Ai and Interdependent Pricing: Combination Without Conspiracy?

• Jurisdiction: United States,Federal
• Author: By Joshua P. Davis and Anupama K. Reddy
• Publication year: 2020
• Citation: Vol. 30 No. 2

AI AND INTERDEPENDENT PRICING: COMBINATION WITHOUT CONSPIRACY?

By Joshua P. Davis and Anupama K. Reddy¹

I. INTRODUCTION

Artificial Intelligence (AI) holds the potential to change our landscape in many ways. Some of them are positive. AI may empower automobiles to drive themselves, greatly reducing traffic accidents² and freeing up our time to pursue other tasks. It may be able to detect cancer—or its absence—far better than radiologists can, saving lives and avoiding unnecessary surgeries.³ It can help us to identify remedies for viruses and perhaps greatly accelerate our development of vaccines. The list goes on and on.

But AI also has the potential to cause great harm. One place that harm may take place is in the marketplace. AI is a great player of games. It has defeated the world chess champion.⁴ It has done the same to the world champion of Go,⁵ a game with even more permutations. Market participants may be able to harness AI's game-playing power to cause market distortions to their benefit. That possibility and how to deal with it are the topics of this article.

More specifically, this article will address civil liability for interdependent pricing⁶ without an agreement as facilitated by AI. There is perhaps no area of greater consensus in antitrust law—and competition law more generally—than that competing sellers should not be permitted to agree to elevate their prices above competitive levels.⁷ Doing so is a crime in the U.S., as well as conduct that is treated as per se illegal under our civil antitrust laws.⁸

[Page 1]

In contrast, courts often hold that the law allows sellers to act independently in a way that approximates—or even replicates—the effects of a price-fixing conspiracy.⁹ Particularly when there are only a small number of sellers in a market, they may all manage to charge prices above competitive levels. True, each one could gain market share by lowering its prices. But the others would likely respond in kind—that is, retaliate—and the ultimate result would be to decrease all of their profits. So they might each price at levels similar to a monopolist and none may break ranks, benefiting all of them. This approach is sometimes called "interdependent pricing" because the strategy works for one seller only if other sellers respond in kind; the pricing strategy of each seller depends on the pricing strategies of the others. The same conduct is also sometimes called "conscious parallelism."¹⁰

Interdependent pricing can cause the same kinds of harms as price-fixing conspiracies.¹¹ It transfers wealth from buyers to sellers, decreases output, and results in deadweight loss (sales of a product or service that fail to occur because prices are elevated above competitive levels). How significant these harms are depends on which notion of efficiency one adopts. A focus on consumer welfare would include all of those effects as significantly harmful.¹² In contrast, a focus on total welfare might suggest that a wealth transfer from a buyer to a seller—all else equal—is not necessarily bad. Regardless, there is general agreement that interdependent pricing, to the extent it has the same consequences as a price-fixing agreement, is anticompetitive.

[Page 2]

Where there is far less agreement is whether the law should prohibit interdependent pricing as it does price-fixing agreements and, relatedly, how to draw the line between the two. In principle, an agreement to fix prices is illegal while independent conduct that has the same effect is not. But it is difficult to articulate how to distinguish them, if they are in fact distinct, a problem made more challenging in practice by various legal doctrines. One of them allows plaintiffs to rely on circumstantial evidence¹³ to establish the existence of an agreement and another recognizes tacit agreements as sufficing for liability. In part because of similarities in effects, and the difficulties of line-drawing, theorists have suggested at times that the law should condemn interdependent pricing as it does price fixing. Richard Posner endorsed that approach over five decades ago,¹⁴ although, as he has acknowledged, he was not alone in doing so and he has since changed his mind.¹⁵ Louis Kaplow recently wrote an excellent book offering reasons to believe Posner may have been right in the first place and may be wrong now—a policy conclusion, he concedes, that "must be viewed as quite tentative."¹⁶

The thrust of Kaplow's argument is that from an economic perspective what we care about is whether prices are elevated above competitive levels, not the communications that we often require to prove the existence of an agreement.¹⁷ As a result, requiring a conspiracy can end up focusing courts on issues that are a distraction at best and counterproductive at worst. We will explore in a bit some of the main points he makes. Note, however, that Kaplow's analysis assumes markets remain much as they have in the past. That assumption may not be realistic.

