The Price of Algorithmic Pricing: Investigating Collusion in a Market Simulation with AI Agents

Due to the rising availability and adoption of Artificial Intelligence in e-commerce, many of the online-prices are not set by humans, but by algorithms. The consequence is an opaque pricing situation that raises the potential of concealed, unfair competition by means of collusion. To examine this phenomenon, we study deep-Reinforcement-learning-based pricing algorithms by conducting an experiment involving an oligopoly model of repeated price competition. Our market model facilitates a variable environment spanning from economic theory to more realistic consumer demand models. We find that the algorithms learn to enter a collusive state and charge supra-competitive prices, without explicitly communicating with one another, and even without seeing each other's prices.