Computational Models of Player Strategy in Roguelike Games

Abstract

Formulating problems algorithmically and developing systematic solutions is central to computer science education, yet teaching algorithmic reasoning remains difficult. We argue that roguelike video games provide environments where players naturally develop algorithmic thinking without explicit instruction. We formalize the relationship between roguelike gameplay mechanics and AI problem structures. We demonstrate that player strategies in survivor-like roguelikes instantiate solutions to constraint satisfaction problems, Markov decision processes, multi-armed bandits, and other canonical frameworks. Players develop heuristics and refine strategies through iterative experimentation that mirrors algorithm design in AI courses. We then demonstrate how player-developed strategies correspond to classical algorithms, and discuss pedagogical implications for leveraging game-based intuitions in algorithm instruction. We argue algorithmic thinking emerges naturally when problem structures demand it. This suggests opportunities to bridge implicit computational patterns in gameplay with explicit formal frameworks. This framework inverts the standard relationship between AI and games: rather than applying AI to games, we establish games as a domain in which canonical AI problems are naturally instantiated.