Toward a Metrology for Artificial Intelligence: Hidden-Rule Environments and Reinforcement Learning
This work addresses the challenge of AI metrology in complex, rule-based environments, which is incremental as it builds on existing RL methods with new experimental setups.
The paper tackles the problem of reinforcement learning in environments with hidden rules, specifically the Game Of Hidden Rules (GOHR), where an agent must infer and execute rules to clear a board; it explores state representation strategies and uses a Transformer-based A2C algorithm, evaluating models across various setups to analyze transfer effects and learning efficiency.
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$\times$6 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.