Explaining Learned Reward Functions with Counterfactual Trajectories

Learning rewards from human behaviour or feedback is a promising approach to aligning AI systems with human values but fails to consistently extract correct reward functions. Interpretability tools could enable users to understand and evaluate possible flaws in learned reward functions. We propose Counterfactual Trajectory Explanations (CTEs) to interpret reward functions in reinforcement learning by contrasting an original with a counterfactual partial trajectory and the rewards they each receive. We derive six quality criteria for CTEs and propose a novel …

ai systems counterfactual cs.ai cs.lg extract feedback flaws functions human interpretability reinforcement systems tools trajectory values