Fast Company - - Contents - By Harry Mccracken

For giants and up­starts alike, ar­ti­fi­cial in­tel­li­gence is the big­gest new op­por­tu­nity in busi­ness. Here’s who will win—and why.


In­spired by news that Vladimir Putin had told Rus­sian stu­dents the coun­try that leads in ar­ti­fi­cial in­tel­li­gence will rule the world, the Tesla and Spacex CEO de­clared the global race to dom­i­nate AI might turn into real war—and that the first strike could well be launched by an al­go­rithm rather than a fle­s­hand-blood leader. Chas­tised by one of his fol­low­ers for the gloomy prog­nos­ti­ca­tion, he apol­o­gized and then con­fessed, “I was depressing my­self too. :-( ”

Musk is a techno-provo­ca­teur with few equals. How­ever, plenty of peo­ple share his take on AI. Even sun­nier fore­casts about the fu­ture of AI, de­tail­ing how self-driv­ing cars might rad­i­cally re­duce high­way car­nage, are typ­i­cally too long-range to of­fer much of a sense of com­fort.

Mean­while, as ev­ery­one muses about where AI might take us, the tech­nol­ogy has ar­rived. First given its name by sci­en­tists at a sem­i­nal con­fer­ence held at Dart­mouth Col­lege in 1956 (they had pre­dicted that pro­gram­mers would be able to sim­u­late the work­ings of the hu­man brain in just a few years), AI now has a per­va­sive and ob­vi­ous im­pact, par­tic­u­larly when it comes to the branch known as ma­chine learn­ing and, in es­pe­cially

ad­vanced form, as deep learn­ing. AI is how Google Pho­tos knows that two snap­shots taken 50 years apart are both of your great-un­cle. It’s how Face­book weeds spam out of your feed. It’s even how the iphone ekes as much life as pos­si­ble out of a bat­tery charge.

In­creas­ingly, smart­phones, smart-home giz­mos, and other de­vices are mor­ph­ing into front ends for Ai-in­fused ser­vices, such as Ap­ple’s Siri, Ama­zon’s Alexa, and Google’s As­sis­tant. “If you un­pack what’s be­hind Alexa, be­hind the Echo, it’s not just a speaker,” says Swami Si­va­sub­ra­ma­nian, VP of Ama­zon AI. “It’s ac­tu­ally an in­tel­li­gent, cloud-en­abled dig­i­tal as­sis­tant, us­ing deep-learn­ing-driven speech recog­ni­tion and nat­u­ral lan­guage un­der­stand­ing.”

As AI be­gins to touch ev­ery as­pect of their busi­nesses, the tech giants are joust­ing to re­cruit su­per­stars in the field, pil­fer­ing brain­power from academia (New York Univer­sity’s Yann Le­cun is now at Face­book) and each other (Jia Li left Snapchat to co-run Google Cloud’s ma­chine-learn­ing group). “Be­cause the tech­nol­ogy is so pow­er­ful, there’s a large de­mand for tal­ent that un­der­stands how to ap­ply it,” says Scott Pen­berthy, di­rec­tor of ap­plied AI for Google Cloud. Re­search firm Paysa re­leased a study in April that showed Ama­zon in­vest­ing $228 mil­lion in new AI po­si­tions, fol­lowed by Google ($130 mil­lion) and Mi­crosoft ($75 mil­lion).

Only a hand­ful of com­pa­nies can com­pete at this level. Ama­zon, Ap­ple, Face­book, Google, and Mi­crosoft “have all these Phds, and they have all this PHD tech,” says Ali Gh­odsi, the CEO of Databricks, a startup that works with busi­nesses to add AI to their pro­cesses. “But the rest of the For­tune 2000, they don’t,” lead­ing to what Gh­odsi states is an emerg­ing “1% prob­lem,” in which only the largest play­ers have the where­withal to take full ad­van­tage of this new tech­nol­ogy.

The sav­ing grace is that busi­nesses of all sizes can avail them­selves of some of these AI in­no­va­tions. In fact, Ama­zon, Mi­crosoft, and Google are count­ing on it. Their cloud-com­put­ing plat­forms—ama­zon Web Ser­vices, Azure, and Google Cloud, re­spec­tively—in­clude en­ter­prise AI of­fer­ings such as im­age recog­ni­tion, nat­u­ral lan­guage pro­cess­ing, and lan­guage trans­la­tion. All three com­pa­nies see AI as the key to driv­ing fu­ture growth of their cloud plat­forms; at present, Ama­zon Web Ser­vices is a $16 bil­lion busi­ness that’s in­creas­ing 42% year over year, though that pace has slowed as Mi­crosoft and Google be­gin to catch up. Then there’s IBM, which calls its fla­vor of ar­ti­fi­cial in­tel­li­gence “cog­ni­tive com­put­ing” and has ef­fec­tively branded it as “Wat­son” to sell as a ser­vice. While Face­book and Ap­ple don’t of­fer their own plat­forms, they pub­lish aca­demic pa­pers on their re­search—and, in Face­book’s case, it open-sources some of the tech­nolo­gies it’s cre­ated.

