Canada’s AI Ex­perts Head South

U.S. com­pa­nies re­cruit its ar­ti­fi­cial-in­tel­li­gence sci­en­tists The coun­try’s lead is “slip­ping away right un­der our nose”

Bloomberg Businessweek (Asia) - - Technology - −Jack Clark and Ger­rit De Vynck

Canada, with a tech in­dus­try one-third the size of Cal­i­for­nia’s, has be­come a leader in the boom­ing mar­ket for ar­ti­fi­cial in­tel­li­gence. Pi­o­neer­ing tech­nolo­gies de­vel­oped in Cana­dian labs can be found in Face­book’s fa­cial-recog­ni­tion al­go­rithms, used to tag peo­ple in im­ages, Pho­tos app, and smart­phone voice iden­ti­fi­ca­tion.

Over the past three years, a hand­ful of lead­ing Cana­dian re­searchers and pro­fes­sors, su­per­stars whose AI work will un­der­pin ev­ery­thing from self­driv­ing cars to smart pros­thetic limbs, have de­fected to U.S. tech com­pa­nies and uni­ver­si­ties, tak­ing their ex­per­tise, and of­ten their stu­dents, with them. Canada’s losses could un­der­mine a decade-long ef­fort by the na­tional and provin­cial gov­ern­ments to leapfrog other coun­tries in ar­ti­fi­cial in­tel­li­gence.

Univer­sity of Toronto com­puter science pro­fes­sor Ge­of­frey Hin­ton spent decades build­ing the tech­niques now used in im­age- and speechreco­g­ni­tion sys­tems. He and two of his stu­dents joined Google in 2013 when the com­pany bought an im­agere­cog­ni­tion startup Hin­ton co-founded. Nando de Fre­itas, who taught com­puter science at the Univer­sity of Bri­tish Columbia, de­vel­oped tech­niques for data anal­y­sis that are now be­ing used by Google’s skunkworks AI di­vi­sion, Deep­Mind. He joined the com­pany in the U.K. in 2014 and oc­ca­sion­ally teaches at Ox­ford. Rus­lan “Russ” Salakhut­di­nov, also in the com­puter science depart­ment at the Univer­sity of Toronto, has done ground­break­ing work that lets com­put­ers iden­tify ob­jects af­ter see­ing only a small set of ex­am­ples, mim­ick­ing how young chil­dren learn. He’s join­ing the ma­chine­learn­ing depart­ment at Carnegie Mel­lon Univer­sity in the spring.

“We had a lead in a field that is po­ten­tially go­ing to be very im­por­tant … and it’s slip­ping away right un­der our nose,” says Ajay Agrawal, a pro­fes­sor at the Univer­sity of Toronto’s Rot­man School of Man­age­ment.

In the mid-2000s the gov­ern­ment­backed Cana­dian In­sti­tute for Ad­vanced Re­search (Ci­far) in Toronto funded work on a then-ob­scure and un­proven tech­nol­ogy, neu­ral net­works, which helps com­put­ers learn to write their own pro­grams for com­plex tasks, in­clud­ing im­age recog­ni­tion and speech pro­cess­ing. A small group of re­searchers made sev­eral break­throughs that wound up in a range of com­mer­cial and con­sumer ap­pli­ca­tions. The Google speech-recog­ni­tion soft­ware in mil­lions of An­droid phones re­lies on tech­niques in­vented by Cana­dian sci­en­tists.

In 2012, Sil­i­con Val­ley started scour­ing Canada for top tal­ent, hir­ing pro­fes­sors, post­grads, and Ph.D.s, and buy­ing star­tups linked to them. In June, Twit­ter bought ma­chine­learn­ing com­pany Whet­lab, whose founders in­clude two Univer­sity of Toronto alumni who worked as post­doc­toral re­searchers with Hin­ton. “With the pull from U.S. com­pa­nies, we run the risk of los­ing our best minds,” says Yoshua Ben­gio, a pro­fes­sor at the Univer­sity of Mon­treal and co-di­rec­tor of Ci­far’s neu­ral net­work pro­gram. “I think it’s im­por­tant that peo­ple in the [provin­cial] gov­ern­ments get to­gether and make it at­trac­tive to stay here in Canada.” Speak­ing at a re­cent AI con­fer­ence at the Univer­sity of Toronto, the city’s mayor, John Tory, said, “I see it as a big part of my job and in­deed the fu­ture of this city to do ev­ery­thing pos­si­ble to make sure they feel they don’t have to leave town, in fact, they shouldn’t leave town.”

Cana­dian com­pa­nies and uni­ver­si­ties are try­ing to pro­tect what they helped build. A pro­gram at the Univer­sity of Toronto to de­velop AI star­tups launched in 2015. Mon­treal is home to sev­eral AI com­pa­nies; au­thor­i­ties there will pro­vide tax cred­its and help nav­i­gate im­mi­gra­tion rules to ease re­cruit­ment of for­eign stu­dents at the Univer­sity of Mon­treal. Malu­uba, a Water­loo, Ont., startup that makes tech­nol­ogy al­low­ing peo­ple to have de­tailed text-based con­ver­sa­tions with com­put­ers, hopes to es­tab­lish in­for­mal links with an AI lab at the univer­sity and is open­ing a re­search of­fice in the city. “I was really ex­cited to find out about Malu­uba, be­cause it meant I could stay Cana­dian,” says Adam Trischler, a re­search sci­en­tist at the com­pany.

At the AI con­fer­ence in Toronto in early De­cem­ber, Salakhut­di­nov, who’s leav­ing for Carnegie Mel­lon, said a ded­i­cated AI cen­ter at a Cana­dian univer­sity could per­suade re­searchers to stay. He noted that Carnegie Mel­lon’s pro­gram has more than 100 Ph.D. re­searchers. “That’s a huge pow­er­house,” he says. The bot­tom line Cana­dian com­pa­nies and uni­ver­si­ties are start­ing pro­grams to en­cour­age AI ex­perts to re­main in the coun­try.

Newspapers in English

Newspapers from Australia

© PressReader. All rights reserved.