While plenty of sci­en­tists are work­ing on hu­manoid ro­bots, Hiroshi Ishig­uro ac­tu­ally wants to build a hu­man. EL­IZ­A­BETH FINKEL re­ports.

Cosmos - - Front Page - EL­IZ­A­BETH FINKEL is the ed­i­tor-in-chief of Cosmos.

HIROSHI ISHIG­URO, the di­rec­tor of the In­tel­li­gent Ro­bot­ics Lab­o­ra­tory (IRL) at Osaka Uni­ver­sity, is well-known for pos­ing with his an­droid twin. It’s not just a weird pub­lic­ity stunt; this might be the an­swer to Ja­pan’s labour cri­sis. With its grey­ing pop­u­la­tion – close to 28% of its 127 mil­lion peo­ple are aged over 65 – be­low-re­place­ment birth rate and re­luc­tance to ramp up im­mi­gra­tion, Ja­pan needs to make its own work­ers.

It al­ready has plenty of in­dus­trial ro­bots. But who will tend to the el­derly in over­flow­ing nurs­ing homes and, per­haps just as im­por­tant, who will make them feel cared for? That’s why Ishig­uro’s lab has gov­ern­ment fund­ing to cre­ate ever-more hu­man-like ro­bots – in­deed ,with US$5 mil­lion ev­ery year for five years, the project to cre­ate his au­ton­o­mous hu­manoid, Erica, re­ceives the largest grant from Ja­pan’s Science and Tech­nol­ogy Agency (JST).

Yet Ishig­uro him­self is a sur­prise. He doesn’t fit the stereo­type of a roboti­cist, some­one more in tune with ma­chines than peo­ple. His first am­bi­tion was to be­come an oil painter, and he re­tains the artist’s ba­sic im­pulse – to ex­am­ine the hu­man con­di­tion. Asked what drives his mis­sion to build hu­manoid ro­bots, he replies: “I want to un­der­stand what it is to be a hu­man be­ing.”

As ar­ti­fi­cial in­tel­li­gence con­tin­ues to de­velop “we will have to ask that ques­tion more and more”, agrees en­gi­neer El­iz­a­beth Croft, who spe­cialises in hu­man­robot in­ter­ac­tion at the Uni­ver­sity of Bri­tish Columbia.

Oth­ers find Ishig­uro’s work puz­zling. “I don’t un­der­stand his sci­en­tific con­cept ex­actly,” says Alin Albu-schäf­fer, di­rec­tor of the In­sti­tute of Ro­bot­ics and Mecha­tron­ics at DLR, the Ger­man Aerospace Cen­tre, but he adds: “I like it from a philo­soph­i­cal per­spec­tive. He’s at the ex­treme, and that pro­vokes change.”

Ishig­uro’s work lies some­where be­tween the prac­ti­cal and the weird. Plenty of places build hu­manoid ro­bots but they are clearly me­chan­i­cal rep­re­sen­ta­tions of hu­man-ness.

Ishig­uro is ac­tu­ally try­ing to build a hu­man. For him it is a way to tackle the mys­ter­ies of the hu­man mind: in­tel­li­gence and con­scious­ness. “We can’t take an an­a­lyt­i­cal ap­proach to find out what a hu­man is,” he says. “We need to take a con­struc­tive ap­proach.”

ISHIG­URO, NOW 54, SWITCHED from paint­ing to pro­gram­ming at uni­ver­sity and was soon drawn to ro­bot­ics. “I saw that AI needs to have a body,” he tells me at a con­fer­ence in Mel­bourne, “be­cause a com­puter needs to have its own ex­pe­ri­ences.”

While AI has pro­gressed in leaps and bounds in re­cent years, it is still enor­mously chal­leng­ing to cre­ate ro­bots that can ma­noeu­vre them­selves in our messy ever-chang­ing world as op­posed to the uni­form con­di­tions of a fac­tory floor. The Google-built Al­phago soft­ware can beat the world Go cham­pion but ro­bots don’t stand a chance at beat­ing a team of kids in a game of foot­ball.

Ro­bot­ics com­pa­nies ev­ery­where are grap­pling with the chal­lenge. DLR has Justin, who is handy with tools. Honda has Asimo, who can serve drinks. Re­think Ro­bot­ics has Bax­ter, who can pass things to a co-worker and whose flat-screen eyes show where its at­ten­tion is. Bos­ton Dy­nam­ics has At­las, whose lat­est trick is back­flips. No one, though, could mis­take these ro­bots for a hu­man. “They are much more R2-D2 than C-3PO,” Croft says.

Most robot mak­ers de­lib­er­ately keep their cre­ations robot-like. This re­flects two guid­ing prin­ci­ples.

