TWO letters can add up to a lot of money. No area of technology is hotter than AI, or artificial intelligence. Venture-capital investment in AI in the first nine months of 2017 totalled $7.6bn, according to PitchBook, a data provider; that compares with full-year figures of $5.4bn in 2016. In the year to date there have been $21.3bn in AI-related M&A deals, around 26 times more than in 2015. In earnings calls public companies now mention AI far more often than “big data”.
At the heart of the frenzy are some familiar names: the likes of Alphabet, Amazon, Apple, Facebook and Microsoft. A similar, though less transparent, battle is under way in China among firms like Alibaba and Baidu. Several have put AI at the centre of their strategies. All are enthusiastic acquirers of AI firms, often in order to snap up the people they employ. They see AI as a way to improve their existing services, from cloud computing to logistics, and to push into new areas, from autonomous cars to augmented reality. Many observers fear that, by cementing and extending the power of a handful of giants, AI will hurt competition. That will depend on three open questions, involving one magic ingredient.
AlphaGone
The tech giants certainly have big advantages in the battle to develop AI. They have tonnes of data, oodles of computing power and boffins aplenty—especially in China, which expects to charge ahead. Imagine a future, some warn, in which you are transported everywhere in a Waymo autonomous car (owner: Alphabet, parent of Google), pay for everything with an Android phone (developer: Google), watch YouTube (owner: Google) to relax, and search the web using—you can guess. Markets with just a handful of firms can be fiercely competitive. A world in which the same few names duke it out in several industries could still be a good one for consumers. But if people rely on one firm’s services like this, and if AI enables that firm to predict their needs and customise its offering ever more precisely, it will be burdensome to switch to a rival.
That future is still a long way off. AI programs remain narrowly focused. Moreover, the ability of the incumbents to perpetuate their advantages is made uncertain by three questions.
The most important is whether AI will always depend on vast amounts of data. Machines today are usually trained on huge datasets, from which they can recognise useful patterns such as fraudulent financial transactions. If real-world data remain essential to AI, the tech superstars are in clover. They have vast amounts of the stuff, and are gaining more as they push into fresh areas such as health care.
A competing vision of AI stresses simulations, in which machines teach themselves using synthetic data or in virtual environments. Early versions of a program developed to play Go, an Asian board game, by DeepMind, a unit of Alphabet, were trained using data from actual games; the latest was simply given the rules and started playing Go against itself. Within three days it had surpassed its predecessor, which had itself beaten the best player humanity could muster. If this approach is widely applicable, or if future AI systems can be trained using sparser amounts of data, the tech giants’ edge is blunted.
But some applications will always require data. How much of the world’s stock of it the tech giants will end up controlling is the second question. They have clout in the consumer realm, and they keep pushing into new areas, from Amazon’s interest in medicine to Microsoft’s purchase of LinkedIn, a professional-networking site. But data in the corporate realm are harder to get at, and their value is increasingly well understood. Autonomous cars will be a good test. Alphabet’s Waymo has done more real-world testing of self-driving cars than any other firm: over 4m miles (6.5m kilometres) on public roads. But established carmakers, and startups like Tesla, can generate more data from their existing fleets; other firms, like Mobileye, a driverless-tech firm owned by Intel, are also in the race.
The third question is how openly knowledge will be shared. The tech giants’ ability to recruit AI expertise from universities is helped by their willingness to publish research; Google and Facebook have opened software libraries to outside developers. But their incentives to share valuable data and algorithms are weak. Much will depend on whether regulations prise open their grip. Europe’s impending data-protection rules, for example, require firms to get explicit consent for how they use data and to make it easier for customers to transfer their information to other providers. China may try to help its firms by having negligible regulation.
The battle in AI is fiercest among the tech giants. It is too early to know how good that will be for competition, but not to anticipate the magic ingredient that will determine the outcome: the importance, accessibility and openness of data.
——
联系客服