Fast Company - - Innovation By Design - BY JOHN PAVLUS

Mo­bile chat is where it’s okay to be dumb. Your Linkedin page may be a work of art wrought in Hr-friendly prose; your tweets, so sharp you could shave with them. But tex­ting? Let the au­to­cor­rect er­rors fly! Why punc­tu­ate? Isn’t that what old peo­ple do? In fact, why use words at all when any num­ber of pic­to­graphic pokes, nods, and grunts avail­able at the press of a but­ton will get the job done faster? Blame the medium, says Ja­son Corn­well, the chief user ex­pe­ri­ence de­signer be­hind Google’s com­mu­ni­ca­tions apps. “Chat is in­her­ently lim­ited and low-band­width,” he says. It’s also here to stay: “Our phones are ba­si­cally chat ma­chines at this point. That’s the dom­i­nant ac­tiv­ity that vir­tu­ally ev­ery­one does on their phone.” ¶ Allo, a mo­bile mes­sag­ing app launched by Google in 2016, is the com­pany’s at­tempt to use ma­chine learn­ing to make chat, if not smarter, then at least a hell of a lot more use­ful and ex­pres­sive. Ma­chine learn­ing and ar­ti­fi­cial in­tel­li­gence are be­com­ing the en­gines be­hind nearly ev­ery­thing at Google, from Gmail’s spam fil­ters to Al­phago, the neu­ral-net­work-pow­ered soft­ware that re­cently beat the world’s best player at Go (a 2,500-year-old strat­egy game long con­sid­ered im­preg­nable by AI). As part of CEO Sun­dar Pichai’s strat­egy of trans­form­ing Google into an “AI first” com­pany, Corn­well’s de­sign team was charged with build­ing “an app that was about chat on your phone, but at its core was about ma­chine learn­ing,” Corn­well ex­plains. Ex­actly what ma­chine-learn­ing tech­nol­ogy

can do for peo­ple peck­ing out slang-filled, emoji-stud­ded mis­sives to one another is now his job to fig­ure out.

Corn­well had nav­i­gated a sim­i­lar de­sign chal­lenge two years ear­lier as user ex­pe­ri­ence lead for Gmail’s off­shoot app, In­box, which uses ma­chine learn­ing to power its Smart Re­ply fea­ture—those lit­tle rec­tan­gles of sug­gested text that you can select in­stead of man­u­ally typ­ing your own mes­sage. Bring­ing Smart Re­ply tech­nol­ogy to chat via Allo seemed like an ob­vi­ous ex­ten­sion, be­cause “edit­ing text on a phone is still painful,” Corn­well says. Ex­ist­ing pre­dic­tive-text func­tions and au­to­com­plete help you type in­di­vid­ual words slightly faster, but you still have to do the com­pos­ing. Allo’s ap­proach aimed to leapfrog that step en­tirely. “The goal was to think of it as a smarter ex­ten­sion of au­to­com­plete, to help you say the thing that you al­ready were think­ing about say­ing, as close to in your voice as pos­si­ble,” he ex­plains.

That’s where Google’s ma­chine-learn­ing ca­pa­bil­i­ties come in: The more you use Allo, the more its al­go­rithms can as­cer­tain what you sound like and gen­er­ate prewrit­ten re­sponses that don’t sound canned. What’s more, Allo can learn how you text with dif­fer­ent re­cip­i­ents—so it can of­fer up a “nice dude” in re­sponse to your best friend, but not when you’re mes­sag­ing your mom. You can’t edit Allo’s Smart Replies, though, so you’re stuck with us­ing—or ig­nor­ing—what­ever it serves up. But that’s on pur­pose: Chat is “a rapid-fire medium,” says Corn­well, and test­ing showed that “it’s al­most just as fast to type some­thing new out” and send it on the heels of a Smart Re­ply that isn’t quite per­fect.

Smart Re­ply also has a stealthy pur­pose: to in­tro­duce users to Google As­sis­tant, the real brains in­side Allo, along with the com­pany’s new Google Home smart speaker. If you’ve ever been forced to pop out of your mes­sag­ing app in or­der to Google some­thing—say, the lo­ca­tion of the restau­rant where you’ll be meet­ing friends, or flight prices for a va­ca­tion you’re plan­ning with a loved one—you’ll un­der­stand the as­sis­tant’s ap­peal. In Allo, you can just text your query to Google in nat­u­ral lan­guage, as if it’s another per­son in the chat thread. This user ex­pe­ri­ence is au­then­ti­cally con­ver­sa­tional—the as­sis­tant doesn’t use punc­tu­a­tion in its texts, ei­ther!— and it’s so seam­less that new­com­ers might not even re­al­ize they’re us­ing this sep­a­rate and so­phis­ti­cated tech prod­uct. (To help en­cour­age users to ex­per­i­ment and dis­cover fur­ther ca­pa­bil­i­ties, Smart Re­ply pro­vides a but­ton to tap that in­vokes the as­sis­tant in an Allo chat win­dow.) It also avoids the dreaded “Mi­crosoft Clippy” prob­lem: In­stead of a chirp­ing ro­bot in­ter­rupt­ing your con­ver­sa­tion to be “help­ful,” Smart Re­ply cre­ates op­por­tu­ni­ties for the as­sis­tant to in­tro­duce it­self or­gan­i­cally. For in­stance, when a friend mes­sages you ask­ing if you’d like to grab a bite down­town, Smart Re­ply may of­fer up a “sure,” “nah,” and an op­tion from the as­sis­tant sug­gest­ing a search for lo­cal restau­rants. “It’s not like the as­sis­tant is jump­ing up and down at you,” Corn­well says. “It’s of­fer­ing you, in this na­tive for­mat, the abil­ity to take the next step.”