One is to steer well clear of the ‘un­canny val­ley’ – the creepy feel­ing when you see al­most-but-not-quite hu­man char­ac­ters in com­puter games or an­i­ma­tions. The other, Albu-schäf­fer says, is that the large gap in robot vs hu­man in­tel­li­gence and au­ton­omy should be re­flected in the de­sign – “the ap­pear­ance should re­flect the robot’s stage of evo­lu­tion”.

Ishig­uro has headed in the op­po­site di­rec­tion, plung­ing head­long into the un­canny val­ley.

His Gemi­noid se­ries of ro­bots are his trade­mark. The first, made in 2002, was a twin of his five-year-old daugh­ter. Repliee Q1 (2005) was the twin of a Tokyo news­reader. Gemi­noid H1 (2006) was Ishig­uro’s twin. Gemi­noid F (2010) was mod­elled on a woman in her twen­ties (whose iden­tity Ishig­uro won’t di­vulge).

The idea be­hind mak­ing a copy of a real hu­man, Ishig­uro says, was to trans­fer the pres­ence, the son­zai-kan, of that per­son to the robot. “I fo­cused on hu­man like­ness be­cause that’s an ex­treme goal of ro­bot­ics,” he tells me. “In a first con­tact, peo­ple will be sur­prised, but it’s easy to adapt.”

These hy­per­real repli­cas have em­ployed the lat­est that sil­i­cone tech­nol­ogy and mus­cle-like fine-mo­tor cir­cuitry (ac­tu­a­tors) can of­fer. But they are less ro­bots than pup­pets, their speech and move­ments con­trolled by some­one sit­ting at a key­board.

One of Ishig­uro’s key goals is for the hu­manoids to con­vey emo­tion. “When we feel emo­tion that’s when we be­gin to make a con­nec­tion,” he says, “and we for­get about the sta­tus of the part­ner.”

To im­part ex­pres­sive­ness to the ro­bots, Ishig­uro turned to a master of the art – play­wright and di­rec­tor Oriza Hi­rata, a cham­pion of re­al­ism (or ‘quiet drama’) in Ja­pa­nese theatre. With mo­tion de­tec­tors at­tached to his face, Hi­rata mod­elled the ges­tures Ishig­uro wanted his hu­manoids to ex­press.

The col­lab­o­ra­tion led to the Robot Theatre Project, which has staged plays around the world. In these per­for­mances com­puter-con­trolled ro­bots fill in for hu­man ac­tors, de­liv­er­ing pre-recorded lines and chore­ographed move­ments.

The com­pany’s reper­toire in­cludes Say­onara, a play writ­ten by Hi­rata where an an­droid (played by Gemi­noid F) tries to con­sole a girl suf­fer­ing from a fa­tal ill­ness un­til its own me­chan­ics go awry. In I, Worker a robot maid loses its mo­ti­va­tion to work. The dou­ble bill toured North Amer­ica in 2013. The robot theatre has also per­formed Franz Kafka’s Me­ta­mor­pho­sis and

An­ton Chekhov’s Three Sis­ters. A planned per­for­mance of Jean Paul Sartre’s No Exit for a ma­jor French arts fes­ti­val in 2015 was can­celled af­ter Sartre’s es­tate re­fused per­mis­sion for robot ac­tors.

Hi­rata has pro­vided the emo­tional X-fac­tor to many of Ishig­uro’s cre­ations. “We call it the Oriza fil­ter,” says the roboti­cist. It’s a cod­i­fied and pro­gram­mable pat­tern based on the di­rec­tor’s ut­ter­ances and ex­pres­sions: a move­ment of the body and hands, then the eyes, then the head, then an ut­ter­ance af­ter a 0.2 sec­ond de­lay. “If we ap­ply the Oriza fil­ter,” Ishig­uro says, “our ro­bots be­come so hu­man-like.”

This chore­og­ra­phy of con­ver­sa­tion is very con­sis­tent be­tween peo­ple, he says – so much so that he de­scribes a patent based on Oriza’s move­ments as “how to rep­re­sent hu­man like­ness”. BUT WHILE SOME OF Ishig­uro’s hu­manoids grow ever more ex­pres­sive and hu­man, oth­ers have de­vel­oped in the op­po­site di­rec­tion.

I shriek in hor­ror when Ishig­uro shows me the hu­manoid he has de­vel­oped for el­derly peo­ple with de­men­tia. It re­sem­bles a thalido­mide child with half arms end­ing in nubs and a torso with­out legs. “It’s a bit creepy,” Ishig­uro ad­mits, “but this works very well.” These ‘te­lenoids’ have been used in more than 70 hos­pi­tals in Ja­pan, he says, as well as in Den­mark, Ger­many and Aus­tria.

