Investigating Using Reinforcement Learning Agents to Encourage Player Behaviour Change
Jawdat W Toume · ERA: Education and Research Archive (University of Alberta) · 2026
Players can grow bored in video games when they discover a single dominant strategy and so stop exploring the space of possible strategies. Player adaptation refers to approaches to adapt the game to the player, but they typically focus on changing the space of the game or the number or types of enemies, rather than enemy strategy. In this thesis, we present a system that aims to select an enemy strategy as a novel player adaption approach with the goal of encouraging the player to adapt by countering said strategy. We do this by learning a cycle of strategies via Reinforcement Learning that counter each other. We ran a human subject study to evaluate our system and determined that it was significantly more likely for players to change their strategy when presented with a counter strategy, which was itself counterable.