As a busi­ness tool, AI is still in its in­fancy. Re­cent stud­ies by both the Mckin­sey Global In­sti­tute and Mit/bos­ton Con­sult­ing Group re­ported that only about 20% of com­pa­nies have im­ple­mented the tech­nol­ogy in a mean­ing­ful way. But un­like past tech­no­log­i­cal in­flec­tion points—such as the emer­gence of e-com­merce in the 1990S—AI doesn’t nat­u­rally fa­vor nim­ble star­tups. Be­cause AI craves data of the sort that can take years to ac­cu­mu­late, “there’s ac­tu­ally an ad­van­tage to in­cum­bency, be­cause the more knowl­edge you have to train your AI, the more valu­able it is,” ar­gues David Kenny, IBM’S se­nior VP for Wat­son and IBM Cloud.

What AI does share with past tech trends, how­ever, is a ten­dency to be over­hyped in such a way that can ob­scure its real ca­pa­bil­i­ties. In Septem­ber, for ex­am­ple, an in­ves­ti­ga­tion by med­i­cal news site Stat de­ter­mined that IBM’S Wat­son for On­col­ogy ser­vice, for med­i­cal in­sti­tu­tions, failed to live up to a pro­mo­tional cam­paign that sug­gested it was a can­cer-fight­ing break­through. “We’re prob­a­bly at the peak of hype and ex­pec­ta­tions: ‘AI is go­ing to do ev­ery­thing for us, it’s go­ing to take over the world, if you don’t touch AI you’re go­ing to be left be­hind,’ ” says data sci­en­tist Steven Fin­lay, the au­thor of Ar­ti­fi­cial In­tel­li­gence and Ma­chine Learn­ing for Busi­ness.

Still, AI is no mere fad. The dol­lars flow­ing into R&D are huge—at the rate of more than $30 bil­lion a year—and the ul­ti­mate im­pact on pro­duc­tiv­ity and en­hanced con­sumer de­mand are pro­jected to be in the tril­lions. It’s no sur­prise that AI has be­come the fo­cal point in the war among tech’s big­gest play­ers, and that it is af­fect­ing how many en­ter­prises, in tech and be­yond, view their fu­tures.

To un­der­stand where AI is to­day—and where it is head­ing—busi­nesses must ac­knowl­edge both the ex­cite­ment and the un­cer­tainty. Here are seven prac­ti­cal lessons from the front lines, in in­dus­tries from tech to re­tail, craft brew­ing to real es­tate. AI is of­fi­cially ev­ery­where, in ways we all need to ap­pre­ci­ate.


The leader of a ma­jor fash­ion busi­ness re­cently de­cided that AI needed to be a tool in his com­pany’s ar­se­nal. But he wasn’t sure what that meant. The en­ter­prise had worked with Google, IBM, and Mi­crosoft in the past. Should it align with one of them on AI? And to do what, ex­actly?

Even those in the AI field warn against be­com­ing smit­ten with the tech­nol­ogy just be­cause it’s trendy. “Some­times I talk to cus­tomers who are like, ‘Hey, we want to use AI,’ with­out re­ally think­ing about why, or what it can do for them,” says Marco Casalaina, VP of prod­uct for Sales­force’s AI, which is branded as Ein­stein.

A good way to start is by iden­ti­fy­ing busi­ness prob­lems that AI might help with—and, in­stead of try­ing to tackle all of them at once, choos­ing a man­age­able pi­lot project. “If you think you’re go­ing to solve this in one go, it’s never go­ing to hap­pen,” says Deep Varma, VP of en­gi­neer­ing at real es­tate in­for­ma­tion provider Tru­lia, who ad­vises AI new­bies to “pick very spe­cific pain points.”

