I'M FRANCIS COLLINS, THE NIH DIRECTOR AS YOU PROBABLY KNOW BY NOW AND IT'S MY GREAT HONOR TO BE GIVEN THE OPPORTUNITY TO INTRODUCE TODAY'S LECTURE AT THE WEDNESDAY AFTERNOON LECTURE AND THIS IS ERIC TOPOL WHO WILL TALK ABOUT TOWARDS A DEEPER AND MORE HUMAN CENTRIC--MUSEUM NANOG CENTRIC MEDICINE, I THINK HIGHS TITLED A BIT. TELL BE SOMETHING LIKE THIS. DR. TOPOL, ONE OF THE BETTER KNOWN NAMES IN MEDICINE AND PARTICULARLY IN PRECISION MEDICINE, GOT HIS UNDERGRADUATE DEGREE AT THE UNIVERSITY OF VIRGINIA. I NOTICED HIS THESIS WAS ON PROSPECTS FOR GENETIC THERAPY IN MAN, SO HE WAS A BIT OF HIS TIME BECAUSE NO 1 HAD DONE THAT WHEN HE WAS WRITE THAGOREAN PARTICULAR THESIS, WENT TO GET HIS MD DEGREE AT UNIVERSITY OF ROCHESTER AND DID POST DOCTORAL TRAINING AT HOPKINS AND IN CORONY ANGIO PLAOF THEY AT THE SAN FRANCISCO HEART EYEWITNESSITUTE AND THEN ASSOCIATE AND FULL PROFESSOR THE UNIVERSITY OF MICHIGAN WHERE HE AND I OVERLAPPED FOR FOR QUITE A FEW YEARS, HE MOVED TO CASE WESTERN RESEARCH IN UNIVERSITY, BECAME PROFESSOR OF MEDICINE AT THE CLEVELAND CLINIC AND WAS THE FOUNDER AND PROVOST OF A VERY SIGNIFICANT MEDICAL SCHOOL, THE CLEVELAND CLINIC LEARNER COLLEGE OF MEDICINE, WHICH WOULD BE AN INTERESTING STORY IN ITS OWN RIGHT. IN 2007 HE MOVED TO THE SCRIPPS RESEARCH INSTITUTE IN LAHOYA, IN ACADEMIC SCRIPTS HEALTH DIRECTOR OF TRANSLATIONAL RESEARCH INSTITUTE AND LOTS OF TITLES INDEED. HIS RESEARCH AS A CARDIOLOGYST HAS CERTAINLY HAD MAJOR IMPACTS IN TERMS OF WHAT WE KNOW AND WHAT WOE DON'T KNOW ABOUT THE MANAGEMENT OF HEART DISEASE, BUT IT IS EXTENDED BROADLY ACROSS THIS IN MANY WAYS BUT PARTICULARLY IN THE AREA OF PRECISION MEDICINE. HE IS WELL KNOWN FOR HIS ROLE AS EDITOR IN CHIEF OF MED SCAPE. HE'S WRITTEN 3 BOOKS ALL OF WHICH I HAVE READ WHICH ARE QUITE PROVOCATIVE IN A FUTURISTIC SORT OF WAY, THE CREATIVE DESTRUCTION OF MEDICINE, THE FIRST 1, OKAY NOT EVERYBODY IN THE MEDICAL PROFESSION PROBABLY LIKED THAT TITLE BUT IT WAS WELL CHOSEN. IN 2015, THE PATIENT WILL SEE YOU NOW. EMPOWERING PATIENTS. AND THEN MOST RECENTLY AND SOMETHING THAT WILL TAP INTO TODAY'S LECTURE, DEEP MEDICINE OW ARTIFICIAL INTELLIGENCE CAN MAKE HEALTHCARE HUMAN AGAIN. MANY PEOPLE WORRY THAT IT'S GOING TO GO THE OTHER WAY, WE WILL HEAR ABOUT THAT. ON TOP OF THAT, ERIC IS, I WOULD SAY A WORLD CLASS TWITTER MEISTER, HIS TWEETS I LOOK AT EVERY DAY ARE, I THINK PROBABLY THE INFORMATION-RICH OF ANYTHING IN BIOMEDICINE. ERIC DOESN'T JUST PUT A TWEET WITH A FEW CLEVER WORDS, ALTHOUGH THERE ARE CLEVER WORDS HE JUST DOESN'T JUST PUT UP CARTOONS BUT SOME ARE CARTOONS AND IT'S NICE TO HAVE A BREAK THERE, TOO, BUT MOST ARE IN FACT BASED UPON PUBLICATIONS WHERE IT'S NOT JUST SUMMARIZING THE RESULT BUT SHOWING YOU THE DAILY BASIS AT SO YOU WILL SEE A GRAPH OR A TABLE AND SOME COMMENTARY IN APPROPRIATE HIGHLIGHTING WHICH HE SAYS IS HIS WAY OF ARCHIVING THE LITERATURE AND I HADN'T REALLY THOUGHT ABOUT HOW THAT COULD BE USEFUL TILL I SAW HOW HE DID THAT AND SO HIS 432,000 TWITTER FOLLOWERS WHICH INCLUDE A LOT OF YOU AT NIH ARE REALLY GRATEFUL FOR THE WAY IN WHICH HE KEEPS PEOPLE ABREAST OF WHAT'S HAPPENING IN THE LITERATURE AND OF COURSE FOR THE LAST 16 MONTHS MUCH OF THAT HAS BEEN COVID BUT HE IS INTO MANY OTHER SPACES AS WELL. SO WE'RE IN FOR A TREAT. I'M GLAD TO SAY ERIC IS SOMEBODY I'VE HAD THE PRIVILEGE OF WORKING WITH IN HIS ROLE RIGHT NOW WITH THE ALL OF US PRECISION MEDICINE EFFORT, IS PARTICULARLY CRITICAL TO THE WHOLE FORCE SUCCESS OF THIS STRONG PERSPECTIVE COHORT STUDY AND THE WAY IN WHICH HE WITH ALL KINDS OF CREATIVE ACTIVITY IS TRYING TO PUSH THE LIMIT OF WHAT'S POSSIBLE AND DO AN EVEN MORE EXCITING TECHNOLOGICAL SPACE WHILE RETAINING THIS CRITICAL ASPECT OF KEEPING MEDICINE ABOUT PEOPLE AND MAINTAINING THAT COMPASSION AND THAT HUMAN FOCUS. SO WITHOUT FURTHER ADO, LET ME ASK YOU TO WELCOME AS BEST YOU CAN VIRTUALLY PROFESSOR ERIC T OPOL, AND ON THE AI PATH TO DEEPER AND MORE ACCURATE MEDICINE, DELIGHTED TO HAVE YOU. >> WELL, FRANCIS, THRILLED TO BE WITH YOU AND THANKS SO MUCH FOR YOUR KIND INTRODUCTION AND MORE THAN THAT THANKS FOR YOUR JUST REMARKABLE LEADERSHIP OF THE NIH OVER MANY YEARS. SO WE WILL GET INTO THE STATUS OF AI TODAY IN HEALTHCARE AND MEDICINE AND IT'S MOVING QUICKLY AND FOR SHORT-TERM BUT ALSO FAR REACHING. SO THE FIRST POINT TO MAKE IS THAT WE HAVE A LOT OF DIAGNOSTIC ERRORS THAT WE ARE RELUCTANT TO MOVE UP TO BUT WE ARE LOOKING AT LEAST 12 MILLION SIGNIFICANT ERRORS A YEAR AND WHEN WE SAW THE NATIONAL ACADEMY OF MEDICINE REPORT ON THAT ISSUES THE FACT THAT EACH OF US WILL EXPERIENCE 1 OF THESE IN OUR LIFETIME IS OF COURSE DISCONCERTING. IF A PHYSICIAN THINKS OF THE CORRECT DIAGNOSIS IN THE FIRST 5 MINUTES TO UNDERSCORE THE IMPORTANCE OF TIME, WHICH WE WILL GET BACK TO, THE ACCURACY'S GREAT, BUT IF IT DOESN'T HAPPEN QUICKLY, IT FALLS OFF DRAMATICALLY, AND THEN INTERESTINGLY THAT DOCTORS WHO ARE SAYING WHAT THE DIAGNOSIS IS, AND WHEN THERE'S AN AUTOPSIES TO COMPARE THE ACTUAL DIAGNOSIS, THERE'S ACTUALLY WHEN THEY SAY THEY'RE COMPLETELY CERTAIN, THEY'RE WRONG 40% OF THE TIME SO WE HAVE A WAYS TO GO TO IMPROVE DIAGNOSTIC ACRASE MODEL SKPE BAKUGAN THAT'S WHERE AI WILL HAVE ITS BIGGEST EARLY IMPACT. AS FRANCIS MENTIONED PRECISION MEDICINE HAS BECOME AN IMPORTANT OBJECTIVE AND A BUZZ WORD BUT ACTUALLY WHAT WE REALLY WANT IS NOT JUST PRECISION MEDICINE THAT IS MAKING THE SAME MISTAKES CONSISTENTLY. WE WANT ACCURATE MEDICINE, BUT THAT'S A MOUTHFUL TO SAY PRECISION AND ACCURATE MEDICINE BUT THAT'S REALLILY THE END GOAL AND THAT'S WHERE AI CAN MAKE A BIG DIFFERENCE. THE WAY WE CAN GET THERE ARE THESE DEEP NEURAL NETWORKS WHICH IS REALLY JUST A SUBTYPE OF AI OVER RECENT YEARS AND THESE ARTIFICIAL NEURONS, THEY'RE NOT ALL THAT SIMILAR TO OUR REAL NEURONS BUT THEY SET UP A NETWORK WITH THE INPUTS OF DATA, SPEECH, TEXT AND THAT GOES THROUGH THE ARTIFICIAL LAYERS OF NEURONS TO THEN DERIVE AT OUTPUT AND THAT DEEP NEURAL NETWORK IS THE BASIS OF THIS ACCURACY STEP UP. SO TO GIVE YOU AN EXAMPLE, IT WAS A CHEST X-RAY