINKS AND CAN 971 00:34:49,127 --> 00:34:50,395 TRANSLATED INTO ANY LANGUAGE OR 972 00:34:50,395 --> 00:34:51,896 ANY LEVEL OF EDUCATION BUT ALSO 973 00:34:51,896 --> 00:34:55,967 IS USED TO GET PREAUTHORIZATION, 974 00:34:55,967 --> 00:35:01,506 TO SET UP PRESCRIPTIONS, RETURN 975 00:35:01,506 --> 00:35:03,141 VISITS, FOLLOW-UP LABS, 976 00:35:03,141 --> 00:35:04,642 PROCEDURES, BILLING CODES 977 00:35:04,642 --> 00:35:05,243 EVERYTHING, INCLUDING NUDGING 978 00:35:05,243 --> 00:35:06,844 THE PATIENT ABOUT WHAT WAS 979 00:35:06,844 --> 00:35:09,047 DISCUSSED AT THE VISIT, WHAT 980 00:35:09,047 --> 00:35:10,014 LIKE WHAT WERE THE BLOOD 981 00:35:10,014 --> 00:35:11,749 PRESSURES WE ASKED TO CHECK, ALL 982 00:35:11,749 --> 00:35:12,583 THROUGH AN AUTOMATED NOTE AND 983 00:35:12,583 --> 00:35:15,353 THIS IS NOW IN MANY HEALTH 984 00:35:15,353 --> 00:35:17,422 SYSTEMS AROUND THE COUNTRY, WE 985 00:35:17,422 --> 00:35:19,490 ADOPTED VERY QUICKLY AND THE 986 00:35:19,490 --> 00:35:23,061 AVERAGE PHYSICIAN REPORTS 2-3 987 00:35:23,061 --> 00:35:24,562 HOURS LESS ELECTRONIC HEALTH 988 00:35:24,562 --> 00:35:26,197 RECORD DATA CLERK FUNCTION AND 989 00:35:26,197 --> 00:35:29,300 IT JUST KEEPS GROWING IN TERMS 990 00:35:29,300 --> 00:35:32,904 OF THE BENEFIT AND TIME, THE 991 00:35:32,904 --> 00:35:34,939 GIFT OF TIME FROM AI AND THIS IS 992 00:35:34,939 --> 00:35:36,541 ALSO THE PREAUTHORIZATION WHICH 993 00:35:36,541 --> 00:35:37,709 IS 1 OF THE WORST THINGS TO GO 994 00:35:37,709 --> 00:35:40,378 THROUGH THAT IT'S HELPING IN A 995 00:35:40,378 --> 00:35:41,245 VERY BIG WAY. 996 00:35:41,245 --> 00:35:43,848 SO THIS GIFT OF TIME IS WHAT I 997 00:35:43,848 --> 00:35:47,385 THOUGHT WE WOULD BE ABLE TO TURN 998 00:35:47,385 --> 00:35:48,186 AROUND THIS PATIENT-DOCTOR 999 00:35:48,186 --> 00:35:49,821 RELATIONSHIP BUT I NEVER 1000 00:35:49,821 --> 00:35:50,655 ENVISIONED GIVING PATIENTS MORE 1001 00:35:50,655 --> 00:35:52,890 CHARGE WITH THEIR DR. TOPOLA 1002 00:35:52,890 --> 00:35:53,858 WITH AI ALLEGOR RYTH EPIGENETICS 1003 00:35:53,858 --> 00:35:56,160 TO SUPPORT THAT IT WOULD ALSO 1004 00:35:56,160 --> 00:35:59,497 LEAD TO THIS FACILITATION OF 1005 00:35:59,497 --> 00:35:59,964 EMPATHY. 1006 00:35:59,964 --> 00:36:02,367 MACHINES HAVE NO IDEA WHAT 1007 00:36:02,367 --> 00:36:03,768 EMPATHY IS, BUT INTERESTINGLY 1008 00:36:03,768 --> 00:36:05,003 BECAUSE THEY ARE TRAINED FROM 1009 00:36:05,003 --> 00:36:07,705 OUR CULTURE, FROM OUR LANGUAGE, 1010 00:36:07,705 --> 00:36:10,408 THEY ARE ABLE TO PROMOTE EMPATHY 1011 00:36:10,408 --> 00:36:12,110 AS YOU'VE SEEN IN MANY STUDIES. 1012 00:36:12,110 --> 00:36:15,646 THIS IS 1 OF THE FIRST STUDIES 1013 00:36:15,646 --> 00:36:17,382 FROM UC SAN DIEGO THAT'S 1014 00:36:17,382 --> 00:36:18,750 CONSISTENT WITH EVERY STUDY 1015 00:36:18,750 --> 00:36:20,618 SINCE AND I'VE WRITTEN ABOUT 1016 00:36:20,618 --> 00:36:21,185 THIS, IT'S REALLY EXCITING, 1017 00:36:21,185 --> 00:36:22,787 WOULD NOT HAVE BEEN PREDICTED, I 1018 00:36:22,787 --> 00:36:23,788 NEVER THOUGHT IT WOULD BE 1019 00:36:23,788 --> 00:36:25,990 POSSIBLE BUT IT JUST SHOWS YOU 1020 00:36:25,990 --> 00:36:28,559 HO AI CAN DO THINGS THAT WERE 1021 00:36:28,559 --> 00:36:30,228 NOT NECESSARILY ANTICIPATED. 1022 00:36:30,228 --> 00:36:33,698 SO, HUMAN TOUCH, THAT IS 1023 00:36:33,698 --> 00:36:36,467 BRINGING BACK A MUCH MORE OF A 1024 00:36:36,467 --> 00:36:37,535 PHYSICAL EXAM, THAT'S WHAT 1025 00:36:37,535 --> 00:36:44,108 PATIENTS WANT WHEN THEY GO TO 1026 00:36:44,108 --> 00:36:45,943 SEE A LOSS, AND THEY RECOGNIZE 1027 00:36:45,943 --> 00:36:46,778 IT'S ALL IMPORTANT AND NOW I 1028 00:36:46,778 --> 00:36:48,846 WANT TO TOUCH ON THE DOWN SIDES 1029 00:36:48,846 --> 00:36:50,048 OF AI BECAUSE MOST EVERYTHING I 1030 00:36:50,048 --> 00:36:51,416 TALKED ABOUT GIVING YOU THE 1031 00:36:51,416 --> 00:36:52,850 IMPRESSION AI IS GLOWING AND 1032 00:36:52,850 --> 00:36:54,419 DOING ALL THESE WONDERFUL 1033 00:36:54,419 --> 00:36:56,187 THINGS, WELL, NOT NECESSARILY, 1034 00:36:56,187 --> 00:36:56,454 RIGHT? 1035 00:36:56,454 --> 00:36:59,190 YOU ALL KNOW IT HAS ANOTHER DARK 1036 00:36:59,190 --> 00:37:04,028 SIDE, IT COULD MAKE FOR A LOT 1037 00:37:04,028 --> 00:37:07,432 MORE BLURRING OF WHAT IS TRUTH 1038 00:37:07,432 --> 00:37:08,099 AND FACT. 1039 00:37:08,099 --> 00:37:15,940 IT ALSO HAS THE ABILITY TO 1040 00:37:15,940 --> 00:37:16,774 PROMOTE [INDISCERNIBLE] 1041 00:37:16,774 --> 00:37:17,809 HALLUCINATION AND ERRORS ITSELF, 1042 00:37:17,809 --> 00:37:20,545 IT HAS THE PROBLEM AS THESE 2 1043 00:37:20,545 --> 00:37:22,180 BOOKS THE MIRROR OF OUR CULTURE 1044 00:37:22,180 --> 00:37:24,148 AND SO ALL THE DIFFERENT BIASES 1045 00:37:24,148 --> 00:37:26,451 THAT ARE EMBEDDED IN THE DATA 1046 00:37:26,451 --> 00:37:29,287 INPUTS NOT NECESSARILY THE 1047 00:37:29,287 --> 00:37:32,824 ALGORITHM, AND THIS RECENT BOOK 1048 00:37:32,824 --> 00:37:34,459 WHICH WILL BE INTERVIEWED 1 OF 1049 00:37:34,459 --> 00:37:36,661 ITS AUTHORS AND WILL BE A 1050 00:37:36,661 --> 00:37:38,930 SUBSEQUENT GROUND TRUTHS IN THE 1051 00:37:38,930 --> 00:37:40,798 DAYS AHEAD, AI NCI--AND OIL, A 1052 00:37:40,798 --> 00:37:45,002 VERY TOUGH TITLE OF A BOOK AND I 1053 00:37:45,002 --> 00:37:46,170 GAVE THE AUTHOR SOME TOUGH TIMES 1054 00:37:46,170 --> 00:37:47,705 BECAUSE DURING THE INTERVIEW, HE 1055 00:37:47,705 --> 00:37:49,440 SAID OH AI IS GREAT, WE JUST 1056 00:37:49,440 --> 00:37:51,509 DIDN'T WRITE ABOUT IT IN THE 1057 