The notorious Cartels characterised as the “supreme evil of antitrust,”[1] are a group of firms who have agreed explicitly among themselves to coordinate their activities, including price fixing,[2] to raise prices at the detriment of consumers, and competition closure against competitors. This ultimately affects innovation since firms will not bother to innovate their products for profit, if they could merely agree to charge higher prices. Under European Union Law, cartels are prohibited under Article 101 TFEU,[3] requiring a form of agreement or communication. Absent an agreement or collusive practice however, mere parallel conduct is not prohibited[4] since it can result from independent and rational behaviour.[5]
Nevertheless, could the introduction of pricing algorithms result in a new form of a price fixing cartels, specifically AI Cartels? Nowadays, dominant market sellers delegate their product’s pricing role to autonomous pricing algorithms instead of discussing themselves on their products price in accordance with the market trend. By using the pricing algorithm, they effectively allow significantly quicker responses to changing prices in various sectors, such as taxi fares, sports tickets, or hotel rooms with perishable goods, and parallelly reduce labor costs by removing the paid human price setter. However, as it turns out, the algorithms in the honest and humble pursuit to bring the most profit to their firms, may conspire with each other to price fix and significantly raise prices.
How? The algorithms in their quest for profit maximisation agreed with each other, on an incremental basis with a tit-for-rat strategy to increase the book’s price so no firm will have a cheaper alternative and thereby consumers will be left with no other choice. This is exactly what happened with a book by Peter Lawrence entitled ‘The Making of a Fly,’ which had two copies reach the price of $23,698,655.93 on Amazon, due to the automatic pricing algorithms lacking a built-in sanity check on their produced prices.[6] Although the “fly” book example represents a clear algorithmic anomaly, and frankly that no sane buyer would exist, what if the price increment is more subtle, shifting from 30 euros to 60? In such cases, the change might be perceived as a natural market progression, causing the same harm as a mainstream cartel with the “supreme evil of antitrust” left roaming free.
The mechanism
Big Data includes any information relating to an identified or identifiable individual (data suspect).[7] As the CMA explains, Big Data consists of competing firms’ prices, customer information, firms costs, such as production storage and fulfilment.[8] Importantly, the utility and significance of Big Data have surged alongside the emergence of Big Analytics: the ability to design algorithms that can access and analyse vast amounts of information.[9] This is the baseline which allows a computer-based system to analyse input and give out the necessary output. Further, advancements in AI and machine learning enable algorithms to discern relevant data for objectives like profit maximization. Q-learning, a subspecies of machine learning, attempts to maximise total discounted profit over time, using ‘trial- and-error’ to interact with its environment to learn the optimal pricing policy.[10] A step further, there are Artificial Neural Networks (ANN’s) systems which are based on collections of connected units that receive, process, and transmit signals to other connected neurons.[11] In essence, these algorithms learn over time, through trial and error to set prices and achieve their core objective of price maximization. As sellers and buyers migrate to an online world, and as technology, big data, and big analytics reach new highs, the use of algorithms to conduct pricing decisions is inevitable.
Current status of AI cartels
The case of the Wood Pulp Cartel [1993] ECR I-1307 of the ECJ established a clear distinction between legal tacit collusion and illegal concerted practice, emphasizing the necessity of an agreement[12] or at least a “concurrence of wills.”[13] Even if legal provisions are extended to cover such conduct, imposing liability in the AI cartel context proves infeasible. As Mehra puts it, “there are three choices in attributing responsibility: to the robo-seller itself, to the humans who deploy it, or to no one.”[14] Since, a machine is not a legal person, and the humans who deploy it do not even have an anti-competitive intention or knowledge, asserting liability not possible. The ECJ is a stranger to this new dilemma as no cases of AI cartel have been recorded thus far, and therefore did not provide any assistance on how AI cartels are dealt with when put on the stand. Of course, when human cartelists use algorithms as a mode of communication to price fix, then the ECJ can follow its normal procedure. Regarding autonomous algorithms however, the issue escapes the legal realm. Consequently, the effects of the colluding algorithms, are hidden, unmonitored, and even worse, ignored.