AND READ BY AN EXPERIENCED CHEST KRAI RADIOLOGIST BUT WHEN THE CHEST X-RAY IS PUT THROUGH A DEEP NURRAL NETWORK, IT PICKS UP A NODULE THAT PROVED TO BE CANCEROUS WHEN BIOPSIED SO THESE ARE THE TYPES OF MISTAKES OR LACK OF PICKING UP THAT WE CAN BY TRAINING A DEEP NEURAL NETWORK WITH HUNDREDS OF THOUSANDS OF CHEST X-RAYS IF NOT EVEN MILLIONS WE CAN DO THIS ROUTINELY. HERE'S AN EXAMPLE WITH MAMMOGRAPHY, THE SAME TYPE OF PHENOMENA, THIS IS THE LARNLEST STUDY TO DATE OUT OF NYU WITH 14 RADIOLOGIST COMPARED TO OVER A MILLION MAMMOGRAMS AND YOU SEE THE ABILITY THAT THIS AUGMENTATION OF INTERPRETATION AND ACCURACY WITH A TRAINED DEEP NEURAL NETWORK. AND THIS IS REGINA BARSLY'S WORK, THIS HAS BEEN AN ISSUE, DOES THE NEURAL NETWORK ONLY 1 IN 1 PLACE FOR THE MAMMOGRAPHY SYSTEM, HERE'S 1 THAT, WITHED AROUND THE WORLD AND PROJECTED 5 YEAR PROCEEDINGINOSEIS BETTER THAN ANYTHING PREVIOUSLY. NOW TO CONVEY THE POWER OF DEEP NEURAL NETWORKS THIS HAS NO PRACTICAL VALUE BUT IT'S POWERFUL, I THINK PERHAPS THE BEST DEMONSTRATION OF AIs CAPACITY IF WE PRECEPT THIS PICTURE OF THE RETINA TO THE LEADING RETINAL EXPERTS AND ASKED THEM IS THAT A MALE OR FEMALE RETINA PHOTO FROM THE MALE OR FEMALE. THE CHANCE OF THEM GETTING THAT RIGHT IS 50%. BUT IF WE PUT THAT THROUGH A DEEP NEURAL NETWORK, IT'S 97% ACCURACY. NOW THERE ARE BETTER WAYS TO DETERMINE THE GENDER THAN THE RETINAL PICTURE BUT I THINK IT CAN BE THAT WE CAN HAVE MACHINES SEE THINGS BETTER THAN HUMANS AND CERTAINLY THE BEST IS AFFUSION OF EFFORTS AND RECENTLY, PIERCE KEEN AND COLLEAGUES FROM PIERCE MOORE FIELD HOSPITAL IN THE UK THIS PAST WEEK PUBLISHED THE FACT THAT THIS WAS REPLICATED AND NOT ONLY WAS IT REPLICATED BUT THIS IS NOTE WORTHY. IT WAS DONE BY CLINICIANS WHO HAD ABSOLUTELY THOUGH CODING EXPERIENCE. AND THIS IS SOMETHING THAT IS EMNITRIC OXIDELY TEACHABLE TO PROFESSIONS TO USE THESEOT MACHINE LEARNING ALGORITHMS. NOW IT ISN'T JUST THE GENDER, AS I MENTIONED, THAT IS REALLY LITTLE PRACTICAL VALUE BUT IT TURNED OUT THE RETINA IS A GATEWAY FOR DEEP NEURAL NETWORKS THAT EXTENDS FROM DIAGNOSING AND TRACKING KIDNEY DISEASE FOR AC1 C, THE BLOOD PRESSURE CONTROL, A WINDOW INTO ALZHEIMER PATIENT HYMER'S DISEASE, PREDICTING THE COURSE AND WHEN TO INTERVENE FOR MACKULAR DEGENERATION AND THIS PAPER PUBLISHED LAST MONTH WAS ALSO A SURPRISE, IN CARDIOLOGY WE USE CALCIUM SCORES THAT IS A CT SCAN TO GAUGE THE RISK BUT YOU CAN ACTUALLY PREDICT THE CARDIOVASCULAR RISK JUST AS WELL THROUGH THE DEEP NEURAL NETWORK AND ANALYSIS OF THE RETINAL VESSELS WHICH IS REALLY PRETTY SURPRISING. SO THE POWER OF AI TRAINED RETINAL IMAGES IS ACTUALLY FAR SURFACING WHAT WE WOULD HAVE SURMISED. NOW, AS IT TURNS OUT, IT'S EVEN DEEPER THAN THAT. BECAUSE NOW WE CAN LOOK AT THE CELLULAR AND SUBCELLULAR LEVEL WITHOUT ANY STAIN, WITHOUT FLUORESCENCE, WE HAVE THE ABILITY TO DO THINGS THAT WERE NOT EVEN CONCEIVABLE, SUCH AS THIS: THAT IS WHEN LOOKING AT THE RETINA TO BE ABLE TO SEE LIVE MACROFAINCH THESE CELLS ARE VERY IMPORTANT. THEY DOMAINON STRAIGHT INFLAMMATION IN THE CENTRAL NERVOUS SYSTEM AND TO BE ABLE TO SEE THAT BECAUSE OF THE ABILITY TO TRAIN MACHINE ALGORITHMS THIS IS ACTUALLY I THINK NOTE WORTHY. BUT IT GOES BEYOND THAT. THAT IS TO TRACK HEPATITIS EAT O BILIARY DISEASE AND WE'RE NOT JUST TALKING ABOUT JAUNDICE THAT WE COULD EASILY DISCERN BUT WE'RE TALKING ABOUT THE HEPAT O BILIARY TREE THROUGH THE RETINAL EYE GROUND. NOW IT'S NOT JUST THE RETINA, HERA THE ELECTROCARDIO GRAMS, I LOVE READING THEM, I'VE BEEN DOING IT FOR A FEW DECADES ABOUT I WOULD NEVER BE ABLE TO TELL WHAT THE AGE OR SEX OF THE PATIENT BY LOOKING AT THE CARDIO GRAM. BUT THAT IS PRETTY STRAIGHT FORWARD WITH DEEP NEURAL NETWORKS AS IS THE EJECTION FRACTION, AS IS THE HEMOGLOBIN SURPRISINGLY AND THEN MAKING DIFFICULT DIAGNOSIS THAT SOMETIMES CAN BE MISSED LIKE AMYLOID HEART OR PULL MONITORARY HYPER TENSION. INTERESTINGLY JUST FINGERTIPS ON A SENSOR CAN DETECT A CONGENITAL LUNG QT SYNDROME WHICH HAS A RISK OF SUDDEN DEATH. THIS IS A RECENT STUDY WITH WELL OVER 1.6 MILLION CARDIO GRAMS THAT WERE TRAINED AND EVALUATED AT MAYO CLINIC. THIS IS A RARE RANDOMIZED STUDY FROM MAYO CLINIC THAT WAS JUST PUBLISHED AND INTERESTING, WHAT IT DID WAS USING THE MACHINE INTERPRETED ECG, WITH PRIMARY CARE PHYSICIANS WHO ARE NOT NOTED FOR BEING EXPERTS IN ECG INTERPRETATION AND IT WAS RANDOMIZED TO SHOW THAT THERE WAS FAR MORE ACCURATE DIAGNOSIS WITH THE GROUP THAT HAD A PRIMARY CARE PHYSICIAN THAT HAD THE AUGMENTATION WITH MACHINE DEEP LEARNING. NOW MOVING FROM THE RETINA AND ECG, WELL THE'S LOOK AT PATHOLOGY. THIS IS ACTUALLY I THINK PRETTY ASTOUNDING THAT WHEN YOU LOOK AT A SLIDE AS A PUTTINGOLOGYST, I THINK YOU KNOW THERE'S LOTS OF DISAGREEMENTS BETWEEN PATHOLOGISTS WHAT THEY'RE SEEING JUST ON AN H& E SLIDE BUT AS IT TURNS OUT YOU CAN NOT ONLY GLEAN THAT INFORMATION CONSISTENTLY WHEN YOU HAVE DEEP LEARNING SUPPORT BUT CAN YOU ALSO SEE THINGS THAT ARE NOT SEEABLE BY HUMAN EYES, TRAINING WITH AGAIN HUNDREDS OF THOUSANDS IF NOT MILLIONS OF SLIDES, SUCH AS THE STRUCTURAL VARIANTS, THE DRIVER MUTATION, THE ORIGIN, PRIMARY OF THE TUMOR WHICH SOMETIMES IS NOT KNOWN. AND SO THIS IS SOMETHING NOW WE'RE SEEING MANY REPLICATIONS OF, SO, THIS LAST WEEK, A NEW ALGORITHM CALLED TOE FOR TUMOR ORIGIN ASSESSMENT REPLICATE THAD WORK THAT IS DEFINING VERY ACCURATELY THE PRIMARY TUMOR WHEN OFTEN THAT IS NOT KNOWN OR NOT DEFINITIVE. SO THESE ARE WAYS THAT PATHOLOGY IS MAKING A DIFFERENCE WITH DEEP LEARNING SUPPORT, BUT AUSJUST PUBLISHED, THIS WEEK, WAS AN EXCELLENT REVIEW IN NATURE MEDICINE WHICH REALLY GETS TO THE FACT THAT WE'RE VERY EARLY IN PATHOLOGY, THERE HASN'T BEEN ENOUGH VALIDATION AND PROOF OF CLINICAL UTILITY, NO LESS THE FACT THAT MOST PATHOLOGY LABS ACROSS HEALTH SYSTEMS HAVE NOT SWITCHED TO POST SLIDE IMAGES THAT, IS DIGITIZING THE PATHOLOGY. NOW WHAT ABOUT SKIN. THIS WAS 1 OF THE EARLIEST NATURE PAPERS OUT OF STANFORD SHOWING YOU COULD TRAIN THE DIAGNOSIS OF SKIN LESIONS. WE WILL GET BACK TO THAT SHORTLY BUT JUST TO POINT