Exploratory Not Explanatory: Counterfactual Analysis of Saliency Maps for Deep Reinforcement Learning
This addresses the reliability of interpretability methods for deep RL practitioners, but it is incremental as it critiques and refines existing approaches without proposing a new paradigm.
The paper tackled the problem of unfalsifiable and subjective explanations from saliency maps in deep reinforcement learning by introducing a counterfactual reasoning approach to test hypotheses, finding that saliency maps are better suited as exploratory rather than explanatory tools in Atari games.
Saliency maps are frequently used to support explanations of the behavior of deep reinforcement learning (RL) agents. However, a review of how saliency maps are used in practice indicates that the derived explanations are often unfalsifiable and can be highly subjective. We introduce an empirical approach grounded in counterfactual reasoning to test the hypotheses generated from saliency maps and assess the degree to which they correspond to the semantics of RL environments. We use Atari games, a common benchmark for deep RL, to evaluate three types of saliency maps. Our results show the extent to which existing claims about Atari games can be evaluated and suggest that saliency maps are best viewed as an exploratory tool rather than an explanatory tool.