00:37:51,509 --> 00:37:51,709 BOOK. 1058 00:37:51,709 --> 00:37:55,646 I SAID OH, THAT'S REALLY NICE. 1059 00:37:55,646 --> 00:37:57,815 [LAUGHTER] 1060 00:37:57,815 --> 00:38:00,551 , 1061 00:38:00,551 --> 00:38:03,554 SO THE BIASES AND POTENTIAL OF 1062 00:38:03,554 --> 00:38:05,623 WORSENING INEBS QUITYS, WE DON'T 1063 00:38:05,623 --> 00:38:06,324 WANT THAT. 1064 00:38:06,324 --> 00:38:08,059 WE'VE ALREADY SEEN EXAMPLES OF 1065 00:38:08,059 --> 00:38:09,794 HOW IT CAN IMPROVE INEQUITIES 1066 00:38:09,794 --> 00:38:11,429 AND OF COURSE THE FEAR OF AI 1067 00:38:11,429 --> 00:38:12,663 THAT IS PARTICULARLY CLINICIANS 1068 00:38:12,663 --> 00:38:14,365 THINK IT WILL REPLACE THEM 1069 00:38:14,365 --> 00:38:15,867 SOMEHOW WHICH IT'S NOT GOING TO 1070 00:38:15,867 --> 00:38:18,302 REPLACE THEM BUT THESE ARE 1071 00:38:18,302 --> 00:38:22,473 EXAMPLES OF IMPROVING EQUITY 1072 00:38:22,473 --> 00:38:26,477 WHETHER IT'S RHEUMATIC HEART 1073 00:38:26,477 --> 00:38:29,080 DISEASE IN AFRICA OR GETTING 1074 00:38:29,080 --> 00:38:32,517 PEOPLE WITH MENTAL HEALTH IN THE 1075 00:38:32,517 --> 00:38:33,684 UK TOA ACCESSIBILITY OF CHAT 1076 00:38:33,684 --> 00:38:35,119 BOTS WHERE THEY HAD THEM BEFORE, 1077 00:38:35,119 --> 00:38:36,521 THIS IS WHERE AI IS HELPING TO 1078 00:38:36,521 --> 00:38:40,324 BRIDGE AND REDUCE THE PROBLEMS 1079 00:38:40,324 --> 00:38:40,791 OF HEALTH INIQUITIES. 1080 00:38:40,791 --> 00:38:42,627 HOPEFUL THERE THERE WILL BE MANY 1081 00:38:42,627 --> 00:38:43,895 MORE EXAMPLES OF THAT IN THE 1082 00:38:43,895 --> 00:38:44,395 FUTURE. 1083 00:38:44,395 --> 00:38:46,030 JUST TO SUMMARIZE, WE HAVE LOTS 1084 00:38:46,030 --> 00:38:47,431 OF UPHILL BALGTS WITH,A I. 1085 00:38:47,431 --> 00:38:50,201 I MENTIONED IT'S NOT BEEN 1086 00:38:50,201 --> 00:38:51,169 IMPLEMENTED VERY MUCH, I 1087 00:38:51,169 --> 00:38:52,436 MENTIONED THERE'S A PROBLEM WITH 1088 00:38:52,436 --> 00:38:56,340 TRUST BECAUSE THE DATA FROM 1089 00:38:56,340 --> 00:38:58,309 THESE FDA AUTHORIZED ALGORITHMS 1090 00:38:58,309 --> 00:39:01,412 AND FREQUENTLY, MOST FREQUENTLY 1091 00:39:01,412 --> 00:39:04,282 NOT PUBLISHED OR IT IS PUBLISHED 1092 00:39:04,282 --> 00:39:05,783 IT'S RETROSPECTIVE MODELS AND A 1093 00:39:05,783 --> 00:39:11,556 LOT OF IT IS LEFT IN PROPRIETARY 1094 00:39:11,556 --> 00:39:13,858 SPHERE, SO LACK OF TRANSPARENCY, 1095 00:39:13,858 --> 00:39:16,394 LACK OF COMPELLING TRIALS, 1096 00:39:16,394 --> 00:39:17,728 REGULATORY LOW THRESHOLD FOR 1097 00:39:17,728 --> 00:39:19,363 GETTING THESE AUTHORIZED, LACK 1098 00:39:19,363 --> 00:39:20,965 OF REIMBURSEMENT AND ALSO THE 1099 00:39:20,965 --> 00:39:22,133 GENERAL RESISTANCE OF CHANGE IN 1100 00:39:22,133 --> 00:39:23,034 THE MEDICAL COMMUNITY. 1101 00:39:23,034 --> 00:39:26,103 SO I JUST WANT TO LEAVE YOU WITH 1102 00:39:26,103 --> 00:39:27,939 MY FAVORITE QUOTE IN MEDICINE 1103 00:39:27,939 --> 00:39:31,709 AND THIS IS FROM FRANCIS 1104 00:39:31,709 --> 00:39:33,344 PEABODY, ALMOST A HUNDRED YEARS 1105 00:39:33,344 --> 00:39:36,581 AGO, A SINGLE AUTHOR, PAPER, 1106 00:39:36,581 --> 00:39:38,449 1927 JAMA, IF YOU HAVEN'T READ 1107 00:39:38,449 --> 00:39:40,084 IT, YOU WILL GET INSPIRED ABOUT 1108 00:39:40,084 --> 00:39:42,486 MEDICINE AND WHAT HE SAYS AT THE 1109 00:39:42,486 --> 00:39:43,754 END IS 1 OF THE ESSENTIAL 1110 00:39:43,754 --> 00:39:46,357 QUALITIES OF THE CLINICIAN IS 1111 00:39:46,357 --> 00:39:47,992 INTEREST IN HUMANITY FOR THE 1112 00:39:47,992 --> 00:39:50,561 SECRET OF THE CARE OF THE 1113 00:39:50,561 --> 00:39:53,598 PATIENT IS IN CARING FOR THE 1114 00:39:53,598 --> 00:39:54,532 PATIENT AND HOPEFULLY AI WILL 1115 00:39:54,532 --> 00:39:55,366 HELP US GET THERE. 1116 00:39:55,366 --> 00:39:58,803 SO THANK YOU VERY MUCH. 1117 00:39:58,803 --> 00:40:00,338 THIS IS AN ACKNOWLEDGMENT OF THE 1118 00:40:00,338 --> 00:40:01,939 PEOPLE I GET TO WORK WITH AT 1119 00:40:01,939 --> 00:40:04,108 SCRIPTS RESEARCH AND OUR FUNDING 1120 00:40:04,108 --> 00:40:05,309 SOURCES AND I LOOK FORWARD TO 1121 00:40:05,309 --> 00:40:07,678 YOUR QUESTIONS AND CRITIQUE, 1122 00:40:07,678 --> 00:40:08,713 THANK YOU. 1123 00:40:08,713 --> 00:40:14,552 [ APPLAUSE ] 1124 00:40:14,552 --> 00:40:17,788 >> THANK YOU DR. TOPOL FOR SUCH 1125 00:40:17,788 --> 00:40:18,489 A WONDERFUL TALK. 1126 00:40:18,489 --> 00:40:20,224 SO THE PODIUMS ARE OPEN IF YOU 1127 00:40:20,224 --> 00:40:21,993 HAVE A QUESTION PLEASE STEP UP 1128 00:40:21,993 --> 00:40:23,928 TO THE MICROPHONE SO THAT WE CAN 1129 00:40:23,928 --> 00:40:28,432 IDENTIFY YOU AND YOU CAN ASK 1130 00:40:28,432 --> 00:40:31,269 YOUR QUESTION, AND SO DR. TOPOL, 1131 00:40:31,269 --> 00:40:35,039 JUST TO LET YOU KNOW WE HAD 1132 00:40:35,039 --> 00:40:37,441 ABOUT 300 PEOPLE ONLINE WATCHING 1133 00:40:37,441 --> 00:40:40,311 SO CLEARLY AN EXCELLENT TOPIC. 1134 00:40:40,311 --> 00:40:42,146 >> HI, I HAVE A QUESTION ABOUT 1135 00:40:42,146 --> 00:40:43,481 THE AMBIENT LISTENING NOTES, I 1136 00:40:43,481 --> 00:40:45,316 DON'T KNOW A SINGLE DOCTOR THAT 1137 00:40:45,316 --> 00:40:47,418 DOESN'T WANT THAT. 1138 00:40:47,418 --> 00:40:47,618 OKAY? 1139 00:40:47,618 --> 00:40:49,020 AND YOU GAVE A LOT OF GREAT 1140 00:40:49,020 --> 00:40:53,324 EXAMPLES OF WHAT IT CAN DO, WHAT 1141 00:40:53,324 --> 00:40:56,527 ARE THE PRIVACY CONCERNS, 1142 00:40:56,527 --> 00:40:57,295 REGULATORY PIECES, GOVERNMENT 1143 00:40:57,295 --> 00:40:58,396 AGENCY, CAN YOU SPEAK A LITTLE 1144 00:40:58,396 --> 00:41:01,265 BIT TO WHY THAT MIGHT NOT BE 1145 00:41:01,265 --> 00:41:01,499 WANTED? 