ANALYSIS: As Pricing Bots Flex New Muscle, Antitrust Watches AI

It’s tempting to ignore the breathy prognostications about generative artificial intelligence (GAI)—the technology is new, and the noise to signal ratio is still high. But antitrust enforcers are already taking the interface of AI and competition very seriously.

DOJ’s antitrust chief, Jonathan Kanter, has an AI and data team called “Project Gretzky,” for example—a reference to the hockey legend’s quip that the key is to skate to where the puck will be. Kanter, and many other enforcers, see that the AI industry is heating up, and they’re preparing for competitive problems to follow.

Antitrust circles are generally focused on two issues: keeping the nascent GAI industry competitive, and how uses of AI may impact competition in other markets. There’s already something of an arms race brewing between enforcers and industry to keep up with where the tools are headed and how to harness them. 2024 will see that race gathering momentum.

What’s New?

GAI is artificial intelligence that creates some form of original content. It might be text, images, code, or audio based on training the program receives using a selected data set, and on prompts from a user.

Many industries have rapidly embraced open GAIs and are experimenting with harnessing them. What’s new about GAI is that it balloons the promise and risk of existing tools like pricing algorithms, and presents some novel potential competitive shocks.

Pricing on Steroids

The glaring example of potential conflict between antitrust and AI is pricing algorithms. To the extent that these programs permit or facilitate price fixing or bid rigging among competitors, they can foster violations.

There’s never been a more powerful tool for policing a cartel, for example. GAI can scan vast amounts of data, like websites and catalogs for competing products, and summarize descriptions or group products for price comparison. Cartelists can find and punish cheating in real time across larger markets than ever before.

GAI can also function as the “hub” in a hub-and-spokes conspiracy, moving between competitors’ pricing systems and coordinating upward movement in prices to look independent or spontaneous. For example, a GAI that checks competitors’ algorithms and generates different triggers that those algorithms want to see to move price upward could effectively coordinate an industrywide price.

That represents a substantial threat, because intent for antitrust violations is difficult to discern in digital markets. Proving tacit collusion through algorithms to the high standard that courts demand is likely a future challenge for regulators. High prices may be evident, but the how and why may prove very hard to show. In fact, if uniform, monopoly pricing becomes a feature of many markets, or price discrimination is widely perfected, antitrust could face a crisis of identity.

Fire With Fire?

The potential is far from one-sided, however. AI models may make it easier for enforcers to spot market deviations with no underlying fundamental cause. The technology may also make it easier to ingest and understand complex chat traffic used to coordinate a cartel, including the use of symbols or slang, to help enforcers prove a cartel in court. Regular queries in a suspect market might by themselves alert authorities to tacit collusion when answers shift.

In other words, AI may well become something of an arms race between antitrust violators and enforcers. Heaven knows that enforcers are already gearing up for this particular fight. But if a few powerful players can keep the most daring innovations to themselves, the battle might become one-sided for consumers and enforcers alike.

Threats to the AI Market Itself

There’s a real threat that capture could squelch GAI’s potential for many consumers and businesses. As the FTC pointed out in a blog post, GAI development depends on a set of inputs (especially data, talent, and computing power). If powerful firms can block access to those inputs, they may be able to “dampen or distort” competition in generative AI markets, the FTC said. If generative AI itself becomes an increasingly critical tool, as many predict it will, “then those who control its essential inputs could wield outsized influence over a significant swath of economic activity,” the FTC said.

GAI might then become another weapon to entrench market power in digital markets among existing online behemoths, as the UK’s Competition and Markets Authority pointed out in its report on AI Foundations.

If only powerful companies are able to build the next generation of AI models, then can control access to the models, they can become even more powerful.

On Your Mark ...

Data access may be the biggest hurdle to surmount in developing a competitive GAI market. The incumbent digital dragons built their data troves in a largely unregulated environment that’s not replicable today. Gaining scale in some data sets may be impossible without mandated access; but mandated access often conflicts with privacy concerns.

Access to computing also poses a credible threat to GAI democratization, and enforcers are sharpening their focus. CMA is beginning an investigation into the cloud computing market in the UK, for example, as concerns also mount about concentrated or unstable markets for chips, rare metals, and energy.

So pay attention as enforcers improve their access to—and regulation of—data and computing to keep up with the new challenges GAI may pose. 2024 promises to be the year when everybody scrambles for a good position on the GAI starting line without knowing for sure if the race is a sprint or a marathon.

Access additional analyses from our Bloomberg Law 2024 series here, covering trends in Litigation, Transactions & Contracts, Artificial Intelligence, Regulatory & Compliance, and the Practice of Law.

Bloomberg Law subscribers can find related content on our Antitrust Practice Center, and on our In Focus: Artificial Intelligence page.

If you’re reading this on the Bloomberg Terminal, please run BLAW OUT <GO> in order to access the hyperlinked content, or click here to view the web version of this article.