Can Games Be AI Explanations? An Exploratory Study of Simulation Games

This exploratory study asks whether simulation games such as Farming Simulator 19, Civilization, RimWorld, The Sims 4, Crusader Kings 3, and Democracy 3 can function as explainable AI interfaces. Its approach is a qualitative affordance analysis that maps the mechanics and interfaces of these games onto Liao et al.'s XAI Question Bank. The authors frame their work as a preliminary investigation rather than a comprehensive survey, noting that few simulation games are deliberately designed to emulate machine learning models, so the conclusions are provisional.

The central finding is that simulation games can support most XAI question types, though not equally well. "What If" and "How" questions are most naturally addressed, since gameplay consists of manipulating inputs and observing outputs, which makes experimentation a built-in mode of explanation. Games address these questions through visualizations, predictive screens, and interactive menus, and players gain understanding through agency and trial-and-error rather than through passive presentation. "Why Not" questions are rarely addressed with explicit examples, partly because they are largely redundant with "Why" questions in binary-outcome games.

The authors also identify challenges that limit this potential. Mechanics can be opaque to inexperienced players, and guiding explanation without undermining player agency remains difficult. To address these tensions, they propose three design considerations for playful explanation interfaces: providing objectives to focus player attention, using failure to prompt exploration of system boundaries, and pairing experimental "What If" and "How" exploration with explicit visualizations that answer supporting questions. Together these suggest that games may represent a new and promising form of AI explanation, though the authors stress the need for further empirical validation.