AI for Games in the Foundation Model Era

arXiv 2026 34.3 method

TLDR

Survey organizing foundation-model and learned-world-model research in games into six roles, examining capability transfer, evidence, and connections across the game lifecycle.

Reasoning

The paper provides a useful cross-role synthesis of AI for games, linking world models, learned environments, and game design workflows. However, as a literature review, it does not present new real-world experiments or benchmarks, and the abstract leaves empirical evidence and limitations underspecified.

Read-first score

Read-first score 34.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 30.

Recency 5%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Methodology quality 15%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=evaluation,result,validation

Topical relevance 25%

42.9

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 15%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=artifact

Graph prestige 15%

5.7

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0016

Citation impact 15%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 10%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 439.

Keyword Scores

world model

world dynamics prediction

model-based reinforcement learning world model

world simulator

generative world model

interactive world model

video world model

Deep Analysis

Innovations

Introduces a six-role taxonomy of AI use across the game lifecycle: playing and acting, modeling players and games, designing games, building and maintaining games, generating and adapting at runtime, and testing and evaluating games.
Articulates cross-role transfer pathways: trajectories train world models, learned environments provide experience for agents, design specifications drive executable implementations, and play or testing feedback guides revision.
Frames the central challenge as reusing or transferring outputs and capabilities across roles while re-establishing evidence for effectiveness in game-specific contexts.

Methodology

The paper is a structured literature survey that organizes AI-for-games research into six roles based on the immediate use of AI output. For each role, it examines what structure the game or workflow supplies, what the AI learns or produces, which capabilities and artifacts transfer, and what evidence supports claims. It also synthesizes cross-role connections and assesses evaluation maturity across settings.

Key Results

The survey finds that cross-role connections exist, but many control schemes, rules, engine interfaces, state representations, and player contexts remain setting-specific, requiring target-setting validation. 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.

Limitations

Research directions have evolved largely separately, obscuring which capabilities transfer across settings and which remain tied to particular games, engines, interfaces, or player populations.
Control schemes, rules, engine interfaces, state representations, and player contexts often remain setting-specific, so downstream claims require validation in the target setting.
Evaluation is less established for persistent state in learned worlds, repeated software revision, validated player modeling, sustained runtime adaptation, and representative automated testing.

Resources

Paper