Beyond Value: CHECKLIST for Testing Inferences in Planning-Based RL
This addresses the problem of improving evaluation and generalization for RL agents, particularly in planning-based settings, but it is incremental as it adapts an existing testing method from another domain.
The paper tackles the limited evidence for post-deployment generalization in RL by extending the CheckList testing methodology from NLP to planning-based RL, allowing users to identify previously-unknown flaws in an agent's reasoning during tree search, as shown in a user study with AI researchers evaluating an agent in a complex real-time strategy game.
Reinforcement learning (RL) agents are commonly evaluated via their expected value over a distribution of test scenarios. Unfortunately, this evaluation approach provides limited evidence for post-deployment generalization beyond the test distribution. In this paper, we address this limitation by extending the recent CheckList testing methodology from natural language processing to planning-based RL. Specifically, we consider testing RL agents that make decisions via online tree search using a learned transition model and value function. The key idea is to improve the assessment of future performance via a CheckList approach for exploring and assessing the agent's inferences during tree search. The approach provides the user with an interface and general query-rule mechanism for identifying potential inference flaws and validating expected inference invariances. We present a user study involving knowledgeable AI researchers using the approach to evaluate an agent trained to play a complex real-time strategy game. The results show the approach is effective in allowing users to identify previously-unknown flaws in the agent's reasoning. In addition, our analysis provides insight into how AI experts use this type of testing approach, which may help improve future instantiations.