1146 00:41:01,499 --> 00:41:02,466 >> I'M SO DPLAD YOU ASKED ABOUT 1147 00:41:02,466 --> 00:41:06,070 IT EMPLOY SO THERE ARE HEALTH 1148 00:41:06,070 --> 00:41:07,738 SYSTEMS, 1 THAT'S NOTABLE IS 1149 00:41:07,738 --> 00:41:09,140 EMORY IN ATLANTA WHERE ALL THE 1150 00:41:09,140 --> 00:41:11,342 DOCTORS HAVE ACCESS TO 1 OF 1151 00:41:11,342 --> 00:41:14,612 THESE AND WHAT'S INTERESTING IS 1152 00:41:14,612 --> 00:41:17,148 THAT THE HEALTH IT PHYSICIAN 1153 00:41:17,148 --> 00:41:18,516 OVERSEEING, HE SELLS, HE SAYS HE 1154 00:41:18,516 --> 00:41:20,284 GETS LOVE NOTES FROM ALL HIS 1155 00:41:20,284 --> 00:41:21,786 DOCTORS, I NEVER HEARD OF THAT, 1156 00:41:21,786 --> 00:41:32,296 LOVE NOTE TO THE IT PHYSICIAN. 1157 00:41:35,533 --> 00:41:36,300 THERE HASN'T BEEN A CONCERN 1158 00:41:36,300 --> 00:41:39,403 WHERE IT DOESN'T HAVE TO GO 1159 00:41:39,403 --> 00:41:43,974 THROUGH FDA, THE POINT ABOUT 1160 00:41:43,974 --> 00:41:45,543 PRIVACY AND WITH THE HEALTH AND 1161 00:41:45,543 --> 00:41:47,044 DATA BUT WHAT'S NICE IS EVERY 1162 00:41:47,044 --> 00:41:49,213 PATIENT GETS THEIR AUDIO AND 1163 00:41:49,213 --> 00:41:51,349 THEIR WRITTEN TRANSCRIPT WHEN 1164 00:41:51,349 --> 00:41:54,685 THEY LEAVE THE OFFICE VISIT. 1165 00:41:54,685 --> 00:41:57,088 I THINK IT WILL BECOME THE NORM 1166 00:41:57,088 --> 00:41:58,889 VERY QUICKLY FOR EPIC CONCERN 1167 00:41:58,889 --> 00:42:00,558 AND THE OTHER HEALTH SYSTEMS ARE 1168 00:42:00,558 --> 00:42:01,826 BASICALLY FIGURING OUT WHICH 1S 1169 00:42:01,826 --> 00:42:06,464 TO WORK WITH AND HOW TO 1170 00:42:06,464 --> 00:42:08,499 INTEGRATE IT AND MAKE IT 1171 00:42:08,499 --> 00:42:09,633 EMBEDDED IN THEIR CAPABILITY, 1172 00:42:09,633 --> 00:42:11,068 IT'S EXTRAORDINARY AND IT'S 1173 00:42:11,068 --> 00:42:12,370 WOORKING MUCH BETTER THAN I 1174 00:42:12,370 --> 00:42:13,137 EXPECTED. 1175 00:42:13,137 --> 00:42:13,904 NOW THERE'S ABOUT 50 DIFFERENT 1176 00:42:13,904 --> 00:42:15,306 COMPANIES IN THIS SPACE, THE 1177 00:42:15,306 --> 00:42:16,941 LARGEST 1 IS MICROSOFT WITH 1178 00:42:16,941 --> 00:42:18,743 NUANCE AND THAT 1 IS NOT 1179 00:42:18,743 --> 00:42:19,710 ACTUALLY NEARLY AS GOOD AS SOME 1180 00:42:19,710 --> 00:42:24,115 OF THE OTHERS THAT ARE 1181 00:42:24,115 --> 00:42:24,648 START-UPS. 1182 00:42:24,648 --> 00:42:25,416 WHAT'S INTERESTING IS SOME OF 1183 00:42:25,416 --> 00:42:26,817 THEM JUST DO THE NOTE, THEY 1184 00:42:26,817 --> 00:42:28,452 DON'T DO ANYTHING ELSE, AND SOME 1185 00:42:28,452 --> 00:42:32,623 OF THEM DO EVERYTHING. 1186 00:42:32,623 --> 00:42:34,759 AND OF COURSE IT DOES REQUIRE 1187 00:42:34,759 --> 00:42:35,893 THE NURSE OR PHYSICIAN OVERSIGHT 1188 00:42:35,893 --> 00:42:38,362 TO MAKE SURE THAT THE 1189 00:42:38,362 --> 00:42:39,830 PRESCRIPTIONS AND--YOU KNOW BUT 1190 00:42:39,830 --> 00:42:41,432 IT'S MUCH QUICKER BECAUSE IT 1191 00:42:41,432 --> 00:42:43,401 DOES ALL THE CLICKS AND ALL THE 1192 00:42:43,401 --> 00:42:45,436 EXTRA WORK AND THE NOTABLE TIME 1193 00:42:45,436 --> 00:42:48,305 IT SAVES IN THE EARLY GOING OF 1194 00:42:48,305 --> 00:42:50,674 SAVE 2-3 HOURS ON AVERAGE AND 1195 00:42:50,674 --> 00:42:52,042 THAT KEEPS GROWING SO RIGHT NOW 1196 00:42:52,042 --> 00:42:56,447 IT'S THE NEAR TERM EXCITEMENT OF 1197 00:42:56,447 --> 00:42:56,914 AI. 1198 00:42:56,914 --> 00:43:00,184 AND WITH THE OVERSIGHT THE ERROR 1199 00:43:00,184 --> 00:43:01,819 CONCERN IS NOT THERE AND IT 1200 00:43:01,819 --> 00:43:05,556 QUICKLY LEARNED THE PHYSICIANS 1201 00:43:05,556 --> 00:43:06,724 WAY OF COMMUNICATING AND HERE'S 1202 00:43:06,724 --> 00:43:09,126 1 OTHER THING I DIDN'T MENTION 1203 00:43:09,126 --> 00:43:12,496 AND THIS IS STRIKING. 1204 00:43:12,496 --> 00:43:14,331 IN THE NEXT COUPLE YEARS EVERY 1205 00:43:14,331 --> 00:43:15,933 PHYSICIAN WILL HAVE TO HAVE 1206 00:43:15,933 --> 00:43:21,739 COACHING BY THE AI. 1207 00:43:21,739 --> 00:43:24,241 NOW THINK ABOUT THIS, WHEN THE 1208 00:43:24,241 --> 00:43:26,544 AI NOTE, AND NOTE IT SAYS, 1209 00:43:26,544 --> 00:43:27,545 DR. JONES, WHY DID YOU INTERRUPT 1210 00:43:27,545 --> 00:43:31,115 THE PATIENT SO QUICKLY? 1211 00:43:31,115 --> 00:43:32,516 WHY DIDN'T YOU LET HER TELL YOU 1212 00:43:32,516 --> 00:43:33,451 ABOUT THIS OR THAT? 1213 00:43:33,451 --> 00:43:36,320 AND SO WHEN YOU SEE THE CRITIQUE 1214 00:43:36,320 --> 00:43:37,721 TAKEN--THEY IT PROVIDES TO THE 1215 00:43:37,721 --> 00:43:39,023 PHYSICIAN, YOU SAY OH MY GOSH 1216 00:43:39,023 --> 00:43:40,858 THIS WILL MAKE DOCTORS BETTER. 1217 00:43:40,858 --> 00:43:44,795 IT DOESN'T HAVE THE SAME AFFECT 1218 00:43:44,795 --> 00:43:47,765 AS MUCH ON NURSES BUT DOCTORS 1219 00:43:47,765 --> 00:43:49,300 REALLY NEED EMPATHY CHECK UPS 1220 00:43:49,300 --> 00:43:50,334 AND COMMUNICATION AND 1221 00:43:50,334 --> 00:43:51,335 SENSITIVITY CHECK UPS SO I SEE 1222 00:43:51,335 --> 00:43:54,038 THAT AS 1 OF THE NEXT FEATURES 1223 00:43:54,038 --> 00:43:55,506 THAT WILL BECOME NEVER AGAIN, 1224 00:43:55,506 --> 00:43:58,042 WHO WOULD HAVE GUESSED AND SO I 1225 00:43:58,042 --> 00:43:59,810 WOULDN'T BE SURPRISED IF IT'S A 1226 00:43:59,810 --> 00:44:02,012 REQUIREMENT OVER THE NEXT COUPLE 1227 00:44:02,012 --> 00:44:03,681 OF YEARS. 