MANY TECH FIRMS’ offices boast luxurious perks such as nap pods, massages and soda fountains that offer employees a choice of exotically flavoured sparkling water. Corporate bosses like to think that finding customised AI solutions is just as easy as selecting a fizzy drink with a hint of grapefruit. They are wrong. Buying AI takes time, can feel like hard work, and the results are often imperfect.
A number of vendors are scurrying to come to would-be users’ aid. The leaders are the West’s biggest providers of cloud storage: Amazon, Google and Microsoft. Cloud computing is a vast market worth $300bn, and fiercely competitive. All three firms offer pre-trained models that corporate clients can use to build AI-enabled systems. For example, they all sell “vision” tools that enable customers to use computer vision to improve their existing services and build new ones. Uber, the ride-hailing firm, worked with Microsoft’s toolset to design a system that scans drivers’ faces to confirm their identity when they start a shift. C-Span, a television network, used Amazon’s vision system to compile a database of politicians so it can quickly name them when they appear on screen.
A broad range of tools is available to help mainstream companies build anything from search and recommendation engines to speech-recognition and translation systems, customer-service bots and more. Jeff Dean, director of Google Brain, the search giant’s AI-research arm, reckons there are 10m organisations in the world that “have a problem that would be amenable to a machine-learning solution. They have the data but don’t have the experts on staff.”
The potential corporate market for AI software, hardware and services is vast: around $58bn by 2021, compared with $12bn last year, according to IDC, a research firm. Amazon has a clear lead in the broader cloud market, with a 44% share of the total, compared with Microsoft’s 7% and Google’s 2.3%, but for AI tools the field remains wide open. Paul Clarke, chief technology officer of Ocado, an online grocer, says it can be good for clients to be promiscuous and use the best tools from each. He thinks it unlikely that any one of them will sweep the board.
Cloud providers try to differentiate their AI offerings in two ways: by ease of use, through a well-designed interface, and by offering better algorithms. Each of the tech giants draws on where its “strength is today”, says Joseph Sirosh of Microsoft. For example, Google offers an excellent tool which companies can use to create or redesign their own search engines, and has especially good engineering talent. Microsoft and Amazon have solid tools for voice recognition. Microsoft’s interface currently has the best design, says Pedro Domingos, a professor of computer science at the University of Washington and author of “The Master Algorithm”, a book about AI and business.
In future tech firms will develop more specialised hardware that will help companies crunch enormous data piles more quickly. Google has a lead in this area; it has built some remarkably powerful custom chips, called Tensor Processing Units (TPUs), and uses other customised accelerators to increase the processing speed of its data centres. The tech firms are also offering free open-source libraries to clients’ machine-learning experts that can be used to design AI-enabled programs. This is “not altruistic”, says Matt Turck of FirstMark Capital, a venture investor. Tech firms want to provide great tools in order to attract clients to their platforms and impress AI experts.
Microsoft has more experience than either Amazon or Google of catering to large firms’ software needs, so it is well placed to serve mainstream companies in need of help with AI. But most such offerings still require a lot of customisation and technical work to make them useful, says Oren Etzioni of the Allen Institute for Artificial Intelligence, a non-profit research group.
The cloud providers are trying to fill the gap by offering consulting services. Google has opened an “Advanced Solutions Lab” that is part consulting service, part tech bootcamp. Whole teams from client companies can come to acquire machine-learning skills and build customised systems alongside Google engineers. Courses typically last from four weeks to several months. Demand has been “overwhelming”, says Vats Srivatsan of Google Cloud, who is now hoping to roll this out much more widely. That is a new departure for tech firms, which in the past have been strong on technical infrastructure but light on people.
The cloud providers will increasingly compete with management consultancies, which charge fat fees for helping clients navigate technological disruption. “The Googles, Amazons and Microsofts of the world may take over from the McKinseys, Boston Consulting Groups and Bains,” says Roy Bahat of Bloomberg Beta, a venture-capital firm. “Consultancies are built for two-by-two matrices. AI’s matrices are a million by a million.” In this race, consultancies with deep expertise in data and technology are better placed than those that focus on general strategy.
The generalists know they are vulnerable. McKinsey has been investing heavily to beef up its expertise in data, for example by buying QuantumBlack, an advanced-analytics firm, for an undisclosed sum in 2015. But many clients seek advice direct from tech firms, which are themselves pioneering users of AI. “All consultants do is listen to you and tell you back what you’ve already told them,” says Morag Watson of BP, an oil giant.
IBM is trying to bridge the gap between the tech wizards and the conventional consultants. “People think this will go the way the digital and mobile revolutions went. I would argue the opposite. If people get their AI right, it’s a great way to extend their incumbent advantage,” says David Kenny, the boss of Watson, IBM’s AI offering. Watson has been heavily marketed on television and enjoys strong name recognition, aided by its victory over human contestants in a game of Jeopardy in 2011. But its bespoke solutions for clients take lots of time to develop, running up hefty bills for consulting hours. “Watson is a branding concept that’s being portrayed as a product,” says Tom Siebel of C3 IOT, an AI startup. “You can’t easily buy it, and you can’t install it.”
IBM also suffers from the same problem as any tech firm other than Google, Amazon and Microsoft: it finds it hard to get hold of the best talent. None of the top doctoral candidates in AI goes to work for IBM, says Mr Domingos of the University of Washington. The old saying that “nobody ever got fired for buying IBM” may no longer apply in the AI era.
Startups, too, are hoping to jump on the AI bandwagon. Many offer services like helping clean up and label data, and take on specific tasks that large tech firms are not yet offering, like helping firms recruit, scan job descriptions and improve customer service. For large companies it makes sense to outsource most of their AI work, except where it directly affects their strategic edge. For example, BP would not want to build AI tools to automate back-office or HR functions, but it would want to develop its own AI system for interpreting seismic imaging to detect oil, says Ms Watson.
If companies want to get products rolled out quickly, they have to work with multiple vendors, says Mr Lippert of MetLife. That may be good for startups, which can be nimble. But the incumbent tech firms’ size, computing infrastructure, proprietary data and balance-sheets give them an unassailable advantage. “Right now everyone thinks they can win. The field will become considerably less democratic,” predicts Martin Reeves of Boston Consulting Group. Having used AI to boost their own fortunes, the incumbents will move on to selling the technology to customers who may become AI-fuelled giants in their own right.