CYIL vol. 11 (2020)

MICHAL PETR CYIL 11 (2020) We put forward that the latter approach is more robust and convincing. It however needs to be added that thanks to the “compliance by design” requirement, avoiding the liability would still be very difficult, though not close to impossible, as in the case of the former approach. It is also more in line with the requirement of fault in order to establish liability V. Conclusions Even though the deployment of self-learning algorithms is believed to revolutionize the commerce in the near future, it does not necessarily imply the need to revisit the basic concepts of competition law. As we discussed in this article, the vast majority of scenarios fit without any problems into the framework of the law as it stands today. The only category creating doubts is the – as of today hypothetical – scenario, under which a deep learning “black box” algorithm finds a way to communicate with another one, without the knowledge of the undertaking itself, and starts to set prices in coordination with other undertakings on its behalf, or under which such an algorithm, employed by a dominant undertaking, “learns” to set vastly different personalised prices, to the detriment of consumers. We put forward that even under such a scenario, the liability may be attributable to the undertaking itself, because it in principle needs to make sure that the algorithms it uses are not capable of engaging in such conduct. Only if it passes the “due diligence” test, it may escape the liability, similarly as it may escape liability for the conduct performed on its behalf by third parties.

256