1228 00:44:03,681 --> 00:44:05,783 NETWORK HI, JEFF, NCI LABORATORY 1229 00:44:05,783 --> 00:44:11,121 PATHOLOGY, I KNOW YOU TOUCHED A 1230 00:44:11,121 --> 00:44:13,624 LITTLE BIT ABOUT FDA AI AND 1231 00:44:13,624 --> 00:44:15,693 MEDICAL TALK AND I'VE DONE A LOT 1232 00:44:15,693 --> 00:44:18,028 OF RESEARCH AROUND A DEVICE 1233 00:44:18,028 --> 00:44:19,797 THAT'S A LIMITED OUTPUT OUTCOME, 1234 00:44:19,797 --> 00:44:23,400 IT'S ASKING A YES OR NO QUESTION 1235 00:44:23,400 --> 00:44:25,703 IN THE AI MULTIMODAL INPUT AND 1236 00:44:25,703 --> 00:44:27,671 OUTPUT, IT SEEMS LIKE THEY'RE 1237 00:44:27,671 --> 00:44:29,540 REALLY LOST ON HOW TO REGULATE 1238 00:44:29,540 --> 00:44:32,109 THOSE AND I WAS WONDERING IF YOU 1239 00:44:32,109 --> 00:44:35,813 COULD EXPAND ON HOW THEY SHOULD 1240 00:44:35,813 --> 00:44:38,449 APPROACH ACCURACY WHEN THE INPUT 1241 00:44:38,449 --> 00:44:40,651 AND OUTPUT ARE VARIED. 1242 00:44:40,651 --> 00:44:43,254 >> WELL THE SIMPLE ANSWER IS THE 1243 00:44:43,254 --> 00:44:46,590 FDA HAS NOT, PROVED ANY MOWLTY 1244 00:44:46,590 --> 00:44:47,825 MODAL MODELS BECAUSE THEY 1245 00:44:47,825 --> 00:44:49,093 HAVEN'T FIGURED OUT HOW THEY'RE 1246 00:44:49,093 --> 00:44:52,096 GOING TO DO IT, AS YOU SAY. 1247 00:44:52,096 --> 00:44:54,298 BUT THE CRITERIA FOR THE 1248 00:44:54,298 --> 00:44:55,566 ORIGINAL MOSTLY RADIOLOGY BUT A 1249 00:44:55,566 --> 00:44:57,334 LOT OF THE OTHERS AS I 1250 00:44:57,334 --> 00:45:00,838 MENTIONED, THE CRITERIA FOR 1251 00:45:00,838 --> 00:45:03,140 APPROVAL WAS SO WEAK. 1252 00:45:03,140 --> 00:45:06,610 I'VE ACTUALLY GIVEN MY FRIEND 1253 00:45:06,610 --> 00:45:07,745 ROB [INDISCERNIBLE] A HARD TIME 1254 00:45:07,745 --> 00:45:09,914 ABOUT THIS BECAUSE THEY'RE NOT 1255 00:45:09,914 --> 00:45:11,215 SHOWING TEETH ABOUT DEMANDING 1256 00:45:11,215 --> 00:45:14,652 THAT THE DATA IS MORE ROBUST AND 1257 00:45:14,652 --> 00:45:15,819 PUBLISHED AND TRANSPARENT AND 1258 00:45:15,819 --> 00:45:18,856 HOW WILL WE GET THAL TO BE 1259 00:45:18,856 --> 00:45:20,057 ACTUALIZED IN DAILY PRACTICE, SO 1260 00:45:20,057 --> 00:45:21,325 RIGHT NOW THEY'RE CAUGHT IN A 1261 00:45:21,325 --> 00:45:22,893 VERY TOUGH SITUATION BECAUSE 1262 00:45:22,893 --> 00:45:25,462 THEY WERE TOO LOW A THRESHOLD 1263 00:45:25,462 --> 00:45:27,565 AND NOW AI HAS GOTTEN MUCH MORE 1264 00:45:27,565 --> 00:45:28,766 COMPLICATED IN THESE NEW MODELS 1265 00:45:28,766 --> 00:45:30,000 SO WE'RE SITTING IN SUSPENSION 1266 00:45:30,000 --> 00:45:32,136 RIGHT NOW EMPLOY THEY HAVEN'T 1267 00:45:32,136 --> 00:45:33,037 OFFERED GUIDANCE AND IT'S GOING 1268 00:45:33,037 --> 00:45:35,005 TO BE INTERESTING TO SEE WHAT IS 1269 00:45:35,005 --> 00:45:36,407 THE NEXT STEP BUT I DON'T KNOW 1270 00:45:36,407 --> 00:45:38,509 HOW THIS IS GOING TO PLAY OUT 1271 00:45:38,509 --> 00:45:40,210 EMPLOY I'M WORRIED ABOUT IT 1272 00:45:40,210 --> 00:45:43,213 BECAUSE THE FACT THAT WE HAVE 1273 00:45:43,213 --> 00:45:46,216 NERL A THOUSAND AI--THEY CALL 1274 00:45:46,216 --> 00:45:47,885 DEVICES, THEY'RE REALLY SOFTWARE 1275 00:45:47,885 --> 00:45:48,852 THAT ARE AUTHORIZED AND MOST OF 1276 00:45:48,852 --> 00:45:50,821 THEM ARE NOT IN RUES, THAT 1277 00:45:50,821 --> 00:45:52,323 DOESN'T SAY--AND MOST OF THEM 1278 00:45:52,323 --> 00:45:53,791 ARE OF COURSE NEVER BEEN 1279 00:45:53,791 --> 00:45:58,095 PUBLISHED IN A MEDICAL PEER 1280 00:45:58,095 --> 00:45:59,363 REVIEW IT'S REALLY WORRISOME 1281 00:45:59,363 --> 00:46:00,698 THAT WE CAN'T KEEP DOING THAT 1282 00:46:00,698 --> 00:46:01,165 SORT OF THING. 1283 00:46:01,165 --> 00:46:03,233 WE NEED TO HAVE MUCH BETTER DATA 1284 00:46:03,233 --> 00:46:05,302 SETS AND TRANSPARENCY AND WE'RE 1285 00:46:05,302 --> 00:46:06,804 GOING TO--IT'S HITTING THE ACID 1286 00:46:06,804 --> 00:46:08,405 TEST RIGHT NOW BECAUSE WITHIN 1287 00:46:08,405 --> 00:46:10,774 THE NEXT FEW MONTHS SOME OF 1288 00:46:10,774 --> 00:46:12,042 THESE MULTIMODAL MODELS WILL BE 1289 00:46:12,042 --> 00:46:16,413 TRYING TO GET FDA APPROVAL. 1290 00:46:16,413 --> 00:46:18,816 >> HI, THANK YOU VERY MUCH. 1291 00:46:18,816 --> 00:46:20,684 I'M JASMINE [INDISCERNIBLE] FROM 1292 00:46:20,684 --> 00:46:22,286 BIOETHICS AND MY BACKGROUND IS 1293 00:46:22,286 --> 00:46:25,289 IN PHILOSOPHY SO I HOPE YOU WILL 1294 00:46:25,289 --> 00:46:27,257 EXCUSE PERHAPS THE SIMPLISTIC 1295 00:46:27,257 --> 00:46:29,593 DESCRIPTION I'M ABOUT TO GIVE OF 1296 00:46:29,593 --> 00:46:30,461 THE DIAGNOSTIC PROCESS BUT I 1297 00:46:30,461 --> 00:46:31,762 WANTED TO ASK YOU ABOUT THE 1298 00:46:31,762 --> 00:46:33,464 TIMING OF WHEN WE SHOULD USE 1299 00:46:33,464 --> 00:46:37,267 THESE AI TOOLS IN DIAGNOSTICS. 1300 00:46:37,267 --> 00:46:39,770 SO IT SEEMS TO ME THAT THERE'S 1301 00:46:39,770 --> 00:46:41,905 SOME REASON TO HAVE A PHYSICIAN 1302 00:46:41,905 --> 00:46:43,540 FIRST TAKE A LOOK AT YOU KNOW 1303 00:46:43,540 --> 00:46:46,043 THE SCAN THEMSELVES AND SEE WHAT 1304 00:46:46,043 --> 00:46:47,678 THEY THINK, AND THEN TURN TO THE 1305 00:46:47,678 --> 00:46:50,748 AI AND SEE IF IT'S FLAGGING 1306 00:46:50,748 --> 00:46:51,949 SOMETHING THEY MISSED ARK 1307 00:46:51,949 --> 00:46:53,651 CORRELATED POSE DOING IT THE 1308 00:46:53,651 --> 00:46:55,519 OTHER WAY AROUND WHICH WILL BE 1309 00:46:55,519 --> 00:46:57,254 TIME SAVING BUT THEN YOU RISK ON 1310 00:46:57,254 --> 00:46:59,056 LEANING ON IT TOO HEAVILY, AND I 1311 00:46:59,056 --> 00:47:00,457 WONDERED WHAT YOU THINK ABOUT 1312 00:47:00,457 --> 00:47:01,759 THESE RISKS, THAT BALANCE ON 1313 00:47:01,759 --> 00:47:05,229 RELYING ON IT TOO MUCH VERSUS 1314 00:47:05,229 --> 00:47:07,331 TIME SAVING COMPONENT? 1315 00:47:07,331 --> 00:47:09,700 >> I LIKE YOUR POINT. 