ART I F I CI AL I NT E L L I GE NCE -- the wa y ma c hine thinks Sta nle y L ia ng , PhD Ca ndida te , L a sso nde Sc ho o l o f E ng ine e ring , Yo rk Unive rsity He lix Sc ie nc e E ng a g e me nt Pro g ra ms 2018
2 AGE OF AI , AGE OF MACHI NE • AI b ring s a lo t o f pa ssio n in the 21 c e ntury • AI po se s o ne o f the g re a te st thre a ts to huma n • I t is the pa th to so me o f the b e st o ppo rtunitie s • We a re a t the ve ry b e g inning o f AI • Re se a rc h: pre dic tio n a b o ut we a the r pa tte rns, pha rma c e utic a ls, me dic a l tre a tme nts • Busine ss: pre dic t c usto me r re q ue st a nd b e ha vio r, virtua l a ssista nt, a uto ma te d driving Sta nle y L ia ng , L a sso nde Sc ho o l o f E ng ine e ring , Yo rk Unive rsity, 2018
3 WHAT I S ART I F I CI AL I NT E L L I GE NCE • T he a b ility to le a rn a nd so lve pro b le ms • Artific ia l inte llig e nc e is inte llig e nc e de mo nstra te d b y ma c hine s, in c o ntra st to the na tura l inte llig e nc e (NI ) displa ye d b y huma ns a nd o the r a nima ls • T he sc ie nc e a nd e ng ine e ring o f ma king inte llig e nc e ma c hine s Jo hn Mc Ca rthy • Just a s the I ndustria l Re vo lutio n fre e huma nity fro m physic a l drudg e ry, AI ha s the po te ntia l to fre e huma nity fro m me nta l drudg e ry Andre w Ng Sta nle y L ia ng , L a sso nde Sc ho o l o f E ng ine e ring , Yo rk Unive rsity
4 A L I T T L E BI T HI ST ORY OF AI • Ge ne ra l pro b le m so lve r (1956) b y Alle n Ne we ll & He rb e rt Simo n • T o so lve a ny pro b le m tha t c a n b y pre se nte d b y ma th • Physic a l symb o l syste m Hypo the sis – symb o ls a re the ke y o f inte llig e nc e , the wa y ho w yo u inte ra c t with the wo rld • Huma n re a so ning is simply c o nne c ting symb o ls • I f ma c hine c a n b e tra ine d to unde rsta nd symb o ls, the y c o uld b e ha ve like huma n Sta nle y L ia ng , L a sso nde Sc ho o l o f E ng ine e ring , Yo rk Unive rsity
5 JOHN SE ARL E AND CHI NE SE ROOM ARGUME NT • A ma n no t kno wing Chine se is lo c ke d in a ro o m with a la rg e b o o k with a lo t o f Chine se pa tte rn • A na tive Chine se spe a ke r puts the Chine se phra se a s se q ue nc e o f c ha ra c te rs thro ug h the ma iling slo t into the ro o m • T he ma n inside ma tc he s the pa tte rns with his b o o k a nd put the c o rre spo nding Chine se pa tte rns to o utput the m a s se q ue nc e o f Chine se c ha ra c te rs • T he na tive Chine se spe a ke r o utside s will think he is ta lking with a na tive spe a ke r • I n fa c t, the ma n inside c a nno t unde rsta nd Chine se a t a ll. • Simply ma tc hing symb o ls is no t a true AI • I f the se q ue nc e is to o lo ng , the ma c hine c a nno t a ffo rd the ma tc hing s – c o mb ina to ria l e xplo sio n Sta nle y L ia ng , L a sso nde Sc ho o l o f E ng ine e ring , Yo rk Unive rsity
6 ART I F I CI AL I NT E L L I GE NCE - DE F I NI T I ON • I n 1950, Jo hn Mc Ca rthy he ld the first wo rksho p o n AI • Artific ia l inte llig e nc e a s a te c hno lo g y is a ny syste m tha t e xhib its b e ha vio r tha t c o uld b e inte rpre te d a s huma n inte llig e nc e • AI ’ s supe r po we r • I BM De e p Blue vs. Ga rry K a spa ro v • Go o g le De e pMind vs. Se do l L e e • I n fa c t, the c o mpute rs ha ve no ide a o f the se g a me s • L e a rn the rule s b y suffic ie nt tra ining • Do pa tte rn ma tc hing • Huma n a nd ma c hine pe rfo rms the ir inte llig e nc e in diffe re nt wa ys: • Co mpute r pro c e sse s fa ste r a nd ide ntify a nd ma tc h mo re pa tte rns Sta nle y L ia ng , L a sso nde Sc ho o l o f E ng ine e ring , Yo rk Unive rsity