OUT MUCH WORK IS BUILT ON THAT INITIAL 2017 WORK FROM STANFORD, NOW, VIRTUALLY ANY TYPE OF SKIN CANCER, SKIN RASH, SKIN LESION, AND THEN OF COURSE THE 2 ARE COMBINED EVEN BETTER FOR PERFORMANCE ACACCURACY, THIS EXTENDS SO MANY WAYS ABOUT YOU ANOTHER 1 IS THE THYROID NODULE QUESTION WHICH FREQUENTLY COMES OUT, SHOULD THIS BE BIOPSIED AND NOW WEB CONNECTED HAVE DEEP LEARNING SUPPORT TO BE ABLE TO HELP MAKE THAT DECISION AS TO WHETHER BIOPSY WOULD BE INDICATED. NOW THIS I THINK HELPS TO VISUALIZE IN REALTIME COLONOSCOPY, FROM JAPAN, WHERE YOU CAN NOT ONLY--THE GASTROENTEROLOGYST SEE POLYPS THAT OTHERWISE WILL BE MISSED BUT IN REALTIME WITH MACHINE VISUAL SAY WHETHER OR NOT IT HAS A HIGH LIKELIHOOD OF BEING CANCEROUS TO HELP GUIDE WHETHER THE BIOPSY SHOULD BE DONE AND THIS IS JUST STARTING TO GET ADOPTED IN THE U.S. BUT THERE'S BEEN SEVERAL RANDOMIZED TRIALS SHOWING THE SUPERIORITY OF REALTIME MACHINE VISION DURINGENDOSCOPY, ALL THOSE TRIALS HAVING BEEN DONE IN CHINA. SO I JUST WANT TO LEAVE YOU AT THE FIRST POINT HERE IS THAT THIS IS SOMETHING, DEEP LEARNING THAT'S GOING TO HAVE ITS EFFECTS ACROSS THE BOARD. THERE IS NO SPECIALTY OR DOMAIN IN MEDICINE OR HEALTHCARE EXTENDING TO PHARMACISTS AND PARAMEDICS, NURSES THAT WILL NOT BENEFIT FROM SOME MACHINE SUPPORT, LEANING ON MACHINES IF YOU WILL. AND IN THE HOSPITAL SETTING, THE USE OF SENSORS WITH MACHINE VISION IS STARTING ON ALSO BECOME IMPORTANT AND THIS IS FROM A REVIEW BY FAYE-FAYE LEE AND COLLEAGUES AT STAND FORD THAT APPEARED LAST FALL AND WHAT YOU CAN SEE IS THAT THE ABILITY TO MONITOR THINGS LIKE THE ADEQUACY OF HAND WASHING OR SUPPORTING PATIENTS IN THE ICU GETTING AWAY WITH LESS 1 TO 1 NURSING TO AVOID INADVERTENT EXTUBATION, ALL SORTS OF THINGS THAT ARE HAPPENING PREVENTING FALLS IN THE HOSPITAL WITH PATIENTS WHO ARE FRAIL, WITH MACHINE VISION. THE OTHER POINT TO MAKE IS THIS IS OCCURRING ACROSS THE HEALTH SPAN THAT, IS EVERYTHING FROM SELECTION OF EMBRYOS WHICH HAS NOT BEEN REVAMPED FROM OLD MANUEL MEANS FROM YEARS AGO WITH LOTS OF INN RACKSYS ALL THE WAY THROUGH PREDICTING OUTCOME WHEN IS A PATIENT PRESENTS TO THE HOSPITAL. BUT IT'S ALSO THE PRACTICAL ISSUE, NOBODY WOULD LIKE TO SPEND A LARGE--LONG TIME IN AN MRI MACHINE OR HAVE IMAGES THAT ARE SUBOPTIMAL, BUT NOW WITH DEEP LEARNING OF THE IMAGES BOTH MRI, CT, PET SCANS, THESE IMAGES CAN BE OBTAINED MUCH MORE QUICKLY AND THE RESOLUTION CAN BE MUCH HIGHER BY TAKING IMAGES THAT ARE NOT SO CRISP TO MAKE THEM MARKEDLY IMPROVE VALUE. NOW I MENTION THAT CLINICAL VALIDATION IS KIND OF A MISSING LINK, THERE'S ONLY A LIMITED NUMBER OF PERSPECTIVE STUDIES AND THEY'RE ALL SUMMARIZED ON THIS TABLE, SO WE DON'T HAVE ENOUGH CLINICAL VALIDATION, MOST OF IT IS ALL FROM RETROSPECTIVE STUDIES, THERE HAVE BEEN HUNDREDS OF THOSE BUT PROSPECTIVE STUDIES ARE DIFFERENT. IT'S THE REAL WORLD, AND IT'S MORE CHALLENGING AND EVEN MORE CHALLENGING ARE RANDOMIZED TRIALS AND I MENTIONED ALMOST ALL THE 1S WITHENDOSCOPY AND ALMOST ALL THE RANDOMIZED TRIALS TO DATE HAVE BEEN IN CHINA AND ONLY RECENTLY ARE WE STARTING TO SEE RANDOMIZED TRIALS ELSEWHERE, THE MOST RECENT 1 I MENTIONED WAS ON USE OF CARDIO GRAMS ON PRIMARY CARE PHYSICIAN. SO THE GOOD THING IN THIS FIELD IS THAT IT'S NEW, ONLY A FEW YEARS OLD, REALLY AND WHAT'S HAPPENED IS THAT THERE ALREADY HAS BEEN A CONSORT IN SPRINT DELPHI GLOBAL CONFERENCE WHICH TOOK CONSIDERABLY EFFORT IN INTERNATIONAL COOPERATION TO PUT OUT GUIDELINES ABOUT WHAT SHOULD BE THE PROTOCOL, WHAT SHOULD BE IN THE PUBLICATIONS AND THEN DISSEMINATION SO NOW WE HAVE STANDARDS WHICH ARE REALLY GOING TO TAKE THIS DEAL TO THE NEW HEIGHTS AND THE JOURNALS, THE LEADING JOURNALS TO BE BE OBLIGED TO FOLLOW THE CONSULLOR AND AI STANDARD. SO I WANT TO LEAVE YOU AT THIS POINT WITH THIS STATEMENT FROM ANTONIO DEEMILIA PHYSICIAN, WHICH IS QUITE ACCURATE HE WROTE MACHINES WILL NOT REPLACE PHYSICIANS, BUT PHYSICIANS USING AI WILL SOON REPLACE THOSE NOT USING IT. I THINK THAT'S ACCURATE. WE'RE NOT THERE YET BUT WE'RE IN. INCHING OUR WAY TOWARDS IT. NOW I WANT TO REVIEW SOME REAL THINGS THAT ARE--THAT ARE NOT RESEARCH, IT'S GENERATING RESEARCH BUT IT'S HELPING SICK NEWBORNS. THIS IS WORK THAT EMANATED FROM STEPHEN KINGS MOORE AND A GROUP AT GRADEY CHILDREN'S HOSPITAL WHICH IS OUR PARTNER FOR OUR CTSA, AND WHAT'S STRIKING HERE IS THAT WITH VARIOUS AI ALGORITHMS, DIFFERENT DEEP LEARNING SOFTWARE, THEY'RE TAKING SICK NEW BORNS AND YOUNG BABIES FROM A SAMPLE OF THEIR GENOME, BLOOD SAMPLE TO ACTUAL MANAGEMENT OF A DIAGNOSED CONDITION ACCURATELY, WITHIN 18 HOURS AND NOT JUST HERE IN SAN DIEGO COUNTY WHICH IS OUR CHILDREN'S HOSPITAL BUT NOW DISSEMINATING THAT THROUGHOUT THE COUNTRY AND THIS IS A STRIKING PROGRAM BECAUSE IN MANY OF THESE BABIES, EVERY MINUTE COUNTS TO GET THE DIAGNOSIS RIGHT AND TO GET THIS DOWN TO 13 HOURS IS A FEAT. SO WHAT YOU'RE SEEING HOOKER IS THE DIFFERENT PHASES FROM A CRITICALLY ILL INFANT. THE FACT THAT THERE'S A SAMPLE THAT HAS TO GO THROUGH PREP AND THEN THERE HAS TO BE THE SEQUENCING AND THE VARIANT CALLING, WHILE THAT'S HAPPENING WHICH IS AN AI PACKAGE IN ITSELF, THERE'S ALSO THE EXTRACTION OF ALL THE DAILY BASIS THEA, FORTUNATELY THERE'S NOT A LOT OF IT ON THE HEALTH RECORD BUT YOUR LABS, THERE'S THE SYMPTOMS AND ANYTHING THAT'S RELEVANT SCANNED ARE ALSO PUT INTO THE MIX AND THEN THERE IS THE AUTOMATED DIAGNOSIS THROUGH DIFFERENT PACKAGES LIKE FABRIC AND THEN FINALLY, NOW, AUTOMATED MANAGEMENT, SO THAT EVEN NEOINATOLOGYST AND PEDIATRICIANS WITHOUT RARE DISEASE BACKGROUNDS WILL HAVE THE BEST ADVICE AS TO HOW TO MANAGE A BABY, SO HERE'S THE DIFFERENT PHASES, THIS IS THE NATURAL LANGUAGE PROCESSING TO GET TO THE DIAGNOSIS THAT IS THE--ALL THE FEATURES OF THE CLINICAL PRESENTATION. THEN THERE'S THE AUTOMATED INTERPRETATION OF THE GENOMICS OF THE GENOMIC DATA AND THERE'S 2 DIFFERENT PACKAGES THAT HAVE BEEN IN USE IN THIS EFFORT. AND THEN, THERE'S THIS RAPID MANAGEMENT TOOL WHICH ACTUALLY IS COMBINING A FEW DIFFERENT AI TOOLS. AND THIS IS REALLY STRIKING, BECAUSE THIS TO ME, IS