1316 00:47:09,700 --> 00:47:11,068 THE HUMAN FIRST, U MAN 1317 00:47:11,068 --> 00:47:12,269 INTELLIGENCE FIRST AND THEN AI 1318 00:47:12,269 --> 00:47:15,673 TO BE KIND OF CHECKING MAYBE 1319 00:47:15,673 --> 00:47:18,142 SOME OTHER DIAGNOSIS TO CONSIDER 1320 00:47:18,142 --> 00:47:21,712 OR ONLY TURNING TURNING TO IT N 1321 00:47:21,712 --> 00:47:23,280 THERE'S LACK OF FULL CONFIDENCE 1322 00:47:23,280 --> 00:47:25,015 FOR THE RIGHT DIAGNOSIS. 1323 00:47:25,015 --> 00:47:27,885 AS FAR AS I KNOW THAT SEQUENCE 1324 00:47:27,885 --> 00:47:29,586 IS VERY IMPORTANT BUT IT HASN'T 1325 00:47:29,586 --> 00:47:30,354 REALLY BEEN STUDIED. 1326 00:47:30,354 --> 00:47:31,522 I THINK IT'S A REALLY 1327 00:47:31,522 --> 00:47:32,256 INTERESTING QUESTION YOU BRING 1328 00:47:32,256 --> 00:47:32,423 UP. 1329 00:47:32,423 --> 00:47:35,559 IT WOULD BE BETTER, I WOULD JUST 1330 00:47:35,559 --> 00:47:38,262 THINK TO HAVE THE AI DONE MORE 1331 00:47:38,262 --> 00:47:39,897 IN THE INTEREST OF CONSULT AND 1332 00:47:39,897 --> 00:47:40,964 THOUGHT PROCESS AS STARTED. 1333 00:47:40,964 --> 00:47:43,200 THE PROBLEM IS WE GOTTA GET 1334 00:47:43,200 --> 00:47:44,468 CLINICIANS MORE TIME SO THEY CAN 1335 00:47:44,468 --> 00:47:50,074 ACTUALLY GO THROUGH ALL THIS AND 1336 00:47:50,074 --> 00:47:51,041 OFTEN THEY'RE THINKING ABOUT 1337 00:47:51,041 --> 00:47:52,443 THESE DAYS, HOW CAN I SAVE TIME, 1338 00:47:52,443 --> 00:47:54,878 I WILL GO TO AI FIRST AND NOT 1339 00:47:54,878 --> 00:47:56,046 EACH HAVE TO THINK, THAT'S 1340 00:47:56,046 --> 00:47:57,581 PROBABLY NOT A GOOD IDEA BUT 1341 00:47:57,581 --> 00:47:59,316 GREAT POINT YOU'RE MAKING. 1342 00:47:59,316 --> 00:47:59,750 >> THANK YOU. 1343 00:47:59,750 --> 00:48:00,918 >> THANK YOU. 1344 00:48:00,918 --> 00:48:02,553 BEFORE WE JUMP TO THE NEXT 1345 00:48:02,553 --> 00:48:03,387 QUESTION, THERE ARE MANY 1346 00:48:03,387 --> 00:48:05,989 QUESTIONS THAT HAVE COME IN 1347 00:48:05,989 --> 00:48:07,157 THROUGH THE ONLINE CHAT SO I'M 1348 00:48:07,157 --> 00:48:09,259 JUST GOING TO PULL 1 AND ASK 1349 00:48:09,259 --> 00:48:11,028 YOU. 1350 00:48:11,028 --> 00:48:11,829 >> SURE. 1351 00:48:11,829 --> 00:48:13,764 >> SO, SOMEBODY ONLINE IS 1352 00:48:13,764 --> 00:48:15,232 DESCRIBING THAT SOMETIMES 1353 00:48:15,232 --> 00:48:16,400 PHYSICIANS BECAUSE OF TIME 1354 00:48:16,400 --> 00:48:18,235 CRUNCHS OR OTHER THINGS HAVE A 1355 00:48:18,235 --> 00:48:20,104 RUSH TO JUDGMENT, THEY MIGHT 1356 00:48:20,104 --> 00:48:21,638 DIAGNOSE HYPERTENSION AFTER A 1357 00:48:21,638 --> 00:48:24,274 SINGLE CERTAINLY--CERTAINLY 1358 00:48:24,274 --> 00:48:27,211 VALENTINEDUATEED ELEVATED BLOOD 1359 00:48:27,211 --> 00:48:29,379 PRESSURE, AND THE QUESTION BEING 1360 00:48:29,379 --> 00:48:32,249 ASKED IS HOW WELL DO AI MODELS 1361 00:48:32,249 --> 00:48:32,950 ACCOUNT FOR GASH CANCER CENTER 1362 00:48:32,950 --> 00:48:35,052 IN AS A WIMBERLY TO ALSO SUPPORT 1363 00:48:35,052 --> 00:48:37,087 DOUBLE CHECKING BASED ON CERTAIN 1364 00:48:37,087 --> 00:48:38,522 DIAGNOSIS SO THAT IT CAN BE MORE 1365 00:48:38,522 --> 00:48:40,624 OBLIGATIONS YECTIVE IN THE 1366 00:48:40,624 --> 00:48:44,828 FUTURE SO THE GARBAGE IN-GARBAGE 1367 00:48:44,828 --> 00:48:45,095 OUT MODEL. 1368 00:48:45,095 --> 00:48:46,463 >> YEAH THIS, IS WHAT I WAS 1369 00:48:46,463 --> 00:48:47,865 TRYING TO GET AT WITH THE 1370 00:48:47,865 --> 00:48:48,799 FORECASTING PART OF THE TALK 1371 00:48:48,799 --> 00:48:53,036 WHICH IS IF WE HAVE IN RICH 1372 00:48:53,036 --> 00:48:54,138 DATA, MULTIPLE LAYERS WE WILL BE 1373 00:48:54,138 --> 00:48:55,839 SO MUCH BETTER, SO MUCH MORE 1374 00:48:55,839 --> 00:48:57,341 ACCURATE BUT WE'RE ONLY DEALING 1375 00:48:57,341 --> 00:48:58,742 WITH 1 THING LIKE A PERSON'S 1376 00:48:58,742 --> 00:49:00,077 BLOOD PRESSURE WHEN THEY SHOWED 1377 00:49:00,077 --> 00:49:03,280 UP TO A CLINIC WHICH IS 1378 00:49:03,280 --> 00:49:04,214 NOTORIOUSLY HARD TO ASSESS, THAT 1379 00:49:04,214 --> 00:49:06,250 WILL LEAD TO PROBLEMS AND SOME 1380 00:49:06,250 --> 00:49:07,217 OF OUR DIAGNOSTIC INACCURACY IS 1381 00:49:07,217 --> 00:49:09,853 THE RUSH, THE LACK OF TIME, RUSH 1382 00:49:09,853 --> 00:49:17,561 TO JUDGMENT, RUSH TO ERONIOUS 1383 00:49:17,561 --> 00:49:17,828 DIAGNOSIS. 1384 00:49:17,828 --> 00:49:19,630 IT'S BEEN SHOWN THAT IF A 1385 00:49:19,630 --> 00:49:21,465 CLINICIAN THINKS OF THE FIRST 1386 00:49:21,465 --> 00:49:23,867 DIAGNOSIS WITHIN THE FIRST 5 1387 00:49:23,867 --> 00:49:27,037 MINUTES THE ACCURACY IS HIGH 1388 00:49:27,037 --> 00:49:29,273 ABOUT 95%, BUT AFTER A WHILE, IT 1389 00:49:29,273 --> 00:49:31,575 DROPS OFF, DOWN INTO THE 20S. 1390 00:49:31,575 --> 00:49:32,910 WE NEED TO DO BETTER ON THIS FOR 1391 00:49:32,910 --> 00:49:34,278 SURE AND THE BIGGEST THING WE 1392 00:49:34,278 --> 00:49:37,214 CAN DO IS HAVE THESE LAYERS OF 1393 00:49:37,214 --> 00:49:38,715 DATA AVAILABLE, SO THE AI, WHICH 1394 00:49:38,715 --> 00:49:41,351 IS WHAT THE U MAN CAPABILITIES 1395 00:49:41,351 --> 00:49:42,886 FOR GOING THROUGH LARGE DATA 1396 00:49:42,886 --> 00:49:44,188 SETS, WE'RE TALKING ABOUT BIG 1397 00:49:44,188 --> 00:49:45,389 DATA THEA THE INDIVIDUAL LEVEL, 1398 00:49:45,389 --> 00:49:49,293 WE'RE NOT GOOD AT THAT. 1399 00:49:49,293 --> 00:49:51,361 SCANNING ALL THE LABS EVEN 1400 00:49:51,361 --> 00:49:54,531 WITHIN THE NORMAL RANGE TO LACK 1401 00:49:54,531 --> 00:49:56,400 AT TRENDS, PULLING IN DATA FROM 1402 00:49:56,400 --> 00:49:58,101 DIFFERENT PROVIDERS AND HEALTH 1403 00:49:58,101 --> 00:49:59,069 SYSTEMS AND EVERYTHING ELSE AND 1404 00:49:59,069 --> 00:50:04,107 ALSO THE UNSTRUCTURED TEXT. 1405 00:50:04,107 --> 00:50:05,409 WE DON'T TEND TO READ THAT IN 1406 00:50:05,409 --> 00:50:06,877 THE NOS. 1407 00:50:06,877 --> 00:50:08,912 SO THAT'S WHERE AI CAN HOPEFULLY 1408 00:50:08,912 --> 00:50:10,914 MAKE A DIFFERENCE. 