7 ST RONG AI VS WE AK AI • Stro ng AI : Ma c hine displa ys a ll • We a k AI : c o nfine d to a ve ry na rro w huma n-like b e ha vio r ta sk • K no wn a s g e ne ra l inte llig e nc e • K no wn a s na rro w AI • A b ro a d AI c o ve rs a wide ra ng e o f • Mimic huma n b a se d o n the ir ta sk pro g ra mming • Ma tc he s g e ne ra l-le ve l inte llig e nc e • Pro g ra mme d ma c hine s a c ting a s if (me nta l sta te s, c o nsc io usne ss, e tc .) the y we re inte llig e nt – All c urre nt AI ’ s a re we a k AI • Po ssib le g ive n inc re a sing ha rdwa re a nd so ftwa re a dva nc e s • Ma c hine le a rning is the ma in pa th fo r we a k AI Sta nle y L ia ng , L a sso nde Sc ho o l o f E ng ine e ring , Yo rk Unive rsity
8 T HE E VOL UT I ON OF AI • AI sta rts with the symb o lic a ppro a c h like the Chine se Ro o m Pro b le m • E xpe rt Syste m: a n e a rly fo rm o f AI : dia g no sis, c re dit c he c k, fill yo ur T a x re turn – lo ng list o f ma tc hing pa tte rns • c o mb ina to ria l e xplo sio n due to infinite pa tte rns e nds in 1980s • Pla nning AI : sho rte n the lo ng list o f pa tte rn b y he uristic re a so ning : limit the se a rc h sc o pe • T he symb o lic syste ms a nd pla nning AI c a n tra c e b a c k to the symb o lic syste m ide a s in the 1950s Sta nle y L ia ng , L a sso nde Sc ho o l o f E ng ine e ring , Yo rk Unive rrsity
9 COMMON APPL I CAT I ONS OF AI • Ro b o tic s: le t ma c hine wo rk o n physic a l ta sks • Use d to c re a te hig hly spe c ia lize d ma c hine s • L imite d in the wo rk to b e a c c o mplishe d • Be st fo r re pe titive wo rk • AI + Ro b o tic s: wide n the sc o pe o f wo rk fo r the ro b o ts • Ro o mb a + AI : le a rn the ma p o f the ro o ms b e fo re wo rking – a b ig da ta pro b le m • Na tura l la ng ua g e pro c e ssing : inte ra c t with ma c hine using huma n la ng ua g e • Go o g le se a rc h, spa m de te c ting • ne e ds a la rg e la ng ua g e c o rpus • Unde rsta nding c o nte xt me a ning Sta nle y L ia ng , L a sso nde Sc ho o l o f E ng ine e ring , Yo rk Unive rsity
10 COMMON APPL I CAT I ONS OF AI • T he I nte rne t o f T hing s (I o T ): the ne two rk o f physic a l de vic e s, ve hic le s, ho me a pplia nc e s a nd o the r ite ms e mb e dde d with e le c tro nic s, so ftwa re , se nso rs, a c tua to rs, a nd c o nne c tivity whic h e na b le s the se o b je c ts to c o nne c t a nd e xc ha ng e da ta – g e ne ra te s b ig da ta • Big Da ta & Da ta Sc ie nc e • Big Da ta : re pre se nts the info rma tio n a sse ts c ha ra c te rize d b y suc h a hig h volume , ve loc ity a nd ie ty to re q uire spe c ific te c hno lo g y a nd var a na lytic a l me tho ds fo r its tra nsfo rma tio n into va lue • Da ta Sc ie nc e : a n inte rdisc iplina ry fie ld fo c us o n e xtra c ting kno wle dg e o r insig hts fro m da ta in va rio us fo rms. Sta nle y L ia ng , L a sso nde Sc ho o l o f E ng ine e ring , Yo rk Unive rrsity
Recommend
More recommend