Summary

General Game Playing usually aims to create agents able to win any game as efficiently as possible, using winning or score as the main utility signal. This report studies a complementary direction: agents that mimic human behaviour in games, where winning is not always the only relevant objective.

The manuscript surveys concepts, roadblocks, and techniques for building more human-like general game playing agents. It discusses why superhuman agents often behave in ways that are difficult for human players to understand, and why more human-like agents could support enjoyable play, realistic game data generation, and deeper analysis of games.

The work is especially relevant to tabletop games, Ludii, the Digital Ludeme Project, and the GameTable COST Action, where artificial agents are used to model, reconstruct, test, and analyse plausible rulesets for historical games.

Reference

Rautureau, A. (2024). Winning is not everything: Human-like agents for tabletop games. Master internship report, UCLouvain.

BibTeX

@mastersthesis{rautureau2024winning,
  author = {Rautureau, Alois},
  title  = {Winning is not everything: Human-like agents for tabletop games},
  school = {UCLouvain},
  year   = {2024},
  type   = {Master internship report},
  url    = {https://piette.info/eric/master/2024%20-%20Winning%20is%20not%20everything%20-%20Human-like%20agents%20for%20tabletop%20games.pdf}
}