Towards counterfactual explanations for sequential decision-making problems

Abstract: Counterfactual Explanations (CEs) are a leading paradigm for enhancing AI explainability, valued for their intelligibility and alignment with human reasoning. Defined as minimally altered inputs that yield a more desirable AI decision, CEs enable systematic "what-if" analysis and automated recourse. However, mainstream CE methodologies focus primarily on single-shot decisions (e.g., in classification), often proving ill-suited for tasks involving sequences of reasoning steps. In this talk, I will present three recent proposals we put forward to generate CEs for sequential decision-making problems. I will consider three different domains – automated planning, neuro-symbolic multi-agent systems, and Markov decision processes – discuss different types of counterfactual edits and present some preliminary solutions to compute CEs for these domains.

Short bio: Francesco is an Assistant Professor in the Department of Computing at Imperial College London. His research focuses on AI safety, explainability and alignment, with special emphasis on counterfactual explanations. Since 2022, he leads the project “ConTrust: Robust Contrastive Explanations for Deep Neural Networks”, a four-year effort devoted to the formal study of robustness issues arising in XAI. More details about Francesco and his research can be found at https://fraleo.github.io/.

https://univr.zoom.us/j/89036897160

ID riunione: 890 3689 7160