CYIL vol. 11 (2020)

CYIL 11 (2020) ARTIFICIAL INTELLIGENCE AND COMPETITION LAW… Laboratory research suggests that it is possible. An article recently published by researchers from several European universities describes interaction of two algorithms designed only to maximise profits established tacit collusion in over 70% of cases. 81 The experiment ran for one million “games”, tacit collusion was however established “already” after 70 000 of them. 82 It needs to be added that even these results are disputed by other researchers, as the laboratory conditions do not fully correspond with practice. 83 Nonetheless, even if the algorithms “spontaneously” arrived to tacit collusion, it would arguably not be covered by competition law, as it stands today, and we will not discuss it further in this article. 3. Pricing Algorithms and Abuse of Dominance As has already been indicated, most of the research has so far been dedicated to using algorithms to establish or sustain coordination among undertakings, i.e. to anticompetitive agreements. However, abuse of dominance is also not without relevance, as will be discussed below. Under competition law, individual (non-coordinated) conduct of undertakings may be assessed as an abuse of dominant position, which is defined as a position of economic strength enjoyed by an undertaking, which enables it to prevent effective competition being maintained on a relevant market, by affording it the power to behave to an appreciable extent independently of its competitors, its customers and ultimately of consumers. 84 A dominant position is typically associated with very high market shares and is therefore not prevalent, especially in online retail. However, if an undertaking is in a dominant position, that position confers a special responsibility on it, the scope of which must be considered in the light of the specific circumstances of each case. 85 In essence, it cannot abuse its dominance. Personalised pricing may in principle constitute an abuse of dominance. Under the most likely scenario, personalised pricing might be treated as an exploitative abuse, a form of excessive pricing, under the rational that some consumers are charged higher prices for reasons not related to costs. 86 In any event, it would be necessary to demonstrate its negative effect on competition, which, as the scholars suggest, is not an unavoidable effect of personalised pricing: The welfare effects of personalised pricing are a priori ambiguous. As we have shown, the economic literature emphasises that price discrimination is not necessarily detrimental to welfare or consumer surplus, and that it can increase welfare and/or consumer surplus 81 CALVANO, E., CALZOLARI, G., DENICOLO, V., PASTORELLO, S. Algorithmic Pricing: What Implications for Competition Policy? (28. July 2018). Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_ id=3209781 (1 June 2020), p. 7. 82 CALVANO, CALZOLARI, DENICOLO, PASTORELLO ( op. cit. sub 81), p. 11. 83 TESAURO, G., KEPHART, J. O. Pricing in Agent Economies Using Multi-Agent Q-Learning. Autonomous Agents and Multi-Agent Systems , 2002 (5), p. 303. 84 CJ EU judgement of 14 February 1978 27/76 United Brands v Commission , ECLI:EU:C:1978:22, par. 65. 85 CJ EU judgement of 9 November 1963 322/81 Nederlandsche Banden Industrie Michelin (Michelin I )

v Commission , ECLI:EU:C:1983:313, par. 57. 86 OECD Report on Personalised Pricing , p. 28.

253