Counterfactual Strategies for Markov Decision Processes
This addresses the need for explainable AI in sequential decision-making tasks, though it is incremental by extending existing counterfactual methods to MDPs.
The paper tackles the problem of applying counterfactual explanations to sequential decision-making by introducing counterfactual strategies for Markov Decision Processes, enabling minimal changes to strategies to reduce the probability of undesired outcomes below a limit, and demonstrates practical viability on four real-world datasets.
Counterfactuals are widely used in AI to explain how minimal changes to a model's input can lead to a different output. However, established methods for computing counterfactuals typically focus on one-step decision-making, and are not directly applicable to sequential decision-making tasks. This paper fills this gap by introducing counterfactual strategies for Markov Decision Processes (MDPs). During MDP execution, a strategy decides which of the enabled actions (with known probabilistic effects) to execute next. Given an initial strategy that reaches an undesired outcome with a probability above some limit, we identify minimal changes to the initial strategy to reduce that probability below the limit. We encode such counterfactual strategies as solutions to non-linear optimization problems, and further extend our encoding to synthesize diverse counterfactual strategies. We evaluate our approach on four real-world datasets and demonstrate its practical viability in sophisticated sequential decision-making tasks.