English

Toward a Metrology for Artificial Intelligence: Hidden-Rule Environments and Reinforcement Learning

Machine Learning 2025-10-24 v3 Artificial Intelligence Machine Learning

Abstract

We investigate reinforcement learning in the Game Of Hidden Rules (GOHR) environment, a complex puzzle in which an agent must infer and execute hidden rules to clear a 6×\times6 board by placing game pieces into buckets. We explore two state representation strategies, namely Feature-Centric (FC) and Object-Centric (OC), and employ a Transformer-based Advantage Actor-Critic (A2C) algorithm for training. The agent has access only to partial observations and must simultaneously infer the governing rule and learn the optimal policy through experience. We evaluate our models across multiple rule-based and trial-list-based experimental setups, analyzing transfer effects and the impact of representation on learning efficiency.

Keywords

Cite

@article{arxiv.2509.06213,
  title  = {Toward a Metrology for Artificial Intelligence: Hidden-Rule Environments and Reinforcement Learning},
  author = {Christo Mathew and Wentian Wang and Jacob Feldman and Lazaros K. Gallos and Paul B. Kantor and Vladimir Menkov and Hao Wang},
  journal= {arXiv preprint arXiv:2509.06213},
  year   = {2025}
}
R2 v1 2026-07-01T05:25:25.183Z