Enter AI. Two features of AI are particularly important for present purposes. First, AI may greatly increase the harms from interdependent pricing. AI holds the potential to enable such pricing to succeed in circumstances where it currently would likely fail and to raise prices closer to monopoly levels when it does succeed. Second, AI may make interdependent pricing immune from legal liability. A reason is that for the foreseeable future AI will likely be incapable of forming any intent, a requirement for civil liability. Let's take these points one at a time.

[Page 3]

AI may make interdependent pricing far more prevalent and far more effective when it occurs.¹⁸ To understand this point, it is important to note a dynamic that lies at the heart of antitrust laws. It is what one might call a collective action problem or a coordination problem. But it isn't really a problem; it's a solution. The difficulty market actors have in acting collectively—in coordinating their behavior—is salutary. The collective action problem ordinarily involves prices (or, what generally amounts to the same thing, output). One of the main mechanisms of antitrust is to encourage sellers in the same market to compete for market share by lowering their prices. That competition ultimately benefits consumers. But it harms the competitors. They would maximize their profits by charging higher prices and producing less. The profit-maximizing model for the competitors would be to charge the same amount and produce the same quantity as would a single seller with complete control of the market, a monopolist. The competitors would do best if they collectively emulate the behavior of the monopolist. But that is not practical for them to do in a competitive market. If there are a large number of sellers of equivalent products—say, pencils—with equivalent cost structures, they generally cannot effectively resist competition. If one of the sellers charges more than the competitive price for pencils, another seller will sell at a lower price and steal market share. The sellers' failure to mimic a single seller of pencils through collective action or coordination redounds to the benefit of society as a whole. Or so antitrust theory runs.

The above analysis explains much of antitrust law. It provides a reason that antitrust law will prohibit mergers and acquisitions if a small number of firms control a large percentage of a market. A merger or acquisition will further consolidate control of the market, threatening competition. On the other hand, if there are a large number of firms in a market—each with a relatively small market share—a merger or acquisition is unlikely to harm competition. We rely on the difficulty that large numbers of actors have in coordinating their behavior to ensure well-functioning markets. So the collective action problem in markets is a feature of antitrust law, not a bug. Ideally, the problem would be insurmountable. Firms would then have to compete on price, or quality, or the like—to "compete on the merits," to coin an antitrust phrase.

AI poses a threat to antitrust law that derives from its ability to overcome collective action or coordination problems. As noted above, there is substantial evidence that firms run by human beings will not be able to emulate monopolists if the firms are sufficiently numerous and sufficiently similar to each other in key respects, such as their costs, and if each has a sufficiently low market share and acts independently. The collective action problem is too formidable. But AI holds the potential to solve previously insoluble problems. AI could be programmed to maximize profits in pricing. And it may be able to act sufficiently quickly and subtly to coordinate with even large numbers of other firms also operated by AI. Just as a small number of firms run by human actors can compete in a way that will collectively benefit them by mimicking a monopolist—or approximating monopolist prices—so may a large number of firms run by AI. Indeed, it is possible that an entrepreneur could develop pricing AI that has precisely this effect—especially if all sellers in a market rely on it—and that the sellers in a market could all buy the pricing AI for that reason. Voila—independent action leads in effect to price fixing.¹⁹ The result could be that many markets—even if they contain large numbers of firms, each with relatively small market shares—may look a lot like they are dominated by a single seller. And the prices the sellers charge in those markets may be closer to monopoly prices than would occur if human beings were in charge of pricing decisions.

[Page 4]

There is no precise, generally accepted definition of AI.²⁰ It includes a suite of technologies capable of self-learning and other behavior that resembles human intelligence.²¹ AI can solve problems or make automated decisions for tasks that are typically thought to require higher order cognitive processes, including vision, spatial reasoning, and conceptualization.

...