Ishig­uro shows me a movie clip of an el­derly Ja­pa­nese lady hug­ging a te­lenoid and chat­ting to it as she might with a favourite grand­child. By be­ing so stripped down, gen­der­less and age­less, “de­mented peo­ple can use their own imag­i­na­tion; they don’t feel any pres­sure,” he ex­plains. For sim­i­lar rea­sons, he says, the te­lenoids have also worked very well for chil­dren with autism.

A more diminu­tive vari­a­tion is the Hugvie – a soft, hug­gable robot you can put a phone into. “It al­lows you to feel the pres­ence of a per­son while you are talk­ing [to them],” Ishig­uro says. He shows me an­other video, of a room of noisy kinder­garten kids who im­me­di­ately quiet down when their Hugvies start talk­ing to them.

No doubt the abil­ity of these stripped-down hu­manoids to ful­fil ba­sic emo­tional needs also says some­thing about what it means to be hu­man. “I THINK ERICA IS the most beau­ti­ful and most hu­man­like au­ton­o­mous an­droid in the world … I hope.” This is how Ishig­uro de­scribes Erica in a video pro­duced by The Guardian last April.

To me, Erica is dis­con­cert­ing. It’s not that her pearly sil­i­cone skin and fea­tures are all that life-like; or that when she speaks her lips move up and down in a doll-like way. But when Eti­enne, a vis­i­tor to Ishig­uro’s lab in Osaka, talks to her, things get un­canny.

Erica turns her head to­wards Eti­enne, her eyes fo­cus­ing on his. “Hello there,” she says. “May I ask your name?” Eti­enne, he tells her. “It’s nice to meet you, Eti­enne,” she re­sponds. “So,” – she nods and pauses – “what coun­try are you from?” South Africa, Eti­enne tells her. “Oh re­ally,” she ex­claims, shrug­ging her shoul­ders. “I’ve never been to South Africa but I did love the film Chap­pie, which was made in South Africa. I think it raises some ques­tions about ar­ti­fi­cial con­scious­ness, and Chap­pie is very cute.”

Erica’s abil­ity to track Eti­enne dur­ing the con­ver­sa­tion comes cour­tesy of two in-built 16-chan­nel mi­cro­phone ar­rays, 14 in­frared depth sen­sors and the abil­ity to move her head 20 de­grees. She can­not move her arms or legs – yet. Her ex­pres­sive ges­tures – blink­ing, shoul­der shrugs, head turn­ing and an up­ward look with her eyes at pen­sive mo­ments – have clearly been run through the ‘Oriza fil­ter’. She also has fa­cial­recog­ni­tion ca­pa­bil­ity and mem­ory, so she knows when she has spo­ken with some­one be­fore, and can re­fer to past con­ver­sa­tions.

But is this ev­i­dence for the work­ings of a mind? Her ar­chi­tect, Dy­lan Glas, sug­gests it is: “For about two years now I’ve been work­ing with Erica to cre­ate her mind, her per­son­al­ity and get all the de­tails work­ing.”

This is where we get into fuzzy ter­ri­tory. No one knows how to cre­ate a hu­man mind. Its fun­da­men­tals – con­scious­ness and in­tel­li­gence – elude even def­i­ni­tion, let alone repli­ca­tion. “No­body can de­fine hu­man in­tel­li­gence,” Ishig­uro tells me em­phat­i­cally. “That is one of our fi­nal goals, to un­der­stand what hu­man in­tel­li­gence is.” He is equally adamant that no one is close to cre­at­ing a hu­man-like ar­ti­fi­cial in­tel­li­gence.

He de­scribes the likes of Al­phago as hav­ing “in­sectlevel in­tel­li­gence”. Ma­chine-learn­ing al­go­rithms learn

This is where we get into fuzzy ter­ri­tory. No one knows how to cre­ate a hu­man mind. Its fun­da­men­tals – con­scious­ness and in­tel­li­gence – elude even def­i­ni­tion, let alone repli­ca­tion.

win­ning pat­terns from vast data sets. Al­phago, for in­stance, learned from 30 mil­lion moves by grand mas­ters, and then from mil­lions more by play­ing against it­self. Ishig­uro is unim­pressed: “A hu­man never does that; if we did we would get old and pass away be­fore learn­ing any­thing.”

The abil­ity to learn pat­terns from data sets means AIS can recog­nise voices, faces and key words, and re­spond with ma­te­rial in their mem­ory. Like Siri, Erica recog­nises key words, finds matches in her mem­ory and an­swers with pro­grammed re­sponses. Erica is also good at fak­ing. She can keep the con­ver­sa­tion go­ing even when it goes off-script. “Re­spond, ac­knowl­edge, pivot; it’s the same trick I oc­ca­sion­ally used with talk­ing to my grandma,” quips Croft.