1409 00:50:10,914 --> 00:50:11,582 >> HI, JONATHAN [INDISCERNIBLE] 1410 00:50:11,582 --> 00:50:15,085 I ALSO IN THE BIOETHICS 1411 00:50:15,085 --> 00:50:16,086 DEPARTMENT HERE. 1412 00:50:16,086 --> 00:50:20,891 SO I WAS REALLY--CAN YOU HEAR 1413 00:50:20,891 --> 00:50:21,625 ME? 1414 00:50:21,625 --> 00:50:21,892 >> YEAH. 1415 00:50:21,892 --> 00:50:23,727 >> SO I WAS REAL E EXCITED ABOUT 1416 00:50:23,727 --> 00:50:25,696 THE TALK, IT WAS INTERESTING. 1417 00:50:25,696 --> 00:50:27,030 I AM WONDERING IF YOU CAN SAY 1418 00:50:27,030 --> 00:50:28,832 MORE, THIS IS LESS OF A 1419 00:50:28,832 --> 00:50:30,000 SHORT-TERM CONCERN, MAYBE A 1420 00:50:30,000 --> 00:50:33,937 LONG-TERM THING, COULD YOU SAY A 1421 00:50:33,937 --> 00:50:36,607 BIT MORE ABOUT WHY YOUR--YOU 1422 00:50:36,607 --> 00:50:37,741 SEEM FAIRLY CONFIDENT THAT THIS 1423 00:50:37,741 --> 00:50:39,409 SORT OF THING WON'T BE A THREAT 1424 00:50:39,409 --> 00:50:44,014 TO YOB CORPSS FOR INSTANCE AND 1425 00:50:44,014 --> 00:50:46,083 IT SEEMS LIKE, I COULD SEE WHY 1426 00:50:46,083 --> 00:50:48,719 YOU WANT SOME KIND OF HUMAN 1427 00:50:48,719 --> 00:50:50,220 OVERSIGHT BUT HISTORICALLY WHEN 1428 00:50:50,220 --> 00:50:52,256 THERE'S SOME AUTOMATION THAT'S 1429 00:50:52,256 --> 00:50:53,423 INTRODUCED, IT'S USUALLY THIS 1430 00:50:53,423 --> 00:50:54,725 WON'T REPLACE YOUR JOB, BUT YOU 1431 00:50:54,725 --> 00:50:57,027 KNOW IF WE'RE TALKING ABOUT AN 1432 00:50:57,027 --> 00:50:58,695 AI CORRECTING A DOCTOR, COACHING 1433 00:50:58,695 --> 00:51:01,298 A DOCTOR AT A CERTAIN POINT, IT 1434 00:51:01,298 --> 00:51:04,735 MAKING ME WONDER, YOU KNOW WHY 1435 00:51:04,735 --> 00:51:06,536 INEVITABLY THIS SORT OF THING 1436 00:51:06,536 --> 00:51:08,939 WON'T LEAD TO JOB LOSS, SOIME 1437 00:51:08,939 --> 00:51:10,641 WONDERING IF YOU CAN EXPLAIN 1438 00:51:10,641 --> 00:51:10,841 THAT. 1439 00:51:10,841 --> 00:51:11,208 >> SURE, SURE. 1440 00:51:11,208 --> 00:51:12,676 I WANT TO BE CLEAR, I'M NOT 1441 00:51:12,676 --> 00:51:13,977 TALKING ABOUT THE ADMINISTRATIVE 1442 00:51:13,977 --> 00:51:16,613 JOBS LIKE CODING AND BILLING AND 1443 00:51:16,613 --> 00:51:18,282 BACK OFFICE, THOSE JOBS WILL BE 1444 00:51:18,282 --> 00:51:19,249 THREATENED BECAUSE ARK I CAN DO 1445 00:51:19,249 --> 00:51:20,651 A LOT OF THAT AND THEY CAN 1446 00:51:20,651 --> 00:51:25,422 REDUCE THE NUMBER OF PEOPLE IN 1447 00:51:25,422 --> 00:51:26,089 THOSE OPERATIONS CONSIDERABLY. 1448 00:51:26,089 --> 00:51:29,293 BUT WE HAVE A SHORTAGE OF 1449 00:51:29,293 --> 00:51:30,761 PHYSICIANS AND NURSES AND 1450 00:51:30,761 --> 00:51:31,828 PHARMACISTS AND MOST 1451 00:51:31,828 --> 00:51:33,430 PROFESSIONALS IN HEALTHCARE, SO 1452 00:51:33,430 --> 00:51:39,836 THE POINT IS IF WE CAN IMPROVE 1453 00:51:39,836 --> 00:51:41,305 THEIR PRODUCTIVITY AND REBOOT TO 1454 00:51:41,305 --> 00:51:44,975 GET BACK THE HUMAN-HUMAN BOND 1455 00:51:44,975 --> 00:51:47,110 THAT IS THE ESSENTIALITY OF 1456 00:51:47,110 --> 00:51:48,712 MEDICINE, THAT SHOULD BE OUR 1457 00:51:48,712 --> 00:51:49,613 OBJECTIVE. 1458 00:51:49,613 --> 00:51:53,417 THE QUESTION THOUGH, GOING BACK 1459 00:51:53,417 --> 00:51:57,621 TO JEFFREY HINTON HE PREDICTED 9 1460 00:51:57,621 --> 00:52:00,023 AGOS THAT WITHIN 5 YEARS WE 1461 00:52:00,023 --> 00:52:01,858 WOULDN'T HAVE ANY RADIOLOGISTS, 1462 00:52:01,858 --> 00:52:03,026 WELL, GUESS WHAT IN IT DOESN'T 1463 00:52:03,026 --> 00:52:04,328 HAVE AN EFFECT AND IT LIKELY 1464 00:52:04,328 --> 00:52:05,862 WILL HAVE AN EECT BUT THE POINT 1465 00:52:05,862 --> 00:52:07,831 IS THAT WE DON'T HAVE ENOUGH 1466 00:52:07,831 --> 00:52:09,333 PHYSICIANS, OUR MAN POWER, IF 1467 00:52:09,333 --> 00:52:13,704 YOU LOOK AT THE YB GROWTH IN THE 1468 00:52:13,704 --> 00:52:16,106 UNITED STATES, WHAT YOU SEE IS 1469 00:52:16,106 --> 00:52:18,275 HEALTHCARE KEEPS EVERY QUARTER, 1470 00:52:18,275 --> 00:52:21,979 EVERY YEAR, LEADS ALL OTHER JOBS 1471 00:52:21,979 --> 00:52:24,247 AND IT'S BASICALLY ACCOUNTING 1472 00:52:24,247 --> 00:52:26,216 FOR INCREASING SHARE OF OUR 1473 00:52:26,216 --> 00:52:30,120 GDPRKS P, SO WE HAVE--GDP SO WE 1474 00:52:30,120 --> 00:52:31,955 HAVE TO FIGURE OUT WAYS SOMEHOW 1475 00:52:31,955 --> 00:52:34,324 TO FIGURE OUT THE WORKFORCE 1476 00:52:34,324 --> 00:52:38,095 ISSUES AND HOW AI CAN HELP WITH 1477 00:52:38,095 --> 00:52:38,628 PRODUCTIVITY. 1478 00:52:38,628 --> 00:52:42,165 SO FOR EXAMPLE, IF A PHYSICIAN 1479 00:52:42,165 --> 00:52:43,900 HAS PAJAMA TIME WORKING ON THIS 1480 00:52:43,900 --> 00:52:45,202 AFTER HOURS, THEY'RE NOT DOING 1481 00:52:45,202 --> 00:52:47,471 THAT, BUT IF THEY HAVE MORE 1482 00:52:47,471 --> 00:52:48,605 TRUST AND WORKING WITH PATIENTS 1483 00:52:48,605 --> 00:52:49,406 BUT THAT'S WHEREY WOO WANT TO 1484 00:52:49,406 --> 00:52:49,873 GO. 1485 00:52:49,873 --> 00:52:51,541 THE HOPE IS TO IMPROVE MEDICINE 1486 00:52:51,541 --> 00:52:52,476 WITHOUT THREATENING THE YOB 1487 00:52:52,476 --> 00:52:57,748 CORPSS, THERE IS NO EVIDENT YET 1488 00:52:57,748 --> 00:52:59,783 OF CLINICIANS HAVE ANY REDUCTION 1489 00:52:59,783 --> 00:53:00,851 IN WORKFORCE, EVEN IN PLACES 1490 00:53:00,851 --> 00:53:04,254 OUTSIDE OF THE U.S., IN CHINA, 1491 00:53:04,254 --> 00:53:05,756 AND UK AND EUROPE EMPLOY SO I 1492 00:53:05,756 --> 00:53:11,395 DON'T THINK IT'S LIKELY. 