The key is to not be se­duced by AI’S po­ten­tial but rather to fo­cus on your own goals. Pro­cesses that hu­man be­ings (read: em­ploy­ees) re­gard as drudgery are of­ten the best place to be­gin. For in­stance, by us­ing nat­u­ral lan­guage pro­cess­ing ser­vices from Mi­crosoft’s Azure, travel tech­nol­ogy com­pany Sabre is ex­per­i­ment­ing with a Face­book Mes­sen­ger bot that can field straight­for­ward ques­tions about ex­ist­ing reser­va­tions. Its cus­tomers in the travel in­dus­try,

says di­rec­tor of Sabre Stu­dios Chad Cal­laghan, “see a fu­ture where agents are fo­cused on highly com­plex itin­er­ar­ies where you re­ally want that per­son-to-per­son in­ter­ac­tion, and the bot is able to sup­port more rou­tine sorts of re­quests.”


Around the turn of the last decade, a bit of tech­no­log­i­cal jar­gon gained cur­rency: “big data.” Its buzzi­ness re­flected a new un­der­stand­ing that there was value in col­lect­ing, or­ga­niz­ing, and an­a­lyz­ing vast amounts of in­for­ma­tion about ev­ery as­pect of a busi­ness, from man­u­fac­tur­ing pro­ce­dures to cus­tomer in­ter­ac­tions. Yet it was far eas­ier to hoard big data than to fig­ure out what to do with it. Many com­pa­nies “kept col­lect­ing data for years and years and years, and it sat on servers and col­lected dust,” says Mark John­son, CEO of geo­graphic AI startup Descartes Labs. En­ter ar­ti­fi­cial in­tel­li­gence, which can iden­tify pat­terns on a scale that would flum­mox a mere mor­tal.

“Data is the food that feeds AI,” says Sales­force’s Casalaina. The more it con­sumes, the smarter it gets. At Google’s I/O de­vel­oper con­fer­ence in May, Google AI chief John Gian­nan­drea ex­plained the con­cept to me by us­ing an ex­am­ple in­volv­ing his 4-year-old daugh­ter. She had spot­ted a gi­ant-wheeled, 19th-cen­tury “penny-farthing” bi­cy­cle and—once he’d told her what it was—she was im­me­di­ately ca­pa­ble of iden­ti­fy­ing any other penny-farthing she might en­counter. With com­put­ers, “we’d have to show them 100,000 penny-far­things and tell them it’s a bike. But once they’d seen 100,000, they’d prob­a­bly be bet­ter at iden­ti­fy­ing them than hu­mans are.”

Even com­pa­nies with plenty of data to mine of­ten need to clean up messy data­bases (Tru­lia’s Varma winces as he re­calls a com­pany that had stored a de­fault time stamp of 00:00:00 on Thurs­day, Jan­uary 1, 1970, for ev­ery record), merge dis­parate repos­i­to­ries, and gen­er­ally make in­for­ma­tion al­go­rithm-friendly. “The first thing to do is to take the data out of the data­bases, make it freely avail­able and ac­ces­si­ble,” rec­om­mends Jean-françois Faudi, se­nior in­no­va­tion man­ager at Air­bus De­fence and Space. For Air­bus, that in­volved mov­ing its satel­lite im­agery to Google Cloud. Now the com­pany can use ma­chine learn­ing to dis­tin­guish be­tween snow and clouds—a feat that, it turns out, com­put­ers are more adept at ac­com­plish­ing than hu­mans.


Com­pa­nies that al­ready care about data have a head start when it comes to AI, no mat­ter what their cat­e­gory. Craft brew­ing, for in­stance, would not make any­body’s list of the in­dus­tries most ob­vi­ously poised to ben­e­fit from the tech­nol­ogy. But Brian Faivre, the brew­mas­ter at Bend, Ore­gon’s Deschutes Brew­ery—the eighth-largest craft brew­ery in the United States— hap­pens to have a com­puter science de­gree. (“I home brewed through­out col­lege but didn’t know that you could have a real job in the craft-beer in­dus­try,” he ex­plains.)

Faivre has long been in­trigued by how data science could be ap­plied to beer mak­ing, and the brew­ery has been log­ging stats about its pro­duc­tion process for years. Mak­ing suds is all about con­trol­ling fer­men­ta­tion, which brew­eries do by ad­just­ing tem­per­a­ture. They know when it’s time to do so by ex­tract­ing liq­uid sam­ples from tanks and mea­sur­ing their den­sity—a cum­ber­some, inex­act pro­ce­dure. But work­ing with a data in­fra­struc­ture firm called Osisoft, Deschutes fed data about past pro­duc­tion into Mi­crosoft’s Cor­tana In­tel­li­gence Suite, part of the Azure plat­form. That has al­lowed Deschutes to be­gin pre­dict­ing the op­ti­mum time to raise the tem­per­a­ture, elim­i­nat­ing the need for the den­sity-mea­sur­ing step and shav­ing a cou­ple of days off the 12-day fer­men­ta­tion cy­cle. The re­sult: The com­pany can pro­duce more beer with­out com­pro­mis­ing qual­ity.