UPENDING THE PATTERN IN MEDICINE. IT USUALLY IS ADULTS FIRST AND THE LAST PEOPLE WHO BENEFIT FROM ANY NEW THING THAT'S VALIDATED ARE BABIES AND CHILDREN. HERE IT'S THE BABIES THAT ARE LEADING THE WAY. AND IT'S REALLY STRIKING HOW MANY HAVE HAD THEIR LIVES SAVED OR THEIR CONDITION PROPERLY DIAGNOSED AND PREVENT THINGS LIKE BRAIN DAMAGE AND OTHER HORRIBLE ADVERSEY QUELL I BY THIS PROGRAM. NOW LET ME TURN TO THE DOWN SIDE OR THE LIABILITIES OF AI IN MEDICINE. THIS IS A REALLY IMPORTANT REPORT FROM THE KLEIN CENTER AT HARVARD AND IT GOT INTO ALL THE ISSUES ABOUT--THAT WE'RE CONCERNED ABOUT, SUCH AS PRIVACY, SUCH AS SECURITY OF THE DATA, THE EXPLAINABILITY THAT IS THE TRANSPEARNSES OR BLACK BOX ISSUES OF THESE ALGORITHMS, FAIRNESS, BIAS, THE INIQUITIES THAT THIS COULD MAKE WORSE, AND ALL THE DIFFERENT CONCERN POINTS. ONE OF THESE WAS EMPHASIZED BY A RECENT PAPER AND I THINK IT'S QUITE ILLUSTRATIVE, THIS IS FROM BERKELEY AND COLLEAGUES IN PUBLISHING NATURE AND MEDICINE, IT SHOWS THAT IT'S NOT THE ALGORITHM THAT'S THE PROBLEM, THE AI PER SE, IT'S THE DATA INPUTS AND HERE IS ABOUT KNEE AFTIO ARTHRITIS AND WHETHER SOMEONE WOULD BE CONSIDERED FOR A KNEE REPLACEMENT. AND WHAT'S FASCINATING IS THAT GOING BACK ON THIS, THE ALGORITHMS, OR THE GUIDELINES, EXCUSE ME, THAT ARE USING CLINICAL PRACTICE THAT ARE NOT ALGAR RHYTHMS AND THE GUIDELINES ARE BASED ON A PAPER IN 1952 FROM COAL MINERS, ALL WHITE COAL MINERS AND WHEN THIS WAS SUBJECT TO REASSESSMENT, IT TURNED OUT THERE WERE MARKED DIFFERENCES IN THE CLINICAL PRESENTATION OF BLACKS VERSUS WHITES AND THAT THIS DATED 60, 70-YEAR DATED GUIDANCE HAD TO BE REASSESSED. SO IT'S THE INPUTS THAT ARE IN MANY WAYS RESULTING IN THE BIASES THAT OCCUR. NOW, I WANT TO TURN TO THE PATIENT SIDE BECAUSE THIS IS REALLY WHAT I WOULD CONSIDER A QUIET REVOLUTION. EVERYONE THINKS ABOUT AI'S HELPING DOCTORS AND CLINICIANS, IT TURNS OUT IT'S ALSO HELPING PATIENTS. AND THIS IS THE IDEA THAT PEOPLE CAN HAVE AI, OVERWORKED CLINICIAN WORKFORCE. AND IT GIVES PEOPLE MORE CHARGE. AND SO, WHEN IT'S DONE PROPERLY, WHICH OBVIOUSLY IS WHEN AND IF CAN HAVE REALLY TRANSFORMATIVE IMPACT. SO THE FIRST 1, I THAT GOT FDA CLEARANCE WHERE THE SMART WATCH DIAGNOSIS OF ATRIOLE FIB ROUGH ATOM LIEWGZ, FIRST BY A SMALL COMPANY, A LIVE CORE AND A YEAR LATER BY APPLE. THIS CAN BE VERY HELPFUL IN PASHTS WHO HAVE RISKS OR PRIOR HISTORY OF ATRIOLE FIB ROUGH ATOM LIEWGZ, UNFORTUNATELY IT'S USED INDISCRIMINATELY AND LEAD TO FALSE DIAGNOSIS AND ALARMS IN PEOPLE WHO HAVE NOR RISK OR HAVE NO ALGORITHM IN THEIR WATCH. BEFORE I TURN TO THIS, OTHER APPLICATION OF PATIENT POWER, I JUST WANT TO MENTION THAT THE TOOL RELIES ON A DEEP LEARNING ALGORITHM OF RESTING HEART RATE. SO IF YOUR RESTING HEART RATE IS ELEVATED SUBSTANTIALLY OVER YOUR NORM, YOUR NORM, THAT IS THE TIP OFF TO GETTING AN ECG THROUGH YOUR WATCH AND NAILING THE DIAGNOSE ACCURATELY OF ATRIOLE FIBRO LIEWGZ HERE'S ABOUT URINARY TRACT INFECTION WHICH IS IS NOW A COMMON PRACTICE IN UK AND OTHER COUNTRY ASKS THERE ARE 5 IN THE CUE FOR THE FDA HERE IN THE U.S. >> PATIENTS CAN NOW DO THE SAME URINE TEST AT HOME. SO THIS ACCURATEY DIAGNOSIS A UTI AS WELL AS CLINICAL LABS AND IT'S ALL DONE THROUGH AN AI KIT THAT CAN BE BOUGHT FROM A DRUG STORE IN MANY PHARMACIES IN OTHER COUNTRIES AND SOME POINT WILL BE HERE IN THE U.S., SO THIS IS JUST ANOTHER EXAMPLE THAT HAPPENED YESTERDAY. THERE IS A PROGRAM THAT GOOGLE PUT TOGETHER CALLED DERM ASSIST AND THEY ARE PUTTING IT OUT FOR FREE FOR DIAGNOSING ALL TYPES OF DIFFERENT SKIN LESIONS, THIS WAS VALIDATED FOR A RECENT PUBLICATION IN JAMMA AND IT WILL BE IN THE U.S. LATER THIS YEAR SO PEOPLE WILL BE ABLE TO TAKE PICTURES AND GET AUTOMATED FEEDBACK BEFORE THEY GO TO SEE A DERMATOLOGIST AS TO WHETHER THIS SHOULD BE SOMETHING OF CONCERN BIOPSY FOR POTENTIAL SKIN CANCER OR KIND OF RASH OR OTHER SKIN PROBLEM THEY'RE HAVING WHICH I THINK YOU KNOW IS 1 OF THE MOST COMMON REASONS WHY PEOPLE GO TO A PHYSICIAN. AND THE OTHER 1 I WANTED TO MENTION WHICH ISN'T A SELF-DIAGNOSIS. IT DOESN'T DOCTORLESS IF YOU WILL BUT IT'S ALSO NOTE WORTHY IS HALF THE PATIENTS IN IN COUNTRY WITH DIABETES NEVER GET SCREENED FOR DIABETIC RETINAL LOCATION NUMBER OF PATIENTSATHY. BUT NOW, AFTER PROSPECTIVE STUDY, THIS SYSTEM IS NOW IN GROCERY STORES IN MANY PLACES THROUGHOUT THE U.S. AND FOR THE FIRST TIME PEOPLE WHO MIGHT NEVER GET SCREENING FOR DIABETIC RETINAL LOCATIONIN OPERATING GLOBALLYATHY A LEADING CAUSE OF PREVENTIBLE BLINDNESS ARE NOW GETTING SCREENED ACLIGHTLY SO THIS IS A STEP IN THE RIGHT DIRECTION AS IT MOVES EVENTUALLY TO SMART PHONE SELF-DIAGNOSIS OF RETINAL LOCATIONIN OPATHY. EVENTUALLY WE WILL GET TO THE VIRTUAL MACHINE MEDICAL SYSTEM. WE'RE IN THE EARLIEST DAYS OF THAT BUT EVENTUALLY ALL THE DIFFERENT LAYERINGS OF A PERSON'S DATA SHOULD THAT INDIVIDUAL WANT TO HAVE THIS INTEGRATED AND FED BACK IN REALTIME IPT GREATER CLUING THE RELEVANT MEDICAL LITERATURE, LITERATURE ABOUT THAT PERSON AND ALL THESE OTHER LAYERS OF DAT ATHE ENVIRONMENT, THE BIOLOGIC, PHYSIOLOGIC LAYERS ALL WILL THEN HELP A PERSON PREVEPT AN ILLNESS THAT OTHERWISE MIGHT HAVE OCCURRED, OR BETTER MANAGED LIKE WE'RE SEEING THE BEGINNING OF FOR PEOPLE WITH DIABETES WITH JUST A 1 FOCUS, 1 DIMENSION OF THIS TYPE OF VIRTUAL COACH PHENOMENA. SO I WANTED TO SHOW YOU 1 TECHNOLOGY WHICH YOU WILL SEE A LOT MORE OF IN THE FUTURE EVEN AT THE PATIENT LEVEL AND THAT IS SMART PHONE ULTRA SOUND. THIS TO ME IS 1 OF THE MOST REMARKABLE EXAMPLES OF WHERE WE CAN GO WITH AI AND CURRENT TECHNOLOGIES, THIS IS A PROBE THAT FITS INTO THE BASE OF A SMART PHONE AND IMMEDIATELY GETS VERY HIGH RESOLUTION IMAGES THAT WE AND OTHER VS PUBLISHED, ARE AS GOOD AS WE CAN GET WITH EXPENSIVE MACHINES AND CLINICS AND HOSPITALS. HERE FOR ME, I HAVEN'T LISTENED TO A HEART IN MANY YEARS, ACTUALLY MORE THAN A DECADE WHICH I CAN SEE EVERYTHING WITHIN SECONDS AND SHARE IT REALTIME WITH THE PATIENT AND WHY WOULD YOU LISTEN TO [INDISCERNIBLE] WHEN YOU CAN SEE THE CHAMBERS OF THE HEART, THE STRENGTH OF THE