CYIL vol. 11 (2020)

MICHAL PETR CYIL 11 (2020) understanding of addressees of competition law, which are undertakings , economic (rather than legal) units responsible for certain conduct; 2 for example, as a parent company and its subsidiary generally work “in tandem” and their conduct is by definition coordinated, they are perceived as a single economic unit, and thus a single undertaking, even though the “unit” is composed of more legal persons. It is therefore the actual conduct of the undertaking that is decisive, and in order to find an undertaking liable for a breach of competition law, it is not necessary to identify a specific natural person who was in fact acting on its behalf. 3 This is, in principal, in line with the general rules on corporate liability, which is based on the presumption that the decision to act in a certain way was adopted by the management of the company, or at least that the management was aware (or should have been aware) of it. This paradigm may however be shifted by the deployment of artificial intelligence, which, in its most advanced form, decides according to rules its “owners” cannot understand, because it has “learned” them on its own (these are known as self-learning algorithms). The question we would like to address in this article therefore is, whether it is possible under the current rules of competition law to input liability for anticompetitive conduct to an undertaking using self-learning algorithms, or whether the rules on liability or definition of addressees of competition law need to be altered. In order to that, this article will first briefly discuss the basic characteristics of self- learning algorithms and their implications for competition law (Chapter II) This is followed by a detailed analysis of competition law and practices relevant for the assessment of an algorithm’s deployment, with a particular emphasis on liability for anticompetitive conduct (Chapter III). And, the question of liability for the conduct of self-learning algorithms it then discussed in the final chapter (Chapter IV). II. Algorithms Without going into the details on the definition of an algorithm, 4 we may start with a widely accepted understanding that an algorithm is “ a sequence of simple and/or well-defined operations that should be performed in an exact order to carry out a certain task or class of tasks or to solve a certain problem or class of problems” ; 5 because this article is only concerned with computing operations, we will further limit the definition to digital algorithms, meaning “ any well-defined computational procedure that takes some value, or set of values, as input and produces some value, or set of values, as output ”, i.e. “ a sequence of computational steps that transform the input into the output ”. 6 There has recently been a lot of attention dedicated to the consequences that the massive deployment of artificial intelligence algorithms may have for competition policy, in particular 2 In detail, see e. g. ROSE, V., BAILEY, D. Bellamy & Child. European Union Law of Competition. Seventh Edition. Oxford University Press, 2013, p. 82 et seq. 3 Ibid , p. 102. 4 In detail, see e. g. MOSCHOVAKIS, Y. N. What is an Algorithm? In: ENGQUIT, B., SCHMID, W. (eds) Mathematics Unlimited: 2001 and Beyond . Springer, 1999, p. 919. 5 KNUTH, D. E. The Art of Computer Programming. Volume 1: Fundamental Algorithms. 3rd edition . Addison- Wesley Professional, 1997, p. 1 et seq . 6 CORMEN, T. H., LEISERSON, E. L., RIVEST, R. L., STEIN, C. Introduction to Algorithms. 3rd edition . MIT Press, 2009, p. 5.

240