Ul­ti­mately, craft brew­ing is not about ruth­lessly ef­fi­cient mass pro­duc­tion, and Deschutes is a long way from us­ing AI to elim­i­nate the hu­man el­e­ment of its busi­ness: Beer mak­ers

are free to tin­ker with the al­go­rith­mi­cally gen­er­ated rec­om­men­da­tions, and they do. “What we’ve al­ways stressed is that our brew­ers are in con­trol,” says Faivre. But the com­pany’s use of AI to in­crease pro­duc­tion is crit­i­cal to its fu­ture: The ad­di­tional sales are help­ing to fund the con­struc­tion of a new brew­ery in Roanoke, Vir­ginia, which will give Deschutes a na­tional foot­print for the first time.


The fact that tech giants are turn­ing their own in-house AI into on-de­mand ser­vices is a boon for or­ga­ni­za­tions that are tight on re­sources. Chris Adz­ima, for ex­am­ple, a se­nior in­for­ma­tion­sys­tems an­a­lyst for the sher­iff’s of­fice in Wash­ing­ton County, Ore­gon, be­came in­trigued last year by a new Ama­zon Web Ser­vices of­fer­ing called Rekog­ni­tion, which in­cludes the abil­ity to rec­og­nize faces. The county’s trove of hun­dreds of thou­sands of book­ing pho­tos taken at the time of ar­rests has be­come so over­whelm­ing that even fil­ter­ing a search by age, gen­der, or race of­ten doesn’t mean­ing­fully nar­row things down. That lim­its its use­ful­ness when po­lice of­fi­cers need to iden­tify a per­son of in­ter­est, such as a sho­plifter caught on cam­era. “I am not a data sci­en­tist, nor do I have any idea how fa­cial recog­ni­tion or ar­ti­fi­cial in­tel­li­gence works,” Adz­ima cheer­fully ad­mits. Within a cou­ple of months, how­ever, he was able to fash­ion a sys­tem

a sys­tem that uses Rekog­ni­tion to match newly taken pho­tos with ones from the ar­chive. So far, it’s helped iden­tify 20 sus­pects.

It was also an ex­tra­or­di­nary bar­gain. The ini­tial setup cost the sher­iff ’s of­fice only around $400; the monthly bill from Ama­zon Web Ser­vices is about $6. “With ev­ery dol­lar I spend, I’m ac­count­able to the tax­pay­ers,” says Adz­ima. “We’re spend­ing such small amounts of money and we’re get­ting a huge re­turn on in­vest­ment.” 5. BUILD IF YOU MUST Fa­cial recog­ni­tion is a type of AI that’s ap­pli­ca­ble in var­i­ous sce­nar­ios, mak­ing Ama­zon’s ver­sion im­me­di­ately use­ful in many fields. In some in­stances, how­ever, com­pa­nies need to em­ploy AI that’s been care­fully tweaked for a par­tic­u­lar pur­pose.

“We don’t tend to ask our ra­di­ol­o­gist for art ad­vice, or our lawyer for stock-pick­ing ad­vice. You go to ex­perts for dif­fer­ent things,” says IBM’S Kenny. That’s why IBM tai­lors Wat­son for spe­cific in­dus­tries, from ed­u­ca­tion to sup­ply-chain man­age­ment. His point re­flects a ba­sic truth about AI: The more am­bi­tious you get, the less likely a plain-vanilla al­go­rithm will suf­fice.

Real es­tate hub Tru­lia hoped to use AI to rum­mage through its col­lec­tion of mil­lions of pho­tos of homes for sale and rent and dis­tin­guish among kitchens, bed­rooms, and bath­rooms— and even no­tice when a kitchen fea­tures such price-boost­ing ex­tras as gran­ite coun­ter­tops. That’s not the sort of in­tel­li­gence that’s avail­able as a com­mod­ity.

“Tru­lia needs to innovate,” says Varma, the com­pany’s data-science guru. To do so, he con­cluded, “we need to own the com­puter vi­sion in­ter­nally.” As a di­vi­sion of Zil­low, the lead­ing dig­i­tal bro­ker val­ued at $5.5 bil­lion, the com­pany could rea­son­ably as­pire to treat AI as a strate­gic im­per­a­tive and in­vest ap­pro­pri­ately. Even though it’s an Ama­zon Web Ser­vices cus­tomer, Tru­lia chose to hire its own ma­chine-learn­ing ex­perts and de­velop its own pro­pri­etary mod­els.