MUSCLES, THE VALVES AND BASICALLY EVERYTHING AND ALSO EVEN TRACK THE BLOOD FLOW TO SEE WHETHER--MOST OF THE ORGAN TO THE BODY IS NOT AS CHALLENGING, BUT WHAT'S INTERESTING NOW IS AI GUIDED ECHO, SO THAT A PERSON WITH NO BACKGROUND OF HOW TO DO AN ECHO AS LONG AS THEY CAN PUT THE PROBE ON THE CHEST, ON THE LEFT SIDE OF THE CHEST, THEY WILL THEN GET DIRECTED TO MOVE THE PROBE UP OR DOWN OR CLOCKWISE OR COUNTER CLOCKWISE AND AS SOON AS THE'MAGE, JUST LIKE IN THE URINE ANALYSIS, AS SOON AS THE IMAGE IS DESIRED THAT IS OBTAINED WITHOUT KNOWLEDGE OF THE PERSON HOLDING THE PROBE, THE IMAGE IS AUTOMATICALLY CAPTURED, THE VIDEO LOOP AND SO THIS IS JUST AN EXAMPLE OF THAT HERE OF AUTOCAPTURE AND AUTOLABELLING. SO IT TELLS WHAT ARE THE DIFFERENT FEATURES THAT ARE BEING IDENTIFIED. IT GIVES AUTOMATED INJECTION FRACTION AND AS WE MOVE FORWARD IT WILL GIVE AUTOMATED HIGH QUALITY INTERPRETATION, WHICH WILL REQUIRE OVERSIGHT BY CARDIOLOGYST OR RADIOLOGYSTS. NOW WHY IS THIS SO NOTE WORTHY? BECAUSE IT ISN'T JUST FOR THE HEART, EVERY PART OF OF THE BODY, EXCEPT FOR THE BRAIN CAN BE IMAGED THROUGH ULTRA SOUND. SO HOW YOU GO THE THIS DEVICE, I WAS SO ENAMORED BY IT, I DID A TOTAL BODY MEDICAL SELFIE INCLUDING CAROTID ARTERY, THYROID, LUNG, HEART, I DIDN'T EVEN KNOW HOW TO IMAGE SOME OF THESE PARTS OF THE BODY LIKE WILL GAL BLADDER OR KIDNEY BUT THERE WERE VIDEO HOW TO TEACH ME ON GOOGLE AND I DID THE ENTIRE BODY IN A MATTER OF MINUTES AND THIS WENTIE IN THE INIF AREIOR VENA CAVA AND TKOUP MARKER TO MY LEFT FOOT. NOW WHY IS THIS IMPORTANT? WELL WE TALKED ABOUT INEQUITIES IN HEALTHCARE AROUND THE WORLD IN FACT BUT NOW SMART PHONE ULTRA SOUND IS BEING USED IN THE HINDER LANS OF PLACES, AFRICA AND INDIA WHERE PEOPLE WHO HAVE NO MEDICAL BACKGROUND ARE OBTAINING IMAGES THAT ARE GETTING THEN INTERPRETED NOT SO MUCH AUTOMATICALLY YET, BUT THEY WILL BE MORE AND MORE OVERTIME AND OF COURSE, THIS IS A STEP FORWARD, LEVELING THE EARTH FOR HIGH-TECHNOLOGY MEDIATED THROUGH A MART PHONE WITH AI HEALTH. NOW THAT TURNS OUT TO BE, I MEAN ANOTHER FACET OF THIS, IS TAKING COMPLEX CONGENITAL HEART DISEASE AND A LIST OF ALL THESE DIFFERENT COMPLEX CONDITIONS IN HEART CONDITIONS AND WHAT YOU CAN SEE IS [INDISCERNIBLE] THE GROUP OF UCSF JUST PUBLISHED DAYS AGO THE ABILITY TO TAKE THOUSANDS OF THESE ECHO CARDIO GRAMS FROM THESE COMPLEX TO INTERPRET AND GET VERY HIGH ACCURACY WITH EXTREMELY HIGH SENSITIVITY AND PREDICTIVE VALUES, SO THIS IS A REAL BIG STEP FOR SOMETHING THAT WOULD OTHERWISE BE REMARKABLY COMPLEX AND I JUST WANT TO MENTION THIS BECAUSE THIS IS ANOTHER FACET OF JUMPS IN AI THAT WE HAVE NOW, THE ABILITY TO PUT A SENSOR DIRECTLY ON THE BRAIN AND WITH PEOPLE WHO ARE LOCKED IN TO BE ABLE TO HAVE BRAIN TO SPEECH AND THEN JUST LAST WEEK BRAIN TO TEXT AND TAKEN--THEY'S BY USING CONVOLUNTEERSUTIONAL NETWORK TO TAKE THE BRAIN WAVES AND TRANSLATE THAT INTO A PERSON'S SPEECH OR TEXT. PRETTY DARN REMARKABLE. NOW THE QUESTION THAT COMES UP IS WHAT ABOUT THE PANDEMIC? HAS AI HELPED ACCELERATE THE PAN DELATWALLIC? I WROTE A BRIEF ESSAY ABOUT THIS IN WALSTREET JOURNEY BUT IT TURNS OUT AI HAS DONE LITTLE TO HELP. THE THINGS THAT WE'VE SEEN REMARKABLE LIKE THE VACCINES AND OTHER IMPROVEMENTS THAT HAVE HELPED US GET THROUGH THIS PANDEMIC HAVEN'T REALLY RELIED ON AI AT ALL, THERE WAS 1 DRUG THAT WAS FACILITATED BY AI DISCOVERY EFFORT, BUT THIS HEIGHT HAS BEEN REMARKABLE, SO, HERE'S AN ARTICLE FROM LAST YEAR WILL AI SAVE US FROM CORONAVIRUS, THE ANSWER IS NO. AND THEN THIS JEFF HINTON WHO IS A FATHER OF DEEP LEARNING FROM UNIVERSITY OF TORONTO, DEEP LEARNING IS GOING TO BE ABLE TO DO EVERYTHING, NO, NONAPOPTOTIC THE REALLY. IT'S IMPORTANT TO NOTICE THAT THE AREA THAT HAD THE MOST INTEREST OF AI ARE THE RAPID DIAGNOSIS OF CHEST X-RAYS OR CT SCANS. AND AT 1 POINT WHEN WE'RE HAVING PROBLEMS WITH TESTING, PC R TESTING FOR COVID, THERE WAS THE PROSPECT OF WELL, MAYBE WE SHOULD JUST USE THE CHEST X-RAY OR THE KREBS CYCLE, TR SCAN, TURNS OUT THIS WAS REVIEWED INDEPTH AND IT TURNS OUT THAT IN THE MANUSCRIPT PER SE,--OR POTENTIAL CLINICAL USE DUE TO FLAWS OR UNDERLYING BIASES. SO WE ARE STILL EARLY IN AI, AND THE COVID EXPERIENCE EMPHASIZES THAT. IT HASN'T MADE ANY SUBSTANTIVE IMPACT WHETHER IT BE ON REMOTE MONITORING, VALIDATION AND MEDICAL IMAGE INTERPRETATION, IT'S REALLY HAD A VERY LIMITED ROLE SO FAR. NOW THE LAST POINT TO LEAVE YOU WITH BEFORE TAKING QUESTIONS, AND GETTING YOUR COMMENTS IS THE MOST IMPORTANT AND MOST OVERARCHING GOAL IN MY VIEW OF AI AND HEALTHCARE AND THAT IS TO BRING BACK THE HUMAN SIDE. THIS IS THE--THE LAST THING IS TO HAVE THE ROBOT WILL SLEEP APNEA AND OBESITY YOU NOW, WE KIND VF THAT NOW BECAUSE WE HAVE SO LITTLE TIME WITH PATIENTS AND THEY'RE NOT EVEN LOOKING AT THE PATIENT BECAUSE THEY'RE WORKING--CLINICIANS ARE WORKING AT KEYBOARDS AND THERE'S LIMITED TIME, AND WHEN I HAD THE FORTUNE OF WORKING WITH THE NHS TO DO THE REVIEW OF THEIR HEALTHCARE SYSTEM, AND PLANNING THE FUTURE FOR THEIR WORKFORCE. AND THIS REVIEW IS FREELYY AVAILABLE I THINK IT'S REALLY HELPFUL BECAUSE IT RELIED ON WHERE AI WILL TAKE US AND WHAT WAS INTERESTING IS KEYBOARD AND THERE IS I CONVERSATION BETWEEN OR A NURSE AND PATIENT OR DOCTOR AND PATIENT TO MAKE SYNTHETIC NOTES TO BRING BACK THE EYE TO EYE CONTACT AND JUST GET RID OF COWBOYS THAT IS ALL THE THINGS THAT THE KEYBOARDS ARE USED FOR, SUCH AS THE ORDERS AND TESTS AND RETURN APPOINTMENTS ALL CAN BE DONE THROUGH VOICE, AND SO ENTHETIC NOTES ARE PROVING TO BE REMARKABLILY BETTER THAN THE NOTES WE HAVE TODAY, THROUGH OUR HEALTH INFORMATION SYSTEMS AND IT TURNS OUT IN THE UK, THE TRANSFORMATIVE IMPACT OF JUST 1 MINUTE LIBERATION FROM KEYBOARDS WAS THE MOST IMPRESSIVE THING FROM OUR TRANSDISCIPLINARY TEAM OF PEOPLE THAT WORKED ON THIS WHOLE CHALLENGE. SO GET BEING GETTING BACK TO THE SYSTEM 1 THINKING WHICH IS THE WAY MEDICINE IS TODAY AND THE DIAGNOSTIC INACCURACIES IS TIED INTO THIS GIFT OF TIME CONCEPT. THAT IS WE DON'T HAVE ENOUGH TIME AND THE LIMITED TIME WE HAVE ARE PECKING AT KEYBOARD