Robust Explanations for Human-Neural Multi-agent Systems with Formal Verification.

The quality of explanations in human-agent interactions is fundamental to the development of trustworthy AI systems. In this paper we study the problem of generating robust contrastive explanations for human-neural multi-agent systems and introduce two novel verification-based algorithms to (i) identify non-robust explanations generated by other methods and (ii) generate contrastive explanations equipped with formal robustness certificates. We present an implementation and evaluate the effectiveness of the approach on two case studies involving neural agents trained on credit scoring and traffic sign recognition tasks.