Some­times the in­di­vid­u­al­iza­tion can be min­i­mal. The car-buy­ing site Ed­munds, which of­fers prospec­tive buy­ers re­sources such as specs, prices, and re­views, has in­te­grated AI into nu­mer­ous as­pects of its busi­ness, from fore­cast­ing rev­enue to se­cur­ing its web­site. Much like Tru­lia, it wanted to use the tech­nol­ogy to help it sort through hun­dreds of thou­sands of pho­tos, in this in­stance, to iden­tify the types of ex­te­rior and in­te­rior shots it has of spe­cific makes and mod­els. “We got 90% there us­ing Google off the shelf, and then we were able to just tweak it at the end to be bet­ter about un­der­stand­ing ve­hi­cle images ver­sus all the images that Google is look­ing at ev­ery­where else,” says VP of prod­uct in­no­va­tion Greg Shaf­fer. 6. GET EV­ERY­BODY IN­VOLVED—AND KEEP THEM IN­VOLVED Whether a com­pany seeks lots of help or takes on more of the heavy lift­ing it­self, AI’S worth is deeply tied to the specifics of in­di­vid­ual busi­ness chal­lenges. Which means that it can only be ef­fec­tive if stake­hold­ers are as en­gaged and com­mit­ted as IT staffers are.

“Or­ga­ni­za­tions have the ten­dency to sit back, just like [af­ter] they pur­chased tech­nol­ogy in the past, and ex­pect tech so­lu­tions to do all the hard work for them,” says Sjo­erd Gehring, global VP of tal­ent ac­qui­si­tion at John­son & John­son. “That’s the one thing that re­ally doesn’t work with AI.” Though Gehring’s job fo­cuses on peo­ple rather than tech­nol­ogy, he cham­pi­oned J&J’S ef­fort, in col­lab­o­ra­tion with Google Cloud and re­cruit­ing soft­ware provider Jibe, to in­cor­po­rate AI into the way it finds ev­ery­one from med­i­cal re­searchers to truck driv­ers. The com­pany says that ap­pro­pri­ate ap­pli­cants are up by 41% since it be­gan us­ing a search en­gine pow­ered by Google’s ma­chine-learn­ing al­go­rithms to match a mil­lion job can­di­dates a year with the 25,000 po­si­tions it fills.

Af­ter that, “it’s a con­tin­u­ous process of re­fine­ment and train­ing to get your im­ple­men­ta­tion bet­ter and bet­ter and bet­ter,” says Meg Sut­ton, di­rec­tor of re­tail client ex­pe­ri­ence at H&R

Block. The tax be­he­moth be­gan in­te­grat­ing ad­vice from IBM’S Wat­son into its rou­tine this year and found that this in­put—based on 74,000 pages of U.S. tax code and de­liv­ered on a sec­ond screen used by its pre­par­ers—in­creased client sat­is­fac­tion. Now the com­pany is work­ing on ver­sion 2.0 for next year’s tax sea­son.


The fi­nal les­son comes back to a sim­ple, hu­man one: pa­tience. There is much to be gained from uti­liz­ing AI, but there’s also much yet to dis­cover. The tech­nol­ogy’s ul­ti­mate cul­tural im­pact—de­spite all the prog­nos­ti­ca­tion, from Elon Musk to Mark Zucker­berg—is im­pos­si­ble to know.

Even­tu­ally, as with pre­vi­ous epoch-shift­ing tech­nolo­gies, “there are peo­ple who will ride the wave and be suc­cess­ful, and other peo­ple who will go against the wave and then be swept away,” says Yunkai Zhou, who spent years build­ing ma­chine-learn­ing tech­nolo­gies into Google’s ad plat­form be­fore co­found­ing startup Thanks to the bold new ex­per­i­ments of com­pa­nies both big and small, we are all learn­ing where the cur­rent might lead.

AI re­search is flow­er­ing be­cause com­put­ing power has caught up with the am­bi­tions of ma­chine-learn­ing spe­cial­ists.

The suc­cess of Ama­zon’s Alexa voice as­sis­tant has re­ver­ber­ated through­out the busi­ness world, mak­ing Aipow­ered chat the next big thing.

Com­put­ers can now iden­tify ob­jects in images bet­ter than hu­mans can, which has mean­ing­ful im­pli­ca­tions for fa­cial recog­ni­tion.

Com­puter vi­sion and speech recog­ni­tion are the two AI dis­ci­plines most likely to have widerang­ing im­pact for busi­ness ap­pli­ca­tions and cus­tomer ex­pe­ri­ence.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.