NOT INTERACTING WITH PATIENTS SO SYSTEM I'M THINKING OF DANNY KAHNEMAN'S WORK, WE NEED THE SYSTEM TO REFLECTIVE RATHER THAN REFLEXIVE THINKING AND SO RIGHT NOW MEDICINE IS FAST AND SHALLOW. NOT ONLY IS THAT LEADING TO THE INACCURACY BUT ALSO THE LACK OF THE HUMAN CENTRIC PART OF MEDICINE. SO WE WANT TO GET BACK THIS GIFT OF TIME THAT'S BEEN TAKING AWAY WITH THIS STEADY EROSION OF MEDICINE OVER DECADES AND THAT WILL BE A COMING THROUGH KEYBOARD LIBERATION, THROUGH THE SYNTHESIS OF A PATIENT'S DATA SET WHICH IS GETTING BIGGER ALL THE TIME. THE PRIMARY SCREENING OF THE IMAGES, THROUGH ALGORITHMS ISSUES THE SIDE THAT I ALLUDED TO OF SCREENING AND ULTIMATELY IN THE YEARS AHEAD THIS MEDICAL COACH, CONSTANT. SO DANIEL [INDISCERNIBLE] HAS DONE WORK IN THIS SPACE AND HAS A RECENT BOOK ON THIS WHOLE PROBLEM OF MEDICAL ERRORS BUT THE CHANCE TO THINK AND THAT HER STATEMENT HERE IN THIS NEW ENGLAND JOURNAL ESSAY BEYOND THE FINANCIAL WAYS MODERN MEDICINE, MEDICAL PRACTICE IS A PETRI DISH FOR MEDICAL HARM AND PHYSICIAN BURN OUT SO THERE ISN'T AN ALGR RHYTHM FOR EMPAGHT SKPE BAKUGAN WE CAN AGREE IN THIS CARTOON, WE TWEAK ANY EMPATHY ALGORITHM. THIS IS WHAT WE ARE FOR, THE HUMAN CONNECTION. AND SO AS WE GO FORWARD, OUR PERFORMANCE, WE WON'T BECOME SUDDENLY MORE INTELLIGENT BUT MACHINES ARE, THEY ARE GOING TO GET PETTER AND WITH DEEP LEARNING AND HYBRID MODELS OF DEEP NEURAL NETWORKS SO OUR CHARGE IS TO GET MORE HUMAN AND SO THE PERSON WHO NAILED THIS, WAS BACK IN 1927, IT WAS ACTUALLY FRANCIS PEABODY, HE WROTE A 6 PAGE PAPER IN JAMMA AT A TIME WHEN SINGLE AUTHOR PAPERS WERE MORE COMMON AND HAD IS AN INSPIRING READ IF YOU HAVEN'T READ IT OR READ IT RECENT LOAMACY AND I JUST WANT TO LEAVE YOU WITH THE LAST 9 OF THIS PAPER, 1 OF THE ESSENTIAL QUALITIES OF THE CLINICIAN IS INTEREST IN HUMANITY, FOR THE SECRET OF THE CARE OF THE PATIENT IS IN CARING FOR THE PATIENT. I WANT TO LEAVE YOU WITH THE CONCEPT THAT IF WE DO THIS RIGHT, AI CAN HELP US ACHIEVE--AND THE REMARKABLE SUPPORT WE GET FROM NIH AND I WANT TO TURN TO YOUR QUESTIONS AND INTERACT. THANKS VERY MUCH FOR HAVING ME. >> WOW DR. TOPOL, THAT WAS A WONDERFUL TOUR DEFORCE OF AI AND MEDICINE. THANK YOU SO MUCH FOR THAT TALK. WE HAVE A LOT OF QUESTIONS. >> THAT'S GREAT. I LOVE IT I SHOULD HAVE LEFT MORE TIME. >> THANK YOU FOR LEAVING TIME FOR THAT SO I WILL JUDGE JUST DIG IN BECAUSE IT WILL GIVE A CHANCE TO SAY MORE ABOUT THESE TOPICS AND THE FIRST QUESTION IS 1 THAT ACTUALLY CAME BEFORE YOU ANSWERED IT BUT I WILL ASK IT ANYWAY, AGAIN AND THE QUESTION WAS IS AI ML GOING TO REPLACE DOCTORS IN THE FUTURE? AND YOU DID ANSWER THAT. SO MAYBE EVEN EXPAND A BIT MORE ON MAYBE WHAT THE UPTAKE IS IN THE PHYSICIAN WORLD AND WHAT PERCENTAGE OF PHYSICIANS ARE ACTUALLY USING AI NOW? >> YEAH, AI HAS NOT HAD MUCH IMPACT YET. I THINK THE BEST EXAMPLE, SHINING EXAMPLE AS I MENTIONED IN THE NEONATE SPACE BUT IN TERMS OF IMAGE INTERPRETATION, MANY RADIOLOGYSTS ARE STARTING TO USE THESE ALGORITHMS FOR THE VARIOUS TYPES OF IMAGES AND I THINK WHAT WE KNOW IS THAT THEIR ALWAYS HAS TO BE OVERSIGHT. YOU DON'T WANT TO ENTRUST AN IMPORTANT DIAGNOSIS TO AN ALGORITHM THAT COULD BE FAULTY, COULD BE SUBJECT TO ADVERSARIAL ATTACKS, IT'S SOFTWARE, WE ALL KNOW THE GLITCHES THAT CAN OCCUR WITH SOFTWARE, THE QUESTION A LOT OF PEOPLE SAY IS WHAT ABOUT WHEN IT KEEPS GETTING BETTER, WHAT ARE RADIOLOGIST GOING TO DO BECAUSE THEY IN PATHOLOGISTS ARE THE 1S THAT ARE MOST VULNERABLE TO CHANGE. MY VIEW THERE IS THAT THERE'S SO MUCH MORE THEY CAN DO THAN JUST OVERSIGHT OF SCANS AND THEIR COMPLIMENTARY ABILITY BECAUSE THEY CAN THEN SEE THE PATIENT. INTERACT WITH THE PATIENT. THEY CAN BE THE BROKER TO RENT UNNECESSARY SCANS OR OPERATIONS AND PROCEDURES BECAUSE THEY HAVE NO VESTED INTEREST, THEY ARE JUST TRYING TO CULL THE DATA TOGETHER. SO BEING GATE KEEPERS AND HAVING MORE PEASHT INTERACTION, AS I TALK TO RADIOLOGISTS AND PATHOLOGIES, THEY ACTUALLY WELCOME THAT POTENTIAL IN THE TIMES AHEAD. SO I DON'T--I THINK WE WILL SEE SOME PHENOTYPIC CHANGES BUT THAT STATEMENT FROM ANTONIO DELEVELS INCREASEIT IS I THINK ACCURATE IS THAT AI WILL BECOME PART OF THE NORM. >> YEAH, THAT'S GREAT. AND IT DOES GET BACK TO THAT LAST POINT YOU LEFT US WITH. IT'S ABOUT THE PATIENT AND CARING FOR THE PATIENT. THERE WAS ANOTHER QUESTION, YOU SHOWED A VARIETY OF IMAGES OF THE RETINA, AND THE QUESTION WAS WHY IS THE RETINA SUCH AN EXCELLENT WINDOW INTO SO MANY ASSPEBTS OF THE BODY? >> YEAH, IT'S MAIRKSZING TO ME, I'M ACTUALLY IN AWE OF THIS. I HAVE TO SAY, THIS WAS A SURPRISE TO ME, HOW DEEP THIS IS GOING ACROSS SO MANY DIFFERENT AREAS IN MEDICINE. WE'VE KNOWN THE RETINA IS GETTING TO THE BRAIN. WE ALSO KNOW THAT BLOOD VESSELS ARE THERE AND THERE'S SO MUCH THAT WE CAN GLEAN, WHO WOULD EVER HAVE THOUGHT WE GD COULD LOOK LIVE MACROPHAGES AND GET AN EARLY INSIGHT ABOUT NEURODEGENERATIVE DISEASES ARE EXAMPLE, BUT THEN JUST THE OTHER DAY WHEN I READ THIS PAPER ABOUT THE CORONARY ARTERY CALCIUM SCORES THROUGH RETINA, I SAID WHOA, IN IS AMAZING. SO IT'S SURPRISING US AND THE FACT THAT THERE ARE NOW LOTS OF EFFORTS SO THAT PEOPLE CAN DO THIS OF THEMSELVES, THEY JUST GET THEIR OWN RETINAL PHOTOS WITHOUT HAVING TO DILATE THEIR EYES SO THIS IS ACTUALLY, I THINK TAKING A LOT OF--IN FACT SOME OF THE LEADING OPHTHALMOLOGISTS ARE SURPRISED HOW OF A GATEWAY THIS IS TO THE HUMAN BODY AND I THINK TO BE ABLE TO SAY THAT WE'LL BE TRACKING TECHNOLOGY TRANSFERS LIKE BLOOD PRESSURE AND DIABETES, AND OTHER CONDITIONS THROUGH THE EYE GROUND IS NO LESS STRIKING THAN THE NEUROLOGIC CONDITIONS. SO I'M ACTUALLY MUCH MORE WORK TO BE DONE BUT THE EARLY DATA IS PRETTY STRIKING. >> THE EYES HAVE IT SO TO SPEAK. >> EYES, EYES, EYE. YEAH. REALLY. >> I WANT TO TAKE A BRIEF PAUSE FOR JUST A SECOND TO REPEAT THE CME CODE. AND I WANT TO SAY THE CME CODE 1 MORE TIME IS 31309. SO THE CME CODE IS 31309. THE BOOK ON DEEP MEDICINE, IN THAT YOU TALKED ABOUT SORT OF 2 POSSIBILITIES THAT COULD HAPPEN, YOU SAID 1 WAS THAT AI WILL IMPROVE HEALTH AND HEALTHCARE AND THE OTHER 