Human-Like Game-Playing Agents Beyond Behavioural Alignment
(2026)
Files
2026 - Human-Like Game-Playing Agents Beyond Behavioural Alignment.pdf
Open Access - Adobe PDF
- 1.72 MB
Details
- Authors
- Abstract
- Research on game-playing artificial intelligence has historically treated games as benchmarks of intelligence. With artificial agents surpassing human players in many games over the years, this original goal has shifted towards designing agents that are aligned with human behavior to provide more engaging experiences, actionable strategies, or simply to study human playing styles. Modern approaches to “human-like” play mainly focus on behavioural alignment, broadly defining human play from a functionalist viewpoint derived from evaluation frameworks similar to the Turing test. Although significant advances have been made through these lenses, we argue that it narrows the definition of play, overlooking social, interactive, and cognitive dimensions. While these aspects formed an active field of research in the past—mainly with the goal of approaching the strong intuition of human experts—they have become secondary in modern works. Beyond their interest to the field of artificial intelligence, a broader definition and study of human-like agents aims to support richer computational models to investigate the cognitive processes involved in decision-making, and the cultural and historical aspects of games. This thesis aims to reframe the definition of “human-like” agents by studying game-playing not only as a sequence of moves to reach a rule-defined goal, but as an inherently social, culturally-situated, and cogntitively-grounded activity. To this aim, we propose a multi-dimensional classification of human-like game-playing agents, review current methods for measuring behavioral alignment, and explore the prospect of cognitively plausible search algorithms in games. Beyond these exploratory assessments, we discuss future work and research directions for a broader study of human-like play.
- Affiliations
Citations
Rautureau, A. (2026). Human-Like Game-Playing Agents Beyond Behavioural Alignment. https://hdl.handle.net/2078.5/278867