The dis­creet way Allo en­cour­ages in­ter­ac­tions with Google As­sis­tant in­formed later de­signs on the as­sis­tant’s stand-alone app (avail­able on both An­droid and IOS). “When you chat with the as­sis­tant in Allo, we also prompt you to ask the next ques­tion—so if you text ‘weather in moun­tain view,’ the as­sis­tant would pro­vide the weather for that day and then of­fer up more spe­cific phrases you could tap on, like ‘how about this week­end?’ ” Corn­well ex­plains. “We found that when peo­ple were first us­ing the as­sis­tant, they would struc­ture their ques­tions very specif­i­cally to get the ex­act an­swer they wanted, sim­i­lar to how you’d type a query in Google search. But we wanted to help peo­ple learn not only the types of ques­tions they could ask, but also [how to] speak


more con­ver­sa­tion­ally with their as­sis­tant.” The more ca­sual a re­la­tion­ship a user has with her as­sis­tant, the more likely she will be to in­ter­act with as­sis­tant-aug­mented prod­ucts that don’t rely on typ­ing or screens at all, like Google Home.

But Allo’s most imag­i­na­tive fu­sion of chat with ma­chine learn­ing, which it rolled out as an ad­di­tional Allo fea­ture in 2017, isn’t about help­ing you re­spond faster or get­ting things done. It’s about help­ing you put your best face for­ward. Just snap a pic of your­self in Allo, and within sec­onds Google’s ma­chine-learn­ing ca­pa­bil­i­ties trans­form it into a suite of 24 “selfie stick­ers”: cute car­toon like­nesses (cre­ated by La­mar Abrams, a sto­ry­board artist for the Car­toon Net­work) that you can text and share like hy­per-per­son­al­ized emoji. “There’s a com­po­nent of chat that’s about iden­tity and how you see your­self—the craft and care that you put into your own com­mu­ni­ca­tion,” Corn­well says.

Bit­moji—a pop­u­lar third-party app that in­te­grates with Snapchat—lets users cre­ate sim­i­lar avatars too, but only man­u­ally. Allo’s au­to­mated ver­sion gets ar­guably as close to a good like­ness as Bit­moji’s does—but not too close. And that’s by de­sign, ac­cord­ing to Corn­well. “Even if the al­go­rithm was per­fect, peo­ple wouldn’t feel good about [selfie stick­ers] un­less they could put their per­sonal stamp on it,” he ex­plains. In other words, there’s some­thing about tweak­ing a car­i­ca­ture of your­self that’s ut­terly es­sen­tial to trust­ing it. Corn­well draws an anal­ogy to in­stant cake mixes from the 1950s, which re­quired that you crack an egg into the bowl: “There re­ally is no tech­ni­cal rea­son to add that to an in­stant cake mix, but it just didn’t feel like cook­ing un­less you broke an egg. In ma­chine learn­ing,” he ex­plains, “there’s this key ques­tion that de­sign­ers need to an­swer: Which pieces do we au­to­mate for you, and which pieces do you need to do your­self in or­der to feel good about the end re­sult?”

In­deed, Allo pro­vides enough com­bi­na­tions (about 563 quadrillion, if you want to get spe­cific) that you could lit­er­ally cus­tom­ize your selfie sticker un­til the sun burns out and still have plenty left to try. But re­ally, what mat­ters most to peo­ple “is the hair,” Corn­well says. “It’s one of the most crit­i­cal fea­tures in terms of mak­ing peo­ple feel like they were see­ing an au­then­tic ver­sion of them­selves on-screen,” or at least the ver­sion they wanted oth­ers to see. It’s true: When I tested a selfie sticker, my ma­chine-learn­ing-gen­er­ated car­toon hair looked bet­ter than the real thing. As Corn­well and his team con­tinue to ex­pand Allo’s ca­pa­bil­i­ties—the app has only a frac­tion of the user base that imes­sage and Face­book Mes­sen­ger boast—he un­der­stands that, when it comes to tex­ting, it’s not just what you type, but how you look.

Pho­to­graph by Damien Maloney

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.