1 IS THAT IT WILL MAKE IT MUCH WORSE, AND SO, WHICH ARE YOU THINKING? HOW CAN WE KIND OF FIGURE OUT WHICH PATH WE'RE GOING TO GO DOWN? >> RIGHT. THIS IS ACTUALLY, I THINK MAYBE THE MOST IMPORTANT POINT OF ALL JOANY AND THAT IS THAT WE HAVE KIND OF BIFURCATION, WE THIS VERY POWERFUL TOOL, YOU KNOW IN MY VIEW, THE 2 MOST POWERFUL THINGS IN BIOMEDICINE TODAY ARE AI AND CRSPR GENOME EDITING AND SO THIS TOOL CAN BE USED 2-EDGED, THAT IS IT CAN MAKE THINGS WORSE LIKE INIQUITIES, IT COULD MAKE THINGS BURN OUT AND THE MORAL OF HEALTHCARE WORKFORCE WORSE AND ALL THE THINGS IT CAN MAKE WORSE OR CAN IT PACIFIC NORTHWEST IT MUCH BETTER, NONAPOPTOTIC YOU, WHAT'S INTERESTING IS IF WE JUST KEEP THINGS THE WAY THEY ARE, WHERE WE HAVE THE MAIN DRIVING FORCE AND REVENUE ASK ADMINISTRATORS AND OVERLOADERS AND HEALTH SYSTEMS, THE BEAN COUNTERS IF YOU WILL, IF WE JUST LEAVE THAT--IF THEY'RE THE 1S MANAGING HOW HEALTHCARE IS PROVIDED WHAT IS GOING TO HAPPEN FROM THIS EFFICIENCY GAIN FROM AI IS OH, WELL, YOU KNOW WHAT? READ MORE SCANS. READ MORE SLIDES. SEE MORE PATIENTS. BECAUSE HAVE YOU ALL THIS MACHINE HELP NOW AND WE CAN BE THE STAND FOR THAT. WE HAVE TO THEN ORGANIZE AND PREVENT THAT PHENOMENA JUST AS WE HAVE TO THEN USE OUR TOOLS TO GET RID OF INIQUITIES BECAUSE IT COULD OBVIOUSLY MAKE THINGS MUCH WOES ARE AND THERE WE HAVE TO BE FESSING UP NOT TO BLAME THE AI, BUT BLAME OURSELVES AND BEFORE WE GO INTO ANY PROJECT OF USING NEURAL NETWORKS, WE HAVE TO DO ANYTHING POSSIBLE TO ERADICATE BIAS, HAVE DIVERSE INPUTS THAT IS FROM ALL TYPES OF PEOPLE WHETHER IT'S RACE ETHNICITY, AGE, I MEAN ANY POSSIBLE DIVERSITY, NEEDS TO BE MAXIMIZED OTHERWISE THE CHANCE FOR OUTPUT WILL BE BE COMPROMISED SOEE HAVE A LOT OF WORK TO DO, THAT'S THE HARD PART OF AI. IS NOT LETTING IT GO IN THE WRONG DIRECTION. >> I'M SO GLAD YOU SAID THAT. IT'S SUCH AN IMPORTANT POINT AND YOU KNOW YOU SHOWED A FEW OF THE IMAGES WITH THE SKIN BIOPSIES AND HOW YOU MIGHT LOOK AT THEM AND WE KNOW ALREADY THAT THE PATIENT DIVERSITY IS SO IMPORTANT TO MAKE SURE WE'RE NOT INSERTING BIAS INTO HOW THOSE IMAGES CAN BE READ AND IT'S SUCH AN IMPORTANT POINT. I WANT TO--THERE'S ANOTHER QUESTION THAT SORT OF RELATES TO THIS, NOT IN TERMS OF BIAS PER SE, BUT ACTUALLY THE--THE SPECIFICITY, AND SENSITIVITY OF THE DEVICES FOR PATIENTS WHO ARE SELF-DIAGNOSING AND IN THINKING ABOUT THE SKIN LESION EXAMPLE THAT YOU SHOWED, THE QUESTION IS HOW IS THE SENSITIVITY AND SPECIFICITY OF SUCH DEVICES COMPARED TO CURRENT STANDARD SYSTEM THAT A REGULATORY PROCESS THAT'S BEING UNDERTAKEN BY THE FDA, HOW ARE THOSE THINGS BEING TAKEN INTO ACCOUNT? >> YEAH THIS, IS ANOTHER GREAT QUESTION BECAUSE WHEN THE FDA REVIEWS THE DATA, IT DOESN'T NORMALLY RARELY GET PUBLISHED, SO THE MEDICAL COMMUNITY DOESN'T GET TO SEE THE SENSITIVITY SPECIFICITY DATA AND MOST OF THESE ARE GETTING 5, 10 K CLEARANCE WHICH IS A LOWER BAR THAN APPROVAL. SO THIS IS A PROBLEM IN MY VIEW, IT'S NOT THE EXPLAINABLE TRANSPARENCY ISSUE THAT THE BLACK BOX ALGORITHM, IT'S MORE THE BLACK BOX FDA ROUTE OF COMPANIES THAT ARE PROPRIETARY ALGORITHMS AND HERE THE PROBLEM IS THEY'RE RETROSPECTIVE. SO PROSPECTIVE PUBLISHED IN PEER REVIEW JOURNAL, THAT ROUTE, IS RARE, AS I POINTED OUT AND THAT'S WHAT WE NEED SO THAT THE MEDICAL COMMUNITY CAN SEE ALL THE DATA AND GET THE SENSITIVITY, SPECIFICITY ACCURACY ALL OUT THERE, BUT UNFORTUNATELY SO FAR MOST OF WHAT WE HAVE THAT'S BEEN FDA CLEARS OR RETROSPECTIVE STUDIES, SMALL, WE DON'T--WE NEVER SEEN THE DATA, IT'S FOR SALE, YOU KNOW? IT'S JUST NOT RIGHT. WE HAVE TO GET THIS THING FIXED EARLY. AND I'M HOPING WE WILL SEE A LOT MORE PROSPECTIVE STUDIES LIKE FOR EXAMPLE, THAT DIABETIC RETINAL LOCATION NUMBER OF PATIENTSATHY WAS A UNIVERSITY OF IOWA SPIN OUT COMPANY CALLED IDX, THEY DID THE FIRST PROSPECTIVE TRIAL AND WHAT'S GREAT IS THEY PUBLISHED THE DAILY BASIS THEA SO EVEN THOUGH THAT SYSTEM SUPPORT GOING TO BE WIDELY USED BECAUSE IT'S TOO EXPENSIVE IT'S 1 OF THE RARE EXAMPLES WE HAVE OF PROSPECTIVE TRIAL, YOU SEE ALL THE DATA IN THE COMMUNITY. >> AGREED AND EVEN, YOU KNOW I THINK FOR THE PATIENTS AS WELL, AND THERE WAS A SECOND PART TO THIS QUESTION THAT SORT OF TALKED ABOUT THE IDEA THAT THERE'S PATIENT ANXIETY OF FALSE-POSITIVES SO TESTS THAT MAY OR MAY NOT NEED TO HAPPEN AND HAVING THAT CONFIDENCE AROUND THESE APPROACHES WILL BE IMPORTANT TO BUILD THAT PATIENT ANXIETY OR LOWER THE PATIENT ANXIETY IN THIS PLACE. >> YEAH, JUDGE UTV A COMMENT ON THAT, YOU GO BACK TO THAT SMART WATCH DIAGNOSIS OF ATRIOLE FIBRILLATION, IT'S FOR THE RIGHT PEOPLE TO USE IT WHEN IT'S BASICALLY MARKED BY APPLE FOR EVERYBODY AND NOW WE HAVE, YOU KNOW 20 YEAR-OLDS THAT ARE GETTING ALERTS THAT THEY HAVE POTENTIAL ATRIOLE FIB ROUGH ATOM LIEWGZ, 99% OF THE TIME IT WILL BE WRONG SO IT INDUCES ANXIETY WHEN YOU USE ALGORITHMS IN THE WRONG PEOPLE. SO WE HAVE TO ACKNOWLEDGE THAT WHEN WE QUEP AI TOOLS, WE HAVE TO CAREFULLY DEFINE WHO ARE THE PEOPLE THAT SHOULD BE APPLIED FOR. >> YEAH, THAT'S A GREAT POINT. THE NEXT QUESTION'S KIND OF GOING TO GET AT MORE CLINICAL DATA AND IT SEEMS LIKE FOR EXAMPLE, ELECTRONIC HEALTH RECORDS OR THEY EXIST ALREADY BUT THERE ARE IN A KIND OF A DIFFICULT STATE THAT THEY AREN'T HARMONIZED VERY WELL, ARE THERE--THERE ARE REASONS TO THINK ABOUT USING AI, FOR ELECTRONIC HEALTH RECORD DATA THAT MIGHT BE HELPFUL FOR DIAGNOSTICS, MAYBE EVEN DRUG REPURPOSING OR OTHER TYPES OF CLUES THAT YOU MIGHT GET FROM USING ELECTRONIC HEALTH RECORD DATA, IS THAT A FAIR STATEMENT TO SAY AND YOU TALKED A BIT ABOUT THOSE--THE KEYBOARD NOTES, AND HOW IMPORTANT THOSE CAN BE AS WELL, SO INTEGRATING THOSE WITH THE STANDARD SORT OF MORE STRUCTURED ELECTRONIC HEALTH RECORD DATA, WILL THAT BE SOMETHING THAT YOU THINK WILL BE MOVING FORWARD IN THE FUTURE AS A WAY TO REALLY GLEAN DEEP INFORMATION? >> NO, I--NO QUESTION THE SOFTSPOT WE HAVE UNFORTUNATELY AS ELECTRONIC HEALTH RECORDS, THE QUALITY OF WHAT'S IN THEM, THEY WERE DONE OF COURSE, AS YOU ALL KNOW FOR BILLING PURPOSES SO THEY DON'T REALLY HAVE A LOT OF THE