The consumer welfare implications associated with the use of algorithmic decision tools, artificial intelligence, and predictive analytics #FTC-2018-0056-D-0028

Submission Number:
FTC-2018-0056-D-0028
Commenter:
Salil Mehra
State:
Pennsylvania
Initiative Name:
The consumer welfare implications associated with the use of algorithmic decision tools, artificial intelligence, and predictive analytics
Please see my full comment in the attached document. An edited version is below in the comment field. Comment re Topic 9 The consumer welfare implications associated with the use of algorithmic decision tools, artificial intelligence, and predictive analytics. by Salil K. Mehra, Charles Klein Professor of Law and Government Temple University, James E. Beasley School of Law My name is Salil Mehra and I am a law professor at Temple Universitys law school in Philadelphia. In 2014, I was the first author to publish legal scholarship on the challenge that algorithms posed for antitrust law: De-Humanizing Antitrust: The Rise of the Machines and the Regulation of Competition, Temple Univ. Legal Studs. Research Paper No. 2014-43 (Sept. 3, 2014), available at https://papers.ssrn.com/sol3/Delivery.cfm/SSRN_ID2520232_code497260.pdf?... (later published in hard copy as Antitrust and the Robo-Seller: Competition in the Time of Algorithms, 100 MINN. L. REV. 1323 (2016)). My comments draw substantially on this paper and my subsequent work, and I would be happy to discuss them more. I am submitting this comment on the issue of algorithms and antitrust law, including algorithmic collusion, in response to the FTCs call for comments in connection with its hearings on Competition and Consumer Protection in the 21st Century. My opinion can be divided into 3 key points. First, algorithmic competition can be a tremendous boon to consumer welfare. Regarding static welfare, it can reduce inefficient allocation by providing better matching between supply and demand. More importantly, the dynamic improvements through innovation especially the potential to increase output at less cost could be tremendous. Because of these benefits, we should not let overly zealous enforcement chill algorithmic competition. As a result, it would be potentially very harmful to consumer welfare to punish algorithmic competition without a better understanding of its impacts. There has been scaremongering based on fears that artificial intelligence will somehow destroy competition as we know it, and that corporate use of mass data collection may, for consumers, drive a descent from king to slave on the data treadmill. See Ezrachi and Stucke, Virtual Competition (2016) ((start of section entitled Final Reflections). However, these fears are premature, first off because technological development is still far from creating some sort of autonomous algorithmic cartel robot. Moreoever, the truth is that algorithmic competition promises real efficiencies and synergies for producers that, in competitive markets, can be passed on to consumers, as well as large improvements in service and product quality. As an example, consider the massive consumer benefits in increased supply, better pricing and improved quality that Ubers matching and pricing algorithm has provided to urban commuters notwithstanding claims in federal district court that Ubers algorithm may foster collusion. Premature antitrust enforcement regarding algorithmic competition could inflict real harm to static and dynamic welfare. Second, while it is true that, all things being equal, in a world with algorithmic competition, Cournot models suggest that parallel/interdependent pricing may become easier for firms to carry out, that alone does not justify imminent enforcement in this area, or new legislation, at this time. Algorithmic competition, or robo-selling, involves using mass data collection, computer-driven algorithmic processing, and automated pricing to digest and respond to market changes at high speed. Drawing a legally enforceable line between parallel pricing that, without more, could be actionable as tacit collusion on the one hand, versus a benign normal response to observable market prices, on the other, will be impossible to do for the foreseeable future without substantial error costs. Third, and finally, algorithmic collusion should nonetheless be a focus for enforcement based on traditional antitrust theories. Antitrust enforcers across the political, ideological and academic spectrum agree on the desirability of deterring explicit agreements to fix prices and restrict output. In cases such as United States v. Topkins and United States v. Aston, we have seen geographically-distant conspirators use algorithm-driven software to fix prices on differentiated products with infrequent sales. In particular, software that fosters anticompetitive collusion could open the possibility of anticompetitive collusion to a wider range of firms due to lower transaction costs over distance and reduced calculation and management cost of cartels. Also, people who ordinarily would not consider committing antitrust offenses IRL may be more likely to do so if off-the-rack pricing software that makes price fixing only a click away.