1493 00:53:11,395 --> 00:53:13,997 >> [INDISCERNIBLE] FROM NIAMS, 1494 00:53:13,997 --> 00:53:14,965 YOU MENTIONED LLMs A LOT 1495 00:53:14,965 --> 00:53:16,800 NOVEMBER TALK AND I HAVE A 1496 00:53:16,800 --> 00:53:17,934 FEELING SOMEONE LISTENING TO 1497 00:53:17,934 --> 00:53:20,670 YOUR TALK MAY HAVE COME OUT WITH 1498 00:53:20,670 --> 00:53:24,741 THE IMPRESSION THAT THESE THINGS 1499 00:53:24,741 --> 00:53:27,077 BY DESIGN ARE NONFACTUAL AND 1500 00:53:27,077 --> 00:53:30,881 NONDETERMINISTIC RIGHT, SO YOU 1501 00:53:30,881 --> 00:53:32,315 DID MENTION THE CONTABUALATIONS 1502 00:53:32,315 --> 00:53:33,383 AT THE END WHICH IS A CLEAR 1503 00:53:33,383 --> 00:53:34,651 EXAMPLE OF THAT AND THE 1504 00:53:34,651 --> 00:53:38,021 CONQUENCES OF THAT ARE OBVIOUSLY 1505 00:53:38,021 --> 00:53:46,396 MAJOR WHEN WE APPLY THESE THINGS 1506 00:53:46,396 --> 00:53:48,031 TO MEDICINE SO SO WHAT WE'VE 1507 00:53:48,031 --> 00:53:50,100 BEGUN TO SEE, IS THE NLM, OR THE 1508 00:53:50,100 --> 00:53:52,102 NLM IS THE TALKING HEAD OF THE 1509 00:53:52,102 --> 00:53:54,071 MULTIMODAL SYSTEM, WHAT'S YOUR 1510 00:53:54,071 --> 00:53:56,239 SENSE OF WHERE NLMs WILL END 1511 00:53:56,239 --> 00:53:59,509 UP IN THE MULTIMODAL SPACE GIIVE 1512 00:53:59,509 --> 00:54:01,178 WE KNOW THAT THE NLM IS NOT THE 1513 00:54:01,178 --> 00:54:04,948 PART THAT WILL BE SORT OF 1514 00:54:04,948 --> 00:54:07,417 SOLVING THE PROBLEM. 1515 00:54:07,417 --> 00:54:09,219 DOES THE QUESTION MAKE SENSE? 1516 00:54:09,219 --> 00:54:10,954 >> YEAH, WE'RE SEEING A LOT OF 1517 00:54:10,954 --> 00:54:13,156 STRATEGIES TO REDUCE ERRORS, SO 1518 00:54:13,156 --> 00:54:15,892 FOR EXAMPLE, OPEN AI IS 1519 00:54:15,892 --> 00:54:17,494 INTRODUCING STRAW BERRY AS AN 1520 00:54:17,494 --> 00:54:19,696 NLM, AND IT THINKS BEFORE IT 1521 00:54:19,696 --> 00:54:21,231 GIVES AN ANSWER, THERE'S A DELAY 1522 00:54:21,231 --> 00:54:23,200 AND THAT THE QUESTION IS WILL 1523 00:54:23,200 --> 00:54:24,835 THAT LOWER--BUT WE HAVE SEEN 1524 00:54:24,835 --> 00:54:26,870 STRONG EVIDENT THAT OVER THE 1525 00:54:26,870 --> 00:54:33,643 LAST 2 YEARS, THE HALLUCINATION 1526 00:54:33,643 --> 00:54:34,811 CONFABULATION RATE KEEPS 1527 00:54:34,811 --> 00:54:35,112 DECREASING. 1528 00:54:35,112 --> 00:54:36,713 IT HAS TO BE KOWPED IN HERE WITH 1529 00:54:36,713 --> 00:54:38,115 THE HUMAN IN THE LOOP STORY 1530 00:54:38,115 --> 00:54:39,316 WHICH IS UNIQUE TO MEDICINE. 1531 00:54:39,316 --> 00:54:41,351 WHO'S GOING TO WANT TO MAKE A 1532 00:54:41,351 --> 00:54:42,652 DIAGNOSIS WITHOUT THAT TYPE 1533 00:54:42,652 --> 00:54:43,887 OF--SO I AM NOT AS CONCERNED 1534 00:54:43,887 --> 00:54:46,389 ABOUT IT BECAUSE I SEE IT AS 1535 00:54:46,389 --> 00:54:49,693 IMPROVING, I SEE VARIOUS 1536 00:54:49,693 --> 00:54:50,994 STRATEGIES, AND THE 1537 00:54:50,994 --> 00:54:52,863 CONVERSATIONAL CAPABILITY, AND 1538 00:54:52,863 --> 00:54:54,397 SO, AT THE PATIENT LEVEL, WE 1539 00:54:54,397 --> 00:54:57,134 DON'T TALK ENOUGH ABOUT THAT, 1540 00:54:57,134 --> 00:54:58,635 THE FACT THAT ANYBODY COULD PUT 1541 00:54:58,635 --> 00:55:00,370 IN DATA AND GET SOME IDEA WHAT 1542 00:55:00,370 --> 00:55:03,240 COULD BE GOING ON, COMTHAT TO A 1543 00:55:03,240 --> 00:55:04,741 GOOGLE SEARCH WHERE A PERSON 1544 00:55:04,741 --> 00:55:05,775 LOOKS UP SOMETHING ASK THEY 1545 00:55:05,775 --> 00:55:06,810 ALREADY HAVE THAT DISEASE AND 1546 00:55:06,810 --> 00:55:08,145 NONE OF THE DATA SPECIFIC TO 1547 00:55:08,145 --> 00:55:09,646 THEM, BECAUSE THEY DON'T HAVE 1548 00:55:09,646 --> 00:55:11,047 THAT CAPABILITY, SO I SEE 1549 00:55:11,047 --> 00:55:13,350 THERE'S LOTS OF POTENTIAL FOR 1550 00:55:13,350 --> 00:55:15,585 IMPROVEMENT BUT CAN WE REST ON 1551 00:55:15,585 --> 00:55:16,019 THAT? 1552 00:55:16,019 --> 00:55:18,488 NO, WE HAVE TO KEEP DEMANDING, 1553 00:55:18,488 --> 00:55:19,723 TO STRIVE, IT'LL NEVER BE 1554 00:55:19,723 --> 00:55:21,491 PERFECT BUT I'M NOT--I'M NOT AS 1555 00:55:21,491 --> 00:55:22,592 CONCERNED ABOUT THAT AS SOME OF 1556 00:55:22,592 --> 00:55:25,562 THE OTHER ISSUES LIKE THE 1557 00:55:25,562 --> 00:55:26,796 BIASES, THE INIQUITIES, THOSE 1558 00:55:26,796 --> 00:55:30,066 SORTS OF THINGS. 1559 00:55:30,066 --> 00:55:33,970 NTHANK YOU. 1560 00:55:33,970 --> 00:55:34,638 >> THANK YOU. 1561 00:55:34,638 --> 00:55:38,041 >> SO I'M MAT MAT 1562 00:55:38,041 --> 00:55:44,581 [INDISCERNIBLE] I'M NICHD, BUT I 1563 00:55:44,581 --> 00:55:47,551 GREW UP IN A RURAL AREA AND IT 1564 00:55:47,551 --> 00:55:51,588 SEEMS LIKE AI HAS A THE POWER TO 1565 00:55:51,588 --> 00:55:52,656 EXTRAPOLATE CLINICAL DATA AND GO 1566 00:55:52,656 --> 00:55:54,824 THERE WOULD YOU BE WILLING TO 1567 00:55:54,824 --> 00:55:56,359 TALK MORE ABOUT OTHER WAYS THAT 1568 00:55:56,359 --> 00:55:59,896 YOU COULD SEE AI HELPING 1569 00:55:59,896 --> 00:56:02,165 HEALTHCARE IN RURAL AREAS WHERE 1570 00:56:02,165 --> 00:56:04,234 DATA COLLECTION FOR PHYSICIANS 1571 00:56:04,234 --> 00:56:05,535 HAS HISTORICALLY BEEN LIMITED? 