ESSENTIAL ELEMENTS AND A LOT OF THAT IS THERE BUT IT'S IN UNSTRUCTURED FORM AND WE DON'T HAVE THE BEST TOOLS TO TAKE UNSTRUCTURED TEXT AT A--HIGH DISCIPLINARY ENGZ MACHINESSAL ANALYTICS. NOW THAT'S WHY INTERESTINGLY PROBABLY WHY THE BABIES DO SO WELL BECAUSE THERE'S NOT A LOT OF UNSTRUCTURED TEXT TO DEAL WITH BUT FOR US ADULTS ESPECIALLY AS WE GET OLDER THERE'S MORE AND MORE OF THAT WE'RE ASKED TO DEAL WITH SO WE ARE SEATING BETTER TEXT PROCESSING AND THE PROJECTION IS OVER THE NEXT COUPLE YEARS THAT WILL GET--IT WON'T BE EVER AS LIKELY AS GOOD AS IMAGES BUT IT WILL BE GETTING, CLOSER AND CLOSER ALL THE TIME, CERTAINLY SPEECH HAS COME A LONG WAY, VOICE, SO, THAT WILL MAKE THE EHR MORE VALUABLE AND ALSO WHAT LOOKS REALLY ENCOURAGING IS THE SYNTHETIC NOTE VS NOR INFORMATIONA THAT IS STRUCTURED THAT WILL THEN BE MORE VALUABLE THAN WHAT WE TYPE IN KEYBOARD AND EPIC AND OTHER IT SYSTEMS. SO I'M HOPEFUL THAT WE SEE A LOT EHR AI PAPERS RIGHT NOW BUT THEY'RE ONLY SCRATCHING THE SURFACE OF WHERE THIS CAN GO BECAUSE THEY'RE BASICALLY JUST USING FOR THE MOST PART STRUCTURED ELEMENTS OF THE EHR. >> SO IT'S A START. >> IT'S A START. >> YEAH, DEFINITELY A START. >> AND THEN HOW DO YOU SEE THE ROOT TO INTEGRATE AI INTO SORT OF THE CLINICAL WORK FLOW? AND TO ALIGN SORT OF THE MAYBE MISALIGNED INCENTIVES FOR MANY OF THE CLINICIANS TO USE AI? >> WELL, I THINK IT'LL BECOME--YOU KNOW WE WON'T EVEN TALK ABOUT IT AT SOME POINT IF WE GO FAST FORWARD BECAUSE IT WILL JUST BE--SO MUCH OF WHAT WE RELY ON ONCE THESE ARE ANYTHING THROUGH THE NATURAL VALIDATION PROCESS BUT RIGHT NOW IF YOU ASK CLINICIANS WHAT THEY WANT THE MOST, YOU TELL THEM YOU NEVER HAVE TO TYPE ON A KEYBOARD AGAIN, THEY LIGHT UP, REALLY, BEAMING. [LAUGHTER] AND YOU CAN ACTUALLY JUST TALK TO THE PATIENT AND THE PATIENT BY THE WAY FEELS SIMILARLY. SO THAT'S GOING TO BE 1 OF THE QUICK HITS. AND THEN IN THE UK THEY ALREADY HAVE VARIOUS, EMERGENCY ROOMS TO DOING THAT, AN EMERGENCY ROOM WHERE THERE'S NO KEYBOARD THAT SAYS SOMETHING RIGHT THERE. SO, YOU KNOW I WAS ON A CALL WITH THEM EARLY THIS MORNING AND THEY WANT TO BE THE AI LEADER AS THEY HAVE BEEN WITH GENOMICS AND THEY'RE MAKING REALLY DEFINITIVE STEPS TO DO THAT, WE'RE NOT AS ORGANIZED WE DON'T HAVE A UNIVERSAL HEALTHCARE NATIONAL STRATEGY BUT WE SHOULD--1 OF THE PROBLEMS WE HAVE IS EACH INDIVIDUAL, EACH PATIENT HAS SO MANY DIFFERENT EHRs THEY GO TO DIFFERENT PROVIDERS AND TO GET ALL THAT DATA, TO GET IT ALL AGGREGATED AND SYNTHESIZED AND CRYSTALLIZE, THAT'S ANOTHER WELCOME FEATURE FOR ANY CLINICIAN, BECAUSE IT'S SO HARD TO GO THROUGH ALL THOSE SCREEN SHOTS TO FIND OUT WHAT'S RELEVANT. SO THIS IS ANOTHER THING THAT WILL HELP IN THE EARLY GOING, BECAUSE THIS IS A PERFECT THING FOR MACHINES AND IT WILL JUST MAKE LIVES EASIER FOR CLINICIANS AND BETTER FOR THE PATIENT. >> THAT'S GREAT. IT'S BEEN REALLY IMPRESSIVE TO WATCH HOW THE UK ALSO HAS HANDLED THE COVID CRISIS AND THE AMOUNT OF DATA COMING FROM THEM, FROM A LOT OF THE CLINICAL TRIALS AND WHAT THEY'VE BEEN DOING SO I THINK WE CAN CONTINUE TO LEARN A LOT FROM EACH OTHER OVER THIS TIME. THIS NEXT QUESTION IS A BIT LONG SO I'M GOING TO TRY TO READ IT AND GET THE CAPTURE THE ESSENCE OF IT WE ONLY HAVE 3 MINUTES LEFT SO THIS MAY BE THE END OF THE QUESTIONS BUT I WILL ASK IT, IN TERMS OF FOR WHEN YOU THINK ABOUT NEURAL NETWORKS AND THEY DON'T WORK LIKE THE BRAIN DOES SO FOR REPETITIVE TASKS WHERE THE SOLUTION IS KNOWN BUT SOMETIMES DIFFICULT FOR A HUMAN TO FIND IN A LIMITED AMOUNT OF TIME ARTIFICIAL INTELLIGENCE IS CLEARLY AN OPTION BUT THE SAME TAME TIME THE NUMBER OF PROFESSIONALS TRUSTING IN THE MACHINE WILL INCREASE AND THE CRITICALLY MINDED PROFESSIONALS TO DISCOVER THE UNKNOWN ARE ESSENTIAL. AND SO THINKING ABOUT THOSE 2 GROUPS OF PEOPLE AND THE FACT THAT SCIENCE IS NOT STATIC, IT EVOLVES OVER TIME AND SO, WITH THE CREATIVITY OF THE HUMAN BRAIN, AND NOT THE EFFICIENCY OF THE MACHINES, HOW DO YOU SEE THE COLLATERAL EFFECT OF THE WIDE SPREAD USE OF AI FOR SCIENTIFIC ADVANCEMENT. >> WELL, THEY'RE REALLY COMPLIMENTARY, RIGHT? THEY'RE YOU'RE NOT GOING IT FIND ANY AI ALGORITHM THAT'S CREATED. THIS IS WHERE WE SHINE. AND THE COMPLEMENTARITY OF THE FUSION HERE IS PRETTY EXTRAORDINARY. THE ARE THE THING IS THE AUTODIDACTIC ASPECT OF YOU HAVE THIS NOW, THESE ARE SUPERVISED LEARNING, DEEP LEARNING BUT MORE AND MORE WE SEE SELF-SUPERVISED EVEN TO THE POINT OF UNSUPERVISED AND THEN ON THE OTHER HAND YOU'VE GOT THE--WHO'S BUILDING THESE NETWORKS? RIGHT? >> RIGHT. >> AND WHAT ARE THE PRIORITIES? I PLEEN WE HAVE ONLY LIMITED AI TALENT ESPECIALLY IN THE MEDICAL SPACE. AND WHAT ARE WE GOING TO PUT OUR RESOURCES TO? AND WHAT--WHERE'S OUR CREATIVITY TO GUIDE THIS AND TO MAKE SURE IT'S MOVING IN THE RIGHT DIRECTION BECAUSE WE KNOW IT HAS SERIOUS LIABILITIES THAT COULD ACTUALLY MAKE THINGS WORSE. I KNOW THAT SOUNDS UNLIKELY BUT IT'S POSSIBLE. SO I THINK THIS IS ACTUALLY, YOU KNOW, WHEN I PUT TOGETHER DEEP MEDICINE, I HAD A WHOLE TABLE OF THE HUMAN QUALITIES LIKE LAUGHTER AND CREATIVITY AND LOVE AND ALL THESE THINGS THAT THAT'S WHAT WE WANT TO EXUDE AND I THINK THAT THEY ARE REMARKABLILY COMPLIMENTARY AND THAT'S WHY OVER TIME, WE WILL RELY MORE ON MACHINES BUT WE NEVER WANT TO LET GO, THAT IS WE NEED TO HAVE THE HUMAN OVERSIGHT AND THE--THIS HAS YOU SAY COLLATERAL, THIS SYMBIOTIC RELATIONSHIP. >> YEAH AND THAT'S I THINK THAT'S SCRUOF THE A PERFECT NOTE TO LEAVE THIS ON. WE NEED EVERYBODY HERE AT THE TABLE TO MAKE IN HAPPEN AND I THINK THAT'S A GREAT MESSAGE. DR. TOPOL I WANT TO THANK YOU SO MUCH FOR THIS WONDERFUL PRESENTATION, AND I WANT TO THANK DR. COLLINS WHO WAS UNFORTUNATELY CALLED AWAY SO SHE WAS ABLE TO GO THROUGH WITH THE QUESTION AND ANSWER PERIOD SO I WAS HAPPY TO STEP IN AND DO SO BUT I WANT TO THANK YOU FOR THIS PRESENTATION AND ALL OF THE SORT OF PROVOCATIVE THINGS YOU'VE SAID TODAY THAT WE CAN CARRY FORWARD AND INSURE THAT WE ARE ON THE PATH FOR AI IN MEDICINE AND IN VERY POSITIVE WAYS. THANK YOU AGAIN AND I THINK WITH THIS I WILL SIGN US OFF.