An un­can­nily hu­man-sound­ing AI

The Week (US) - - 18 News -

The ro­bots are com­ing, and it turns out they “won’t sound like an over­lord, but a, uhh, Mil­len­nial,” said Alexis Madri­gal in TheAt­ Google last week de­buted a “jaw-drop­pingly” life­like ar­ti­fi­cial­in­tel­li­gence tool at its an­nual de­vel­oper con­fer­ence: an au­to­mated voice as­sis­tant ca­pa­ble of mak­ing salon ap­point­ments, book­ing restau­rant reser­va­tions, and con­duct­ing other tasks over the phone. The bot, known as Du­plex, “sounds like a hu­man,” com­plete with pauses, “cheery col­lo­qui­alisms,” and filler sounds like “um” and “hmm” for added re­al­ism. In a pre-recorded on­stage demon­stra­tion, Du­plex spoke with a hair salon re­cep­tion­ist and chat­ted with a restau­rant em­ployee to book a ta­ble; at no point did the hu­mans on the line appear to re­al­ize they were talk­ing to a robot, caus­ing the au­di­ence of coders to cheer. In the out­side world, crit­ics pounced, ac­cus­ing Google of de­vel­op­ing “de­ceit­ful and un­eth­i­cal” tech­nol­ogy, said Alex Hern in The­ “Hor­ri­fy­ing,” tweeted so­cial me­dia the­o­rist Zeynep Tufekci, who added that “Sil­i­con Val­ley is eth­i­cally lost.” In re­sponse to the out­cry, Google said that it will build in fea­tures that “ex­plic­itly let peo­ple know they are in­ter­act­ing with a ma­chine,” but didn’t pro­vide specifics.

“The robot fu­ture, it’s worth point­ing out, isn’t ex­actly here yet,” said Molly Roberts in Wash­ing­ton­ Google’s AI can chat only in cer­tain sit­u­a­tions, so it’s quite lim­ited in what it can do. Still, this tech­nol­ogy ap­pears to be able to ace the fa­mous test of ma­chine in­tel­li­gence de­vel­oped by Bri­tish com­puter sci­en­tist Alan Tur­ing in 1950—to pass the test, a ma­chine must be­have in a way in­dis­tin­guish­able from a hu­man. To date, our ma­chines have been, well, ma­chine-like enough to spare us from hav­ing to an­swer thorny ques­tions about where ad­vanced AI fits into our so­ci­etal codes. But we could soon face a world where we don’t know “who’s the hu­man and who’s the ma­chine.” It’s not hard to imag­ine “how this kind of tech­nol­ogy could be used for all kinds of ques­tion­able or dan­ger­ous tasks,” said Kevin Roose in NY A hacker could mimic a per­son’s voice and fool a lis­tener into hand­ing over valu­able in­for­ma­tion, or use AI to spread mis­in­for­ma­tion to thou­sands of peo­ple as part of an au­to­mated at­tack. The fact that the back­lash to Du­plex “caught Google by sur­prise,” is, “to me, is the most dis­turb­ing piece” of news.

Google is “not the only com­pany de­vel­op­ing these ser­vices,” said James Vin­cent in It bears ask­ing whether we should al­low tech com­pa­nies to man­age AI’s thorny eth­i­cal ques­tions on their own, or if we need new laws to pro­tect the pub­lic. This tech­nol­ogy is com­ing soon, and it will be wide­spread. Google has pledged to “do the right thing,” but will oth­ers? “We need to have a con­ver­sa­tion about all this, be­fore the ro­bots start do­ing the talk­ing for us.”

When a robot is on the other line

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.