1572 00:56:05,535 --> 00:56:07,304 >> YEAH, I MEAN I THINK WHAT 1573 00:56:07,304 --> 00:56:08,939 WE'RE SEEING EVEN NOT JUST IN 1574 00:56:08,939 --> 00:56:19,416 RURAL U.S., BUT WORLD WIDE. 1575 00:56:21,551 --> 00:56:23,620 I MEAN WE HAVE PEOPLE IN AFRICA 1576 00:56:23,620 --> 00:56:25,288 WHO NEVER ACQUIRED AN ULTRA 1577 00:56:25,288 --> 00:56:26,356 SOUND IMAGE USING AI TO SAY PUT 1578 00:56:26,356 --> 00:56:31,461 IT ON THE CHEST, GO CLOCKWISE, 1579 00:56:31,461 --> 00:56:33,530 GO COUNTER CLOCKWISE, LIKE AN 1580 00:56:33,530 --> 00:56:36,266 AUTOMATED CAPTURE WHEN THE IMAGE 1581 00:56:36,266 --> 00:56:38,835 IS APPROPRIATE AND THEN FEETING 1582 00:56:38,835 --> 00:56:41,404 AN INTERPRETATION, SO TAKING 1583 00:56:41,404 --> 00:56:43,273 SOMETHING LIKE THAT THAT CAN BE 1584 00:56:43,273 --> 00:56:44,507 DONE ANYWHERE IN THE WORLD BOTH 1585 00:56:44,507 --> 00:56:46,309 IN THE ACQUIRING OF THE CLINICAL 1586 00:56:46,309 --> 00:56:49,546 GENETICSAGE AND THE ALGORITHM 1587 00:56:49,546 --> 00:56:50,013 FOR INTERPRETATION. 1588 00:56:50,013 --> 00:56:51,381 THIS IS JUST 1 EXAMPLE WHERE I 1589 00:56:51,381 --> 00:56:52,949 THINK THIS CAN GO BUT I DO THINK 1590 00:56:52,949 --> 00:56:56,119 THE REACH, WHEREVER THERE'S 1591 00:56:56,119 --> 00:57:00,390 ADEQUATE BAND WIDTH, WE CAN SEE 1592 00:57:00,390 --> 00:57:02,892 BIG CHANGES SO LEVELING THE TYPE 1593 00:57:02,892 --> 00:57:06,096 OF CAPABILITIES AND REDUCING THE 1594 00:57:06,096 --> 00:57:07,130 WORRY ABOUT INIQUITIES PEOPLE 1595 00:57:07,130 --> 00:57:08,632 DON'T SPEND ENOUGH TIME THINKING 1596 00:57:08,632 --> 00:57:10,634 ABOUT THE RURAL SIDE, USUALLY 1597 00:57:10,634 --> 00:57:11,935 ARE THINKING ABOUT 1598 00:57:11,935 --> 00:57:12,569 UNDERREPRESENTED MINORITIES BUT 1599 00:57:12,569 --> 00:57:14,938 I DO THINK IT'S A BIG ISSUE AND 1600 00:57:14,938 --> 00:57:19,776 I'M SAYING THAT IT WILL BE 1601 00:57:19,776 --> 00:57:20,443 IMPROVED. 1602 00:57:20,443 --> 00:57:21,544 >> ALL RIGHT, SO TO KEEP 1603 00:57:21,544 --> 00:57:22,946 EVERYBODY ON TIME WE WILL TAKE 1 1604 00:57:22,946 --> 00:57:24,447 LAST QUESTION AND FOR THOSE THAT 1605 00:57:24,447 --> 00:57:26,116 DON'T GET YOUR QUESTIONS ASKED 1606 00:57:26,116 --> 00:57:27,183 YOU CAN PERHAPS APPROACH AFTER 1607 00:57:27,183 --> 00:57:28,952 THE SESSION IS COMPLETED AND FOR 1608 00:57:28,952 --> 00:57:30,887 THOSE THAT ARE ONLINE, WE WILL 1609 00:57:30,887 --> 00:57:33,590 PASS YOUR QUESTIONS TO DR. TOPOL 1610 00:57:33,590 --> 00:57:34,357 AS WELL. 1611 00:57:34,357 --> 00:57:36,660 SO PLEASE, FINAL QUESTION. 1612 00:57:36,660 --> 00:57:38,261 >> HI, I'M JACOB [INDISCERNIBLE] 1613 00:57:38,261 --> 00:57:40,297 FROM THE CLINICAL NEUROIMACKING 1614 00:57:40,297 --> 00:57:42,966 RESEARCH CORE AT THE NIAA, THANK 1615 00:57:42,966 --> 00:57:44,868 YOU FOR THE INSPIRING YET 1616 00:57:44,868 --> 00:57:45,201 HUMBLING TALK. 1617 00:57:45,201 --> 00:57:47,304 IN THE CONTEXT OF ALL OF THESE 1618 00:57:47,304 --> 00:57:48,238 DRAMATIC CHANGES THAT WE'RE 1619 00:57:48,238 --> 00:57:50,540 FACING IN THE MEDICAL FIELD, DO 1620 00:57:50,540 --> 00:57:53,376 YOU HAVE ANY PIECE OF ADVICE FOR 1621 00:57:53,376 --> 00:57:54,911 ALL THE TRAINEES HERE THAT ARE 1622 00:57:54,911 --> 00:57:56,079 AT THE BEGINNING OF THEIR 1623 00:57:56,079 --> 00:57:58,248 CAREERS IN SCIENCE AND MEDICINE? 1624 00:57:58,248 --> 00:58:00,984 >> WELL MY ADVICE WOULD BE I 1625 00:58:00,984 --> 00:58:02,285 WISH I WAS YOU BECAUSE YOU'RE 1626 00:58:02,285 --> 00:58:06,623 GOING TO BE IN THE MOST EXCITING 1627 00:58:06,623 --> 00:58:07,157 TIME OF MEDICINE. 1628 00:58:07,157 --> 00:58:08,892 AND YOU DON'T NECESSARILY HAVE 1629 00:58:08,892 --> 00:58:11,861 TO BE AN AI EXPERT BUT YOU 1630 00:58:11,861 --> 00:58:13,596 SHOULD UNDERSTAND THE NUANCES 1631 00:58:13,596 --> 00:58:14,864 AND GET ENOUGH GROUNDING AND 1632 00:58:14,864 --> 00:58:18,268 HAVE THE TYPE OF COLLABORATIONS 1633 00:58:18,268 --> 00:58:20,403 YOU NEED TO TEST THESE NEW 1634 00:58:20,403 --> 00:58:22,839 WATERS AND TOOLS AND HOW THEY 1635 00:58:22,839 --> 00:58:23,873 IMPLEMENT THEM, WHEN TO 1636 00:58:23,873 --> 00:58:25,709 IMPLEMENT THEM BUT IT IS REALLY 1637 00:58:25,709 --> 00:58:26,743 EXCITING BECAUSE WE'VE BEEN IN 1638 00:58:26,743 --> 00:58:29,713 NEED OF A RESCUE IN MEDICINE FOR 1639 00:58:29,713 --> 00:58:31,214 A NUMBER OF YEARS. 1640 00:58:31,214 --> 00:58:35,985 WE HAVE ALL THIS DATA ALL 1641 00:58:35,985 --> 00:58:38,288 DRESSED UP AND NOWHERE TO GO, 1642 00:58:38,288 --> 00:58:39,055 HOARDING DATA, STORING DATA AND 1643 00:58:39,055 --> 00:58:41,157 TO YOU WE HAVE A WAY FORWARD 1644 00:58:41,157 --> 00:58:42,258 THAT IS QUITE EXTRAORDINARY. 1645 00:58:42,258 --> 00:58:45,795 SO IT'S GOING TO TAKE TIME IN 1646 00:58:45,795 --> 00:58:46,830 MEDICINE, HEALTHCARE ALWAYS 1647 00:58:46,830 --> 00:58:50,300 TAKES LONGER THAN IT SHOULD. 1648 00:58:50,300 --> 00:58:52,969 WE HAD TO HAVE A PANDEMIC BEFORE 1649 00:58:52,969 --> 00:58:54,771 TELEMEDICINE EVER CAME IN USE, 1650 00:58:54,771 --> 00:58:55,939 SO HOPEFULLY THIS WON'T REQUIRE 1651 00:58:55,939 --> 00:58:58,575 SOMETHING LIKE THAT BUT IT TAKES 1652 00:58:58,575 --> 00:59:00,076 LONGER BUT YOU'RE BEING YOUNG 1653 00:59:00,076 --> 00:59:03,480 AND ALL THE OTHER TRAINEES HERE 1654 00:59:03,480 --> 00:59:05,849 ARE IN A VERY PREPICIOUS 1655 00:59:05,849 --> 00:59:06,649 SITUATION AND POSITION BECAUSE 1656 00:59:06,649 --> 00:59:07,517 YOU'RE GOING TO SEE IT. 1657 00:59:07,517 --> 00:59:09,986 IT'S GOING TO HAPPEN IN YOUR 1658 00:59:09,986 --> 00:59:13,690 CARE OF PATIENTS AND YOUR CARER 1659 00:59:13,690 --> 00:59:15,225 AND THAT'S SO EXCITING. 1660 00:59:15,225 --> 00:59:17,794 >> THANK YOU DR. TOPOL FOR A 1661 00:59:17,794 --> 00:59:24,167 WONDERFUL TALK . 1662 00:59:24,167 --> 00:59:34,377 [ APPLAUSE ]