Emotionalintelligence

Ouremotionsinfluenceeveryaspectofourlives,fromourhealthandhowwelearn,tohowwedobusinessandmakedecisions,bigonesandsmall.Ouremotionsalsoinfluencehowweconnectwithoneanother.We'veevolvedtoliveinaworldlikethis,butinstead,we'relivingmoreandmoreofourliveslikethis--thisisthetextmessagefrommydaughterlastnight--inaworldthat'sdevoidofemotion.SoI'monamissiontochangethat.Iwanttobringemotionsbackintoourdigitalexperiences.0:47Istartedonthispath15yearsago.IwasacomputerscientistinEgypt,andIhadjustgottenacceptedtoaPh.D.programatCambridgeUniversity.SoIdidsomethingquiteunusualforayoungnewlywedMuslimEgyptianwife:Withthesupportofmyhusband,whohadtostayinEgypt,IpackedmybagsandImovedtoEngland.AtCambridge,thousandsofmilesawayfromhome,IrealizedIwasspendingmorehourswithmylaptopthanIdidwithanyotherhuman.Yetdespitethisintimacy,mylaptophadabsolutelynoideahowIwasfeeling.IthadnoideaifIwashappy,havingabadday,orstressed,confused,andsothatgotfrustrating.Evenworse,asIcommunicatedonlinewithmyfamilybackhome,Ifeltthatallmyemotionsdisappearedincyberspace.Iwashomesick,Iwaslonely,andonsomedaysIwasactuallycrying,butallIhadtocommunicatetheseemotionswasthis.(Laughter)Today'stechnologyhaslotsofI.Q.,butnoE.Q.;lotsofcognitiveintelligence,butnoemotionalintelligence.Sothatgotmethinking,whatifourtechnologycouldsenseouremotions?Whatifourdevicescouldsensehowwefeltandreactedaccordingly,justthewayanemotionallyintelligentfriendwould?Thosequestionsledmeandmyteamtocreatetechnologiesthatcanreadandrespondtoouremotions,andourstartingpointwasthehumanface.2:29Soourhumanfacehappenstobeoneofthemostpowerfulchannelsthatweallusetocommunicatesocialandemotionalstates,everythingfromenjoyment,surprise,empathyandcuriosity.Inemotionscience,wecalleachfacialmusclemovementanactionunit.Soforexample,actionunit12,it'snotaHollywoodblockbuster,itisactuallyalipcornerpull,whichisthemaincomponentofasmile.Tryiteverybody.Let'sgetsomesmilesgoingon.Anotherexampleisactionunit4.It'sthebrowfurrow.It'swhenyoudrawyoureyebrowstogetherandyoucreateallthesetexturesandwrinkles.Wedon'tlikethem,butit'sastrongindicatorofanegativeemotion.Sowehaveabout45oftheseactionunits,andtheycombinetoexpresshundredsofemotions.3:17Teachingacomputertoreadthesefacialemotionsishard,becausetheseactionunits,theycanbefast,they'resubtle,andtheycombineinmanydifferentways.Sotake,forexample,thesmileandthesmirk.Theylooksomewhatsimilar,buttheymeanverydifferentthings.(Laughter)Sothesmileispositive,asmirkisoftennegative.Sometimesasmirkcanmakeyoubecomefamous.Butseriously,it'simportantforacomputertobeabletotellthedifferencebetweenthetwoexpressions.3:49Sohowdowedothat?Wegiveouralgorithmstensofthousandsofexamplesofpeopleweknowtobesmiling,fromdifferentethnicities,ages,genders,andwedothesameforsmirks.Andthen,usingdeeplearning,thealgorithmlooksforallthesetexturesandwrinklesandshapechangesonourface,andbasicallylearnsthatallsmileshavecommoncharacteristics,allsmirkshavesubtlydifferentcharacteristics.Andthenexttimeitseesanewface,itessentiallylearnsthatthisfacehasthesamecharacteristicsofasmile,anditsays,Aha,Irecognizethis.Thisisasmileexpression.4:29Sothebestwaytodemonstratehowthistechnologyworksistotryalivedemo,soIneedavolunteer,preferablysomebodywithaface.(Laughter)Cloe'sgoingtobeourvolunteertoday.4:44Sooverthepastfiveyears,we'vemovedfrombeingaresearchprojectatMITtoacompany,wheremyteamhasworkedreallyhardtomakethistechnologywork,asweliketosay,inthewild.Andwe'vealsoshrunkitsothatthecoreemotionengineworksonanymobiledevicewithacamera,likethisiPad.Solet'sgivethisatry.5:05Asyoucansee,thealgorithmhasessentiallyfoundCloe'sface,soit'sthiswhiteboundingbox,andit'strackingthemainfeaturepointsonherface,sohereyebrows,hereyes,hermouthandhernose.Thequestionis,canitrecognizeherexpression?Sowe'regoingtotestthemachine.Sofirstofall,givemeyourpokerface.Yep,awesome.(Laughter)Andthenasshesmiles,thisisagenuinesmile,it'sgreat.Soyoucanseethegreenbargoupasshesmiles.Nowthatwasabigsmile.Canyoutryasubtlesmiletoseeifthecomputercanrecognize?Itdoesrecognizesubtlesmilesaswell.We'veworkedreallyhardtomakethathappen.Andtheneyebrowraised,indicatorofsurprise.Browfurrow,whichisanindicatorofconfusion.Frown.Yes,perfect.Sotheseareallthedifferentactionunits.There'smanymoreofthem.Thisisjustaslimmed-downdemo.Butwecalleachreadinganemotiondatapoint,andthentheycanfiretogethertoportraydifferentemotions.Soontherightsideofthedemo--looklikeyou'rehappy.Sothat'sjoy.Joyfiresup.Andthengivemeadisgustface.TrytorememberwhatitwaslikewhenZaynleftOneDirection.(Laughter)Yeah,wrinkleyournose.Awesome.Andthevalenceisactuallyquitenegative,soyoumusthavebeenabigfan.Sovalenceishowpositiveornegativeanexperienceis,andengagementishowexpressivesheisaswell.SoimagineifCloehadaccesstothisreal-timeemotionstream,andshecouldshareitwithanybodyshewantedto.Thankyou.(Applause)6:44So,sofar,wehaveamassed12billionoftheseemotiondatapoints.It'sthelargestemotiondatabaseintheworld.We'vecollecteditfrom2.9millionfacevideos,peoplewhohaveagreedtosharetheiremotionswithus,andfrom75countriesaroundtheworld.It'sgrowingeveryday.Itblowsmymindawaytha