Artificial Intelligence: Can Seemingly Collusive Outcomes Be Avoided?

Abstract

Strategic decisions are increasingly delegated to algorithms. We extend the results of Waltman and Kaymak [2008] and Calvano et al. [2020b] to the context of dynamic optimization with imperfect monitoring by analyzing a setting where a limited number of agents use simple and independent machine-learning algorithms to buy and sell a storable good on behalf of a large number of consumers. No specific instruction is given to them, only that their objective is to maximize profits based solely on past market prices and payoffs. With an original application to battery operations, we observe that the algorithms learn quickly to exert market power at seemingly collusive levels, despite the absence of any formal communication between them. Contrary to the findings reported in the existing literature, we show that seeming collusion may originate in imperfect exploration, rather than excessive algorithmic sophistication. We then show that a regulator may succeed in disciplining the market to produce socially desirable outcomes by enforcing decentralized learning or with adequate intervention during the learning process.