Elizabeth Warren is the latest politician to become a card-carrying member of the “hipster antitrust” movement by proposing to break up big technology companies. Subscribers to the hipster antitrust philosophy believe that large companies harm society simply by virtue of their bigness, abandoning the core tenant of modern antitrust law that prioritizes consumer welfare in shaping competition policy. Others have aptly explained why such proposals would be a bad idea with serious unintended consequences. However, hipster antitrust policies could have particularly damaging effects on the development of artificial intelligence (AI) in the United States. Warren’s proposal comes during a time when policymakers on both sides of the aisle are vocally stressing the importance of U.S. competitiveness in AI, making it important for policymakers to recognize that these goals are at odds.
The majority of hipster antitrust calls to break up big tech are vague and rely on dubious justification. For example, venture capitalist Roger McNamee wants to see Google “be broken up into eight or 10 different monopolies” because, perplexingly, he is “OK with Google being a monopolist in search…[but] I don’t want them to use it to make their photo search and all these other things.” Scott Galloway, a business professor at New York University, agrees that tech giants should be broken up, arguing that the bigness of companies like Amazon, Google, and Facebook stifles innovation, but fails to articulate how. And others, such as MIT Technology Review editor Konstantin Kakaes, have argued that companies like Facebook should be broken up because they have access to enough data that it could lead to abuses of power.
Breaking up big tech firms along these arbitrary lines would have a significant negative impact on AI development because their bigness is what often allows big tech firms to advance AI. For example, in 2011, Google Brain, the company’s AI research team, developed a system for developing artificial neural networks called DistBelief to create a scalable, distributed platform to advance AI research throughout Google’s many product teams. Over 50 teams across Google and other Alphabet companies quickly adopted DistBelief and developed machine learning applications for products including Google Search, speech recognition, Google Photos, Google Maps, Google Translate, YouTube, and others. Recognizing the broad utility of Distbelief, Google Brain developed a second generation of the system called TensorFlow, designed to be even more scalable and flexible, and made it freely available as open source in November 2015. TensorFlow turned out to be so popular that just eight months later, in June 2016, GitHub, the popular web-based hosting service for computer code, had over 1,500 repositories that reference TensorFlow, only five of which were from Google. Now, TensorFlow is the most popular deep learning framework in the world, beating other frameworks in GitHub activity, arXiv publications, use by developers, and mentions in job descriptions by a wide margin. It is exceedingly unlikely that Google would have had the capability or desire to create such a flexible, widely useful framework for machine learning without the need to meet a wide variety of needs throughout its many lines of business. Because of this scale, they created a product that has significantly transformed and accelerated the AI development landscape. Chopping Google’s various product teams into separate businesses would preclude similar innovations and their spinoff benefits in the future.
To the extent that Senator Warren and others genuinely care about consumer welfare, they should recognize that calls to break up big tech could substantially weaken the ability of the United States to develop and deploy AI, and thus harm consumers and the U.S. economy.