But there is some­thing more to Erica – the be­gin­nings of some­thing that is dis­tinctly hu­man. “Erica has sim­ple in­ten­tions and de­sires that con­trol the be­hav­iour,” Ishig­uro says. “That is the main dif­fer­ence to Siri.”

In­ten­tions and de­sires! It sounds scary – surely the first step to­wards ro­bots tak­ing over the world. But many roboti­cists think it is a nec­es­sary next step. “If we want ro­bots to serve hu­mans in the home,” says Toby Walsh, an AI ex­pert at the Uni­ver­sity of NSW, “we will need them to have in­ten­tions and de­sires.”

Con­sider load­ing a dish­washer. Step-by-step in­struc­tions won’t cut it, ex­plains Albu-schäf­fer. A robot needs to recog­nise all kinds of ob­jects un­der dif­fer­ent light­ing in dif­fer­ent kitchens, re­trieve them from odd po­si­tions, open a dish­washer door and fi­nally stack dishes in an ef­fec­tive fash­ion: “We can’t de­scribe this at the level of equa­tions; this kind of plan­ning and knowl­edge of en­vi­ron­ment is some­thing we as­sim­i­late through­out our lives.” It is some­thing, Albu-schäf­fer jokes, his 18-year-old son has yet to master.

In ro­bot­ics-speak “in­ten­tion and de­sire” is what ro­bots need to carry out such mis­sions. From in­ten­tion and de­sire come rea­son­ing, plan­ning and ac­tion. “Peo­ple think giv­ing ro­bots in­ten­tions and de­sires means they will take over the world,” says Al­buSchäf­fer. “We just want them to load the dish­washer.”

So what sort of in­ten­tions and de­sires does Erica have? “In her cur­rent im­ple­men­ta­tion she wants to talk, she wants to be well-recog­nised and she wants to take a rest,” Ishig­uro says.

And Erica’s mind? Ishig­uro says it is more in the mind of the be­holder. He ac­knowl­edges that sci­en­tif­i­cally, “no”, she does not have a mind, but “for vis­i­tors, she does”. It is the son­zai-kan cre­ated by her beau­ti­ful sil­i­cone face, Hi­rata’s the­atri­cal moves and the au­ton­o­mous con­ver­sa­tion.

“This is the Tur­ing test af­ter all,” says Walsh, re­fer­ring to com­put­ing pi­o­neer Alan Tur­ing’s pro­posal that the true test of ar­ti­fi­cial in­tel­li­gence is to pass for a hu­man in con­ver­sa­tion. Tur­ing en­vis­aged only textbased di­a­logue; Ishig­uro’s hu­manoids use their bod­ies to en­hance the il­lu­sion. “We’re be­ing fooled by ma­chines that have al­most no in­tel­li­gence,” Walsh notes. JA­PA­NESE CUL­TURE IS FASCINATED by ro­bots. “Un­like North Amer­i­cans,” Croft says, “the Ja­pa­nese don’t seem to have the same prob­lem with the un­canny val­ley.” Com­men­ta­tors of­ten point to Shinto to ex­plain Ja­pan’s com­fort with me­chan­i­cal peo­ple. This an­i­mist re­li­gion, which as­cribes souls to inan­i­mate ob­jects like trees or stones, plays a strong role along­side Bud­dhism in Ja­pa­nese cul­ture.

“That’s the rea­son we are so good for ro­bots,” says Ishig­uro. “We don’t care about flesh bod­ies to de­fine a hu­man.” He hopes that “peo­ple will ac­cept Erica as some type of hu­man”.

But his ul­ti­mate goal re­mains to un­der­stand what it is to be a hu­man, es­pe­cially his own con­scious­ness.

His paint­ing, these days with wa­ter­colours, seems to be pur­su­ing that goal. Equipped with pal­ette and brush, he is strug­gling to con­vey the sense of pres­ence that ob­jects have. How, for in­stance, does his con­scious­ness per­ceive the pres­ence of a chair?

“If I can rep­re­sent my con­scious­ness on the paint­ing,” he says, “I don’t need to de­velop any more ro­bots. I can go back to art.” IM­AGES 01 – 04 Hiroshi Ishig­uro Lab­o­ra­tory

“Peo­ple think giv­ing ro­bots in­ten­tions and de­sires means they will take over the world,” says Alin Albu- Schäf­fer. “We just want them to load the dish­washer.”

Ishig­uro’s idea be­hind mak­ing copies of real peo­ple is to trans­fer the pres­ence of the hu­man to the robot.

Newspapers in English

Newspapers from Australia

© PressReader. All rights reserved.