Foundation models and learned game-world models are no longer confined to playing games. According to a new survey titled AI for Games in the Foundation Model Era, these systems now model players and game dynamics, assist with design and development, adapt player-facing experiences at runtime, and evaluate the artifacts they help produce.
Yet these directions have largely evolved in isolation, the authors argue, leaving unclear which capabilities transfer across settings and which stay tied to specific games, engines, interfaces, or player populations. To bring order to the field, the survey sorts the literature into six roles based on how AI output is immediately used: playing and acting; modeling players and games; designing games; building and maintaining games; generating and adapting at runtime; and testing and evaluating games.
For each role, the authors examine what structure the game or workflow supplies, what the AI learns or produces, which capabilities and artifacts carry over across settings and roles, and what evidence backs the resulting claims. They also trace cross-role connections: gameplay trajectories train world models, learned environments supply experience for agents, design specifications feed executable implementations, and play or testing feedback drives revision.
The survey strikes a cautionary note. Control schemes, rules, engine interfaces, state representations, and player contexts often remain setting-specific, so downstream claims need validation in the target setting. Evaluation is most standardized for bounded game playing and selected learned environments, while persistent state in learned worlds, repeated software revision, validated player modeling, sustained runtime adaptation, and representative automated testing remain less established.
The central challenge, the authors write, is to reuse or transfer outputs and capabilities across roles while re-establishing evidence of effectiveness in the game-specific contexts where they are used. The work is a joint effort from the National University of Singapore and Nanyang Technological University, and is accompanied by a living collection of 439 references, an interactive survey map, and six playable AI-crafted worlds.