Page 62 - The Economist
P. 62

62 Business                                                                  The Economist December 9th 2017
       2 robotics department was famously plun-  plies to the Chinese giants, which are try-
        dered by Uber in 2015). Academic confer-  ing to establish Western outposts and hire  Out in the open  2
        ences, such as this week’s Neural Informa-  American researchers. Baidu has opened  Artificial-intelligence-related research*
        tion Processing Systems in Long Beach,  two research labs with an AI focus in Sili-  By company-affiliation, 2000-16
        California, double up as places to shop for  con Valley, in 2013 and thisyear. Western AI
                                                                                     0   200  400  600  800 1,000
        talent (see page 78). The best recruiters are  researchers rate them highly but prefer to
        academia’s AI celebrities: people like Yann  work for the American giants, in part due  Microsoft
        LeCun of Facebook and Geoffrey Hinton  to theirrelative transparency.   Google
        of Google—both former professors who  If companies can lure the right people
        keep a university affiliation—can attract  in AI, the effect is to extend their work-  IBM
        othersto workalongside them. Proprietary  forces exponentially. AI is “like having a  Facebook
        data can also serve as a draw, if the huge  million interns” at one’s disposal, says
        salaries are not enough.           Benedict Evans of Andreessen Horowitz.  Baidu
           If none of that works, companies buy  Thatcomputational poweristhen integrat-
        whole startups. The tech industryfirst took  ed into firms’ existingbusinesses.  Amazon
        notice of this trend in 2014, when Google  The advantagesofAI are mostvisible in  Source: Ajay Agrawal and Amir  major AI conferences
                                                                                                  *Papers from five
                                                                               Sariri, University of Toronto
        spent an estimated $500m on DeepMind,  firms’ predictions of what users want.
        a startup with no revenue or marketable  Automated recommendations and sugges-
        product but a team of “deep learning” re-  tions are responsible  for around three-  plied in the corporate world. David Kenny,
        searchers; after the deal they designed a  quarters of what people watch on Netflix,  the boss ofWatson, IBM’s AI platform, pre-
        program that beat the world champion at  forexample, and more than a third of what  dicts that there will be “two AIs”: compa-
        “Go”, an ancient board game. Other firms  people buy on Amazon. Facebook, which  nies that profit from offering AI-infused
        have also shelled out to buy money-losing  owns the popular app Instagram, uses  servicesto consumersand otherswhich of-
        startups, which are typicallyvalued not on  machine learning to recognise the content  fer them to businesses. In practice, the two
        future profits or even sales but instead re-  of posts, photos and videos and display  worlds meet because of the tech giants’
        ceive a price for each employee that can be  relevant ones to users, as well as filter out  cloud-computing arms. Providers are com-
        as much as $5m-10m.                spam. In the past it ranked posts chrono-  peting to use AI as a way to differentiate
                                           logically, but serving up posts and ads by  their offerings and lock in customers. The
        Behind closed doors                relevance keeps users more engaged.   three largest—Amazon Web Services, Mi-
        Companies have different philosophies  Without machine learning, Facebook  crosoft’s Azure and Google Cloud—offer
        abouthowto deal with staff. Some, such as  would never have achieved its current  application-programminginterfaces(APIs)
        Microsoft and IBM, invest heavily in AI re-  scale, argues Joaquin Candela, head of its  thatprovide machine-learningcapabilities
        search and publish a large number of pa-  applied AI group. Companies that did not  to other companies. Microsoft’s cloud of-
        pers (see chart 2), but do not require re-  use AIin search, orwere late to do so, strug-  fering, Azure, for example, helped Uber
        searchers to apply their findings to  gled, as in the case of Yahoo and its search  build a verification tool that asks drivers to
        money-making activities. At the opposite  engine, and also Microsoft’s Bing.   take a selfie to confirm their identities
        end of the scale are Apple and Amazon,  Amazon and Google have gone fur-  when they work. Google Cloud offers a
        which do not have enormous research ini-  thest in applying AI to a range of opera-  “jobs API”, which helps companies match
        tiatives, expect all work to feed into pro-  tions. Machine learning makes Amazon’s  jobseekers with the best positions.
        ducts and are tight-lipped about their  online and physical operations more effi-
        work. Google and Facebook are some-  cient. It has around 80,000 robots in its ful-  AI on the brain
        where in between on whether researchers  filmentcentres, and also usesAI to categor-  Many firms in other industries, from retail-
        musttoil onlyon money-makingventures.   ise inventory and decide which trucks to  ing to media, stand to benefit from what
           The intense battle for talent may force  allocate packages to. For grocery ordering,  those in the cloud business tout as the “de-
        secretive companies to become more  ithasapplied computervision to recognise  mocratisation” of AI. Providing AI to com-
        open. “Ifyoutell them, ‘come workwith us  which strawberries and other fruits are  paniesthatdo nothave the skillsor scale to
        but you can’t tell anyone what you’re  ripe and fresh enough to be delivered to  build up sophisticated capabilities inde-
        working on’, then they won’t come be-  customers, and is developingautonomous  pendently could be a money-spinner in
        cause you’ll be killing their career,” ex-  drones that will one day deliverorders.   the $250bn cloud market. But providers of-
        plains Mr LeCun, who leads Facebook’s AI  As for Google, it uses AI to categorise  ten must customise APIs for clients’ com-
        research lab. This trade-off between secre-  content on YouTube, its online-video web-  plex needs, which is time-consuming. Mi-
        cy and the need to attract people also ap-  site, and weed out (some) objectionable  crosoft, with its history of selling software
                                           material, and also to identify people and  to clientsand offeringthem support, seems
                                           group them in its app, Google Photos. AI is  likelyto do well in thisarea. Itis onlya mat-
           Here come the corporations  1   also embedded in Android, its operating  ter of time before  AI offerings become
           Global merger-and-acquisition activity   system, helping it to work more smoothly  “more and more self-help”, countersDiane
           related to artificial intelligence  and to predict which apps people are inter-  Greene, who runs Google Cloud.
           Number of deals      Value, $bn  ested in using. Google Brain is regarded in  IBM is another contender, having
                                           the field of AI as one of the best research  backed a huge marketing campaign for its
           120                       24
                                           groups at applying machine-learning ad-  Watson platform. AI researchers tend to be
           100                       20
                                           vances profitably, for example by improv-  dismissive of IBM, which has a large con-
           80                        16    ing search algorithms. As for DeepMind,  sulting business and a reputation for valu-
           60                        12    the British firm may not ever generate  ing time billed over terabytes. The firm’s
           40                        8     much actual revenue for Alphabet, but it  critics also point out that, although IBM
                                           has helped its parent save money by in-  has invested over $15bn in Watson and
           20                        4
                                           creasing the energy efficiency of its global  spent $5bn between 2010 and 2015 to buy
            0                        0     data centres (and its Go experiment was a  companies, much of that with the aim of
             2010 11  12  13  14  15  16  17*
                                           public-relations coup).           acquiring proprietary data, for the most
           Source: PitchBook     *To Dec 4th
                                             Artificial intelligence is also being ap-  part it does not have unique data of its  1
   57   58   59   60   61   62   63   64   65   66   67