The Hu­mans Hid­ing Be­hind the

▶ ▶ Ar­ti­fi­cial in­tel­li­gence ser­vices are lean­ing on low-tech reps work­ing long hours ▶ ▶ “I left feel­ing to­tally numb and ab­sent of any sort of emo­tion”

Bloomberg Businessweek (North America) - - Technology -

Amy In­gram, the ar­ti­fi­cial in­tel­li­gence per­sonal as­sis­tant from startup, sounds re­mark­ably like a real per­son. Given ac­cess to your cal­en­dar and e-mail, Amy takes over the mun­dane tasks of sched­ul­ing meetings and send­ing ap­point­ment re­minders. After she made her de­but in 2014, beta testers praised her “hu­man­like tone” and “elo­quent man­ners.” “Ac­tu­ally bet­ter than a hu­man for this task,” one tweeted.

This ar­ti­fi­cial in­tel­li­gence, how­ever, isn’t wholly ar­ti­fi­cial. Be­hind most ev­ery e-mail is a per­son—some­one like 24-year-old Wil­lie Calvin.’s soft­ware can usu­ally guess from its e-mail scans when “to­mor­row” means Tues­day, but it’s a ways from be­ing able to op­er­ate on its own. An spokes­woman says train­ers re­view “the vast ma­jor­ity” of in­for­ma­tion in e-mails so the sys­tem can im­prove.

Calvin joined in De­cem­ber 2014, shortly after grad­u­at­ing from the Univer­sity of Chicago with a pub­lic pol­icy de­gree. He was un­der the im­pres­sion his $45,000-a-year job as an AI trainer would be split evenly be­tween spot-check­ing Amy and do­ing more am­bi­tious prod­uct de­vel­op­ment. He says he was asked, as part of the job ap­pli­ca­tion, to write a one-page es­say ar­gu­ing that au­toma­tion would be good for jobs and work­ers.

After Calvin started work, he says, the prod­uct part of the job never ma­te­ri­al­ized. In­stead, he sat in front of a com­puter, click­ing and high­light­ing the bot’s right and wrong re­sponses to re­quests, for up to 12 hours a day. “It was ei­ther re­ally bor­ing or in­cred­i­bly frus­trat­ing,” he says. “It was a weird com­bi­na­tion of the ex­act same thing over and over again and re­ally frus­trat­ing sin­gle cases of a per­son de­mand­ing some­thing we couldn’t pro­vide.” de­clined to com­ment on spe­cific hir­ing prac­tices. Kristal Bergfield, who over­sees’s train­ers, says the job has evolved over time and en­tails hard work. “We’re build­ing some­thing that’s en­tirely new,” she says. “It’s an in­cred­i­bly am­bi­tious thing, and so are the peo­ple who work here.”

A hand­ful of com­pa­nies em­ploy hu­mans pre­tend­ing to be ro­bots pre­tend­ing to be hu­mans. In the past two years, semi­au­to­mated ser­vices have sprung up to of­fer e-mail sched­ulers (, Clara); shop­ping as­sis­tants (Op­er­a­tor, Mezi); and do-any­thing concierges (Magic, Face­book’s M, Gobut­ler). Star­tups in this arena have raised more than $50 mil­lion in ven­ture cap­i­tal fund­ing in the past two years, and most aim to em­ploy as few hu­mans as pos­si­ble. But for now, the com­pa­nies tend to use sim­i­larly vague mar­ket­ing lingo and in truth are pow­ered largely by peo­ple.

Face­book turned the spot­light on hu­man-as­sisted AI last sum­mer when it in­tro­duced M, a chirpy per­sonal as­sis­tant bot that lives in Mes­sen­ger, its chat app. M’s Ai-gen­er­ated re­sponses are re­viewed, edited if nec­es­sary, and sent out by a team of a few dozen con­trac­tors work­ing from the so­cial net­work’s Menlo Park, Calif., cam­pus, ac­cord­ing to the com­pany. Face­book won’t say what hours the con­trac­tors work or how of­ten they cor­rect M’s guesses.

Clara, which of­fers an e-mail sched­ul­ing ser­vice a lot like’s, uses

Newspapers in English

Newspapers from Canada

© PressReader. All rights reserved.