The paper “Winning at Any Cost – Infringing the Cartel Prohibition With Reinforcement Learning”, authored by Michael Schlechtinger, Damaris Kosack, Thomas Fetzer, and Heiko Paulheim was accepted at the 19th International Conference on Practical Applications of Agents and Multi-Agent Systems.
In the paper, we explore how agents based on reinforcements can learn to cooperate, even if they are not explicitly programmed to do so. While the example in the paper is a simple three-player game, it can be transfered to more realistic cooperative scenarios, like algorithmic pricing, where such behavior of automated agents may pose challenges for antitrust legislation.
The paper is a joint publication between the DWS group and the chair of public law, regulatory law and tax law. It is the first publication in the interdisciplinary research project KareKoKI.
A preprint is available here.