Tacit Algorithmic Collusion: Are WE Ready to Face the Future?
Last year, at a Senate subcommittee hearing, Senator Richard Blumenthal asked Assistant Attorney General Makan Delrahim of the Department of Justice (“DOJ”) whether the DOJ needed more resources to “scrutinize the fairness of algorithms and artificial intelligence systems.”1 Rather than answering that specific question, Delrahim noted that companies could use algorithms in an anticompetitive way.2 He added that the anticompetitive use could take a couple of forms, and that one of those forms involves two potential competitors using the same algorithm “as a way of effectuating a price-fixing scheme.”3
In fact, in January 2019, Daniel William Aston, a former e-commerce executive, pleaded guilty to the algorithmic price-fixing conspiracy that took place between 2013 and 2014.4 Aston and his competitors agreed to adopt specific algorithms to coordinate the prices of certain posters sold in the United States through Amazon Marketplace.5 In announcing Aston’s guilty plea, Delrahim stated: “The Division and its law enforcement partners are committed to investigating and prosecuting individuals, wherever located, who collude through new and sophisticated means, including algorithmic pricing software.”6
Cases like Aston’s, wherein algorithms are simply used as a tool for the execution of a price-fixing conspiracy, are easily prosecuted under the Sherman Act’s prohibition on anticompetitive agreements.7 The Federal Trade Commission (“FTC”) Commissioner Maureen Ohlhausen suggested a simple test that captures such easy cases: “If it isn’t ok for a guy named Bob to do it, then it probably isn’t ok for an algorithm to do it either.”8
However, what about the possibility of rival firms independently coding their algorithms to collude by programming their algorithms to punish their rivals for setting low prices, and it is through that threat of punishment that high prices are sustained?9 In that scenario, an algorithm, through trial and error, learns not to lower a price, even though it would yield a short-term profit as it anticipates a retaliatory response by other algorithms that would lower their prices.10 Do the algorithms that collude by using the reward-punishment scheme engage in concerted action in violation of the Sherman Act?
Under the Sherman Act, conscious parallelism is perfectly lawful. For instance, it is lawful for the owners of the two neighboring gas stations to decide not to compete and maintain supra-competitive prices as long as they come to this decision independently by realizing that they both would be better off not engaging in a price war. However, if the independent decision not to compete is reinforced by some concerted actions, for instance, a wink that serves as a commitment to their scheme, their decision can be regarded as a tacit agreement that violates the Sherman Act.
As the name indicates, tacit agreement refers to an agreement which is implied or indicated, but not explicitly expressed.11 To assist in separating conscious parallelism from tacit agreement, lower courts have endorsed the concept of “plus factors,” such as circumstantial factors that go beyond mere conscious parallelism, from which an agreement can be indirectly inferred.12
Michal S. Gal, a President of International Association of Competition Law Scholars (“ASCOL”), argues that designing algorithms that facilitate collusion falls within the definition of an unlawful tacit agreement.13 Gal explains that programming algorithms to collude is the avoidable action that goes beyond mere conscious parallelism.14
To be sure, studies show that designing algorithms with the capability to tacitly collude in a realistic market environment is a challenging technical problem.17 Nevertheless, any technological problem will eventually find its own solution. The antitrust authorities, in turn, should also be prepared to tackle challenges posed by sophisticated pricing algorithms. Today, as the Acting Chair of the FTC Maureen K. Ohlhausen has noted, “The innerworkings of these tools [algorithms] are poorly understood by virtually everyone outside the narrow circle of technical experts that directly work in the field.”18 Perhaps the use of congressional resources to catch up with technical aspects of algorithms is not a bad idea for the antitrust authorities.
John Eggerton, Delrahim: Criminal Case Against Anti-competitive Search Algorithms Coming, Multichannel News (Oct. 4, 2018), https://www.multichannel.com/news/delrahim-criminal-case-against-anti-competitive-search-algorithms-coming. [https://perma.cc/3GXR-AQ2H]↩
Press Release, Dep’t of Justice, Former E-Commerce Executive Pleads Guilty to Price Fixing; Sentenced to Six Months (Jan. 28, 2019), https://www.justice.gov/opa/pr/former-e-commerce-executive-pleads-guilty-price-fixing-sentenced-six-months. [https://perma.cc/E4QX-JBCN]↩
See 15 U.S.C. § 1 (1988)↩
Maureen K. Ohlhausen, Should We Fear The Things That Go Beep In the Night? Some Initial Thoughts on the Intersection of Antitrust Law and Algorithmic Pricing, Fed. Trade Comm’n, 1, 10 (May 23, 2017), https://www.ftc.gov/system/files/documents/public_statements/1220893/ohlhausen_-_concurrences_5-23-17.pdf [https://perma.cc/9SM4-6T9F]↩
See Joseph E. Harrington, Developing Competition Law for Collusion by Autonomous Artificial Agents, 14 J. of Competition Law & Econ. 331, 350 (2018).↩
Michal S. Gal, Algorithms As Illegal Agreements, 34 Berkeley Tech. L.J. 67, 100–01 (2019).↩
Id. at 67.↩
Ai Deng, What Do We Know about Algorithmic Tacit Collusion, 33 Antitrust 88, 89 (2018).↩
Id. at 90.↩
Ai Deng, Cover Story, Litigation Practicean Antitrust Lawyer’s Guide to Machine Learning, 32 Antitrust ABA 82, 82 (Spring, 2018)↩