In The Wealth of Nations, Adam Smith famously wrote, “[p]eople of the same trade seldom meet together, even for merriment and diversion, but the conversation ends in a conspiracy against the public, or in some contrivance to raise prices.”[1] Although now nearly 250 years old, Mr. Smith’s observation about merchants and businessmen remains timeless. But is the same statement true of computers? As we near the 2020 election, both parties are considering changes to the antitrust laws in response to seemingly unstoppable growth by marketplace leaders such as Facebook and Amazon. Particularly in light of the COVID-19 pandemic, consumer shopping is increasingly moving to online platforms and away from traditional bricks-and-mortar retailers. Are the antitrust laws – designed to combat the so-called robber barons of the late 19th Century – sufficient to keep up with a changing marketplace?
Amazon and other online merchants increasingly use a pricing model called dynamic pricing. First introduced on a large scale by American Airlines in the 1980s,[2] dynamic pricing utilizes computers to adjust pricing according to real-time factors, such as supply and demand. Today, this model remains dominant in the airline industry: the prices of a seat on a flight may change from day to day depending on number of seats remaining, route demand, and competitors’ pricing. For dynamic pricing to work properly, a company must rely on a sophisticated inventory management system, like those used by airlines and hotels. Today, advances in computer technology have made dynamic pricing widely available to e-commerce companies. These algorithms, in the blink of an eye, scan prices across the web and adjust its sales price accordingly.[3] Can artificial intelligence learn from the market and, without human interaction, conclude that the best strategy to maximize profits is to collude with other AI merchants? The answer, increasingly, is yes.[4] These computers learn to collude by trial and error and without communicating with each other. Does antitrust law have an answer to robots that learn parallel pricing?
A well-known alternative to traditional taxi cabs is Uber Technologies, Inc. (“Uber”).[5] Uber utilizes a form of dynamic pricing that it refers to as a “surge fare.” Surge fares are prices that increase during periods of high demand. The surge pricing calculated as a multiplier of the standard fare; the higher the demand, the larger the surge multiplier.[6] Surge fares will change in real time from moment to moment according to demand. The amount of the surge fare is calculated by Uber’s algorithms, so the drivers and passengers each see the same price during a period of high demand. Uber claims that its surge pricing is simply an application of supply and demand. It incentivizes its drivers, who are independent contractors, to take additional fares, thereby better ensuring that Uber can meet increased demand.
In 2016, an Uber customer brought an antitrust suit against Uber, alleging that surge pricing algorithm created a conspiracy between Uber and its drivers that was managed by Uber’s AI software. Meyer v. Kalanick,[7] 174 F. Supp. 3d 817, 820, 822-823 (S.D.N.Y. 2016). Although the case was ultimately sent to arbitration, where the arbitrator found in favor of Uber, the AI collusion theory caught the attention of Judge Rakoff, who, before sending the case to arbitration, denied Uber’s motion to dismiss. He wrote that the plaintiff had alleged that each Uber driver agreed with Uber to charge certain fares “with the clear understanding that all other Uber drivers are agreeing to charge the same fares.” 174 F. Supp. 3d at 824.
Although the hundreds of thousands of Uber drivers around the world have never once met together, (as Adam Smith might wryly note) this is the “genius” of Uber, Judge Rakoff held. Id. at 825. By utilizing the “magic of smartphone technology,” Uber was able to invite agreements “hundreds of thousands of drivers in far-flung locations. . . . The advancement of technological means for the orchestration of large-scale price fixing-conspiracies need not leave antitrust law behind.” Id. at 825-826.
The theory that computer algorithms can facilitate, or even implement, price-fixing schemes has not gone unnoticed in the world of criminal antitrust. As far as we are aware, there has been only one criminal antitrust case involving a computer algorithm, but in that case, the algorithm was only the mechanism by which a human-led conspiracy was implemented. In 2015, David Topkins, an art dealer from California, pleaded guilty to coordinating with other art dealers to use price-fixing algorithms for the sale of movie posters on Amazon.com.[8] Unlike the price-fixing AI that could operate without human involvement, Mr. Topkins’s case was decidedly more hum-drum: he admitted to agreeing with his competitors to use the algorithm to set the price of the artwork.[9] What is still unclear five years later, despite the explosion of online pricing algorithms and e-commerce, is how enforcers will view machine learning and where does liability stop? Can programmers be held liable for writing pricing algorithms?
The cases at each end of the spectrum are simple. The ones in the middle, however, may not be so easy. In a 2017 white paper from the OECD, the drafters wrote that “[f]inding ways to prevent collusion between self-learning algorithms might be one of the biggest challenges that competition law enforcers have ever faced[.]”[10] Regardless of the scope of the challenge, this issue will increasingly become one that enforcers and retailers can no longer ignore simply by refusing to play.
[1] The Wealth of Nations, Book I, Chapter X (1776).
[2] See R. Preston McAfee and Vera te Velde, Dynamic Pricing in the Airline Industry, Economic and Information Systems (2006).
[3] A Northeastern University research paper found that in 2015, algorithmic sellers covered one-third of the best-selling products offered by third party merchants on Amazon, and that algorithmic sellers’ pricing was 10 times more volatile than human-priced sellers. Le Chen, Alan Mislove, and Christo Wilson, An Empirical Analysis of Algorithmic Pricing on Amazon Marketplace, available at https://mislove.org/publications/Amazon-WWW.pdf.
[4] Emilio Calvano, Giacomo Calzolari, Vincenzo Denicolo, and Sergio Pastorello, Artificial Intelligence, Algorithmic Pricing, and Collusion, available at https://www.ftc.gov/system/files/documents/public_events/1494697/calzolaricalvanodenicolopastorello.pdf.
[5] Uber uses a smartphone application to match private drivers with passengers. The Uber app calculates and collects the fare, which it remits to driver, minus a licensing fee that the driver must pay for the driver’s use of the Uber app. Drivers and passengers cannot negotiate the price of a fare; it is set by the Uber app.
[6] See Uber Help – How Surge Pricing Works, https://www.uber.com/us/en/drive/driver-app/how-surge-works/
[7] 174 F. Supp. 3d 817 (S.D.N.Y. 2016). The plaintiffs sued Travis Kalanick, Uber’s CEO and founder personally in the complaint; however, for ease of reference, the defendant will be referred to in this article as Uber.
[8] United States v. Topkins, CR-15-201, at Dkt. 7 (N.D. Cal. 2015).
[9] Id. at ¶4(c).
[10] Big Data: Bringing Competition Policy to the Digital Era, at ¶84, available at https://one.oecd.org/document/DAF/COMP(2016)14/en/pdf (last accessed February 19, 2017).