Generative AI and market power: What role for antitrust regulators?

Many thanks to Tommaso Valletti and Cristina Caffarra for organising this spectacularly good panel on antitrust and generative AI. It’s well worth watching in full (see below), but I’ve tried to capture some key points.

First: Prof Daron Acemoglu (MIT, shown above) said the problem of the developing foundation model duopoly is it will shape the direction of innovation in harmful ways, both in large-scale elimination of good jobs, and erosion of democratic capacity. Some of his points I particularly liked:

  • “The direction of technology is malleable but it’s also a societal choice” ??
  • “If we want alternatives, it is highly unlikely to come from a monopolistic or duopolistic market” ????
  • Use antitrust, including breakups, to achieve a more pro-human and democratic outcome ??????
  • “We need a new type of interoperability approach, as well as how we compensate and encourage more creation of creative data. And provide more explicit incentives… for socially valuable practices. And tax policies to create room for more business models, like an ad tax” ??

Next: Andreas Mundt (head of the German Cartel Office) is concerned AI will strengthen existing Big Tech incumbents; that conditions for investment and development are worse in Europe than US; and that competition tools have limited potential, targeting tying/envelopment/mergers rather than (for example) cooperation (such as Microsoft’s deal with OpenAI).

Mundt is concerned that AI development is too early to properly regulate via the EU’s AI Act. That’s why I think adding generative AI as a Digital Markets Act core platform service via its market investigation mechanism in the medium term could be a good way forward.

Next, Prof Josh Gans, who thinks the role of competition economics is to accelerate the uptake of new technology ? I’m much more concerned about the types of innovation which emerge and are adopted, and their impact as a whole on society. Gans is more optimistic about the emergence of new foundation model providers, like Anthropic AI and Cohere, rather than assuming Microsoft/OpenAI and Google are going to dominate this market due to high investments to date.

Finally, the US Department of Justice’s Susan Athey: Tech shifts provide a glimmer of opportunity for changes in industry leadership, but there’s a now well-established playbook for how incumbents try to stop that — buying up nascent competitors, increasing switching costs, using long-term contracts…

Athey expects much change to come industry by industry, so it will be important for competition authorities to stay on top of these changes, and watch out for new bottlenecks, such as staff who can customise and build on foundation models. Andreas Mundt agrees, but points out this will not be true of “the Internet” (the digital economy of today) because it is already so dominated by a few powerful firms.

Gans says we can’t predict the directions of innovation in AI; there will be a monopolist in a decade, because they will solve some big new problem, but we don’t know who they will be. Mundt says we can’t wait around to address many of the emerging harms as they are too significant.

Tommaso Valletti asked: is it a healthy model for our future lives, to have waves of monopolies, and for society to be dominated by a new “exciting” AI monopoly, as Gans put it? Valletti concluded: it’s exciting for the monopolist ? Athey added data and scale will always be important for incumbents, but we are now enabling data markets which weren’t possible before, to extract learning from one firm. She is cautiously optimistic about avoiding future waves of monopolies by using antitrust tools “earlier and more assertively”.

Grazie mille to the speakers, and to Cristina Caffara for her great chairing!