AI be­ing cre­ated by ‘young men with no prob­lem-solv­ing abil­ity’

The Guardian - - NATIONAL - Ian Sam­ple Sci­ence ed­i­tor

Big tech firms are en­trust­ing some of the most pro­found problems in his­tory “to a bunch of very young men who have never solved a prob­lem in their lives”, a lead­ing Sil­i­con Val­ley tech­nol­o­gist has said.

Vivi­enne Ming will join a panel of lead­ing thinkers on ar­ti­fi­cial in­tel­li­gence in Lon­don next week for a Royal So­ci­ety event, chaired by physi­cist and TV pre­sen­ter Brian Cox. It will take ques­tions on the im­pact AI will have on jobs; risks to so­ci­ety, and its abil­ity to make moral and eth­i­cal de­ci­sions.

While Ming be­lieves AI will be­come an ever more pow­er­ful tool, she thinks there is a prob­lem with the train­ing that com­puter en­gi­neers re­ceive and their un­crit­i­cal faith in AI. “These are very smart men. They are not ma­li­cious. But we are ask­ing them who should I hire, how should we deal with mental ill­ness, who should go to prison and for how long, and they have no idea how to solve these problems,” she said.

“AI is a gen­uinely pow­er­ful tool for solv­ing problems, but if you can’t work out the so­lu­tion to a prob­lem your­self, an AI will not work it out for you.”

Ama­zon is a case in point. The tech firm once tried to re­cruit her as a chief sci­en­tist, telling her it would be her job to make em­ploy­ees’ lives bet­ter. “It be­came clear that Jeff Be­zos’s idea of bet­ter was very dif­fer­ent to mine,” she says. Ama­zon’s in­ven­tion of a wrist­band that buzzed when fac­tory staff reached for the wrong pack­age did not meet with her ap­proval.

Ming heard about the firm’s hopes to build an al­go­rithm that could au­to­mate the hir­ing process, an idea she says she crit­i­cised at the time. In Oc­to­ber, news broke that the firm had scrapped the pro­ject be­cause it was bi­ased against women.

The al­go­rithm scoured CVs to rank them. Be­cause it trained on Ama­zon data, it learned that male ap­pli­cants fared best in the work­place. It pe­nalised those with the word “women’s” on their CVs, as in “women’s row­ing cham­pion”, and down­graded grad­u­ates from women’s col­leges. “If a com­pany doesn’t know how to solve the prob­lem of bias, an AI will not solve the prob­lem for them,” said Ming.

Life ex­pe­ri­ence is per­haps why Ming’s take is dif­fer­ent. Vivi­enne Ming was once Evan Smith, a stu­dent at the Univer­sity of Cal­i­for­nia, who dropped out, be­came home­less, and then clawed his way back to glit­ter­ing suc­cess. Bat­tling demons he strug­gled to un­der­stand, he went to Pitts­burgh to study neu­ro­science. There he met his wife, Norma Chang, who stuck with him when he told her of his wish to be a woman. The cou­ple have two chil­dren.

Ming says she turned down of­fers from Uber and Net­flix, tak­ing a job at a startup called Gild. The firm found that traits such as re­silience and what Ming calls a “growth mind­set” – the flex­i­bil­ity to learn from one’s fail­ures – pre­dicted bet­ter soft­ware en­gi­neers, as rated by hu­man coders.

So the firm built small AIs to crawl blogs and so­cial me­dia feeds for the best can­di­dates, whether they were job hunt­ing or not. Some­times, a tweet car­ried huge weight. One read: “Cel­ery is awe­some.” Out of con­text it sounds “like some­one who is wrong about a gross food,” says Ming. But “cel­ery” was a ref­er­ence to an ob­scure job queue tool writ­ten in the pro­gram­ming lan­guage Python. The tweet, and the pas­sion it con­tained, was a “huge pre­dic­tor” of the can­di­date’s cod­ing skills.

PhD grad­u­ates join tech firms with­out the faintest idea how to solve real-world problems. A thor­ough ground­ing in ethics will help, but Ming be­lieves it takes more than learn­ing the rules from a book. “Ethics is like re­silience, you get good at it by fail­ing.”

She said: “I think it’s in­cred­i­bly valu­able for peo­ple who have suf­fered in some way to have a voice in this. If you come from a back­ground like mine, you are scep­ti­cal. You re­alise tech­nol­ogy in­creases in­equal­ity and it only gets bet­ter if we take ac­tive steps to avoid that.”

Vivi­enne Ming: ‘If you can’t work out a so­lu­tion, AI won’t do it for you’

Newspapers in English

Newspapers from UK

© PressReader. All rights reserved.