Looking Forwards
Still, let us not consider this as an unavoidable AI crisis as solutions do exist, notable being a Market Investigation scheme. As the OECD stated, “when there are signs that the market is not functioning well, but there are no indications of any coordination among the market players, competition agencies may decide to engage in market studies or sector inquires in order to understand why the market is failing and to identify possible solutions.”[15] Once a market study is performed and a case of an AI cartel is identified, remedial actions can be imposed to counter this issue which takes the form of a behavioural remedy, requiring the undertaking concerned to perform or refrain certain acts relating to its behaviour on the market.[16] A notable remedy is obliging the concerned undertakings in programming their algorithm to avoid collusive price conduct – a programming remedy. For instance, authorities could order firms to program their algorithms to play competitive instead of cooperative games.[17] Programming restrictions to counter price fixing could also be facilitated ex ante by targeting the creators of the pricing algorithms, to build restrictions on how the AI operates, yet significant care must be upheld so that firms would not be pushed away from a technology that also has efficiency gains.
All things considered, the concept of AI cartels no longer exists in the realm of legal sci-fi, and it is a matter of time before we see a significant increase in pricing algorithms forming agreements with each other. This is a reminder that whilst we build new technologies which elevate our welfare, we should regulate accordingly before we are in a position where we are creating more problems than we solve.
This article was written by Iraklis Kyprianou, a talented trainee lawyer at the firm, who possesses detailed knowledge of the subject, having focused his university dissertation on this area.
[1] Verizon Communications v Law Offices of Curtis V. Trinko, 540 US 398 (2004), 408.
[2] Organisation for Economic Co-operation and Development, “Recommendation of the Council Concerning Effective Action Against Hard Core Cartels” (C (98)35/FINAL, 1998).
[3] Consolidated Version on the Treaty on the Functioning of the European Union (2012) OJ C326/49.
[4] Cases C-89, 104, 114, 116–117, and 125–129/85, Re Wood Pulp Cartel: Ahlström Oy v Commission [1993] ECR I-1307, para 71.
[5] Peter D. Camesasca, Laurie-Anne Grelier, “Close Your Eyes”? Navigating the Tortuous Waters of Conscious Parallelism and Signalling in the European Union” (2016) J.E.C.L. & Pract. 599–607, 601.
[6] Michael Eisen, “Amazon’s $23,698,655.93 Book about Flies” (It Is NOT Junk, 22 April 2011) <http://www.michaeleisen.org/blog/?p=358> accessed 4th March 2024.
[7] Ariel Ezrachi and Maurice E. Stucke, Virtual Competition: The Promise and Perils of the Algorithmic Driven Economy (HUP 2016), 15.
[8] Competition and Markets Authority, Pricing algorithms: Economic working paper on the use of algorithms to facilitate collusion and personalised pricing (CMA94, 2018), <https://assets.publishing.service.gov.uk/media/5bbb2384ed915d238f9cc2e7/Algorithms_econ_report.pdf> accessed 15th April 2024, 15.
[9] Ariel Ezrachi and Maurice E. Stucke, Virtual Competition: The Promise and Perils of the Algorithmic Driven Economy (HUP 2016), 15.
[10] Emilio Calvano and others, “Artificial Intelligence, Algorithmic Pricing and Collusion” (2020) 110 AER 3267-3297, 3295.
[11] Competition and Markets Authority, Pricing algorithms: Economic working paper on the use of algorithms to facilitate collusion and personalised pricing (CMA 94, 2018). <https://assets.publishing.service.gov.uk/media/5bbb2384ed915d238f9cc2e7/Algorithms_econ_report.pdf> at [12].
[12] Re Wood Pulp Cartel: Ahlström Oy v Commission (Cases C-89, 104, 114, 116-117, and 125-129/85) [1993] ECR I-1307.
[13] Case T-41/96 Bayer AG v Commission [2000] ECR II-3383, para 69.
[14] Salil K. Mehra, “Antitrust and the Robo-Seller: Competition in the Time of Algorithms” (2015) 100 Minn. Law Rev. 1323-1375, 1366.
[15] Organisation for Economic Co-operation and Development, “Algorithms and Collusion: Competition Policy in the Digital Age” (OECD 2017) at 40.
[16] Cyril Ritter, “How Far Can the Commission Go When Imposing Remedies for Antitrust Infringements?” (2016) J.E.C.L. & Pract. 587–598, 595.
[17] Francisco Beneke and Mark-Oliver Mackenrodt, “Remedies for algorithmic tacit collusion” (2021) 9(1) Journal of Antitrust Enforcement 152–176, 170.
