July 3 session – 9:00 a.m. – 10:00 a.m.Session chair : Frédéric Maris
Explaining the Space of Plans through Plan-Property Dependencies
Model-based approaches to AI are well suited to explainability in principle, given the explicit nature of their world knowledge and of the reasoning performed to take decisions. AI Planning in particular is relevant in this context as a generic approach to action-decision problems. Indeed, explainable AI Planning (XAIP) has received interest since more than a decade, and has been taking up speed recently along with the general trend to explainable AI.
The talk describes a recent approach to XAIP, a form of contrastive explanation aimed at answering user questions of the kind « Why do you suggest to do A here, rather than B (which seems more appropriate to me)? ». Answers to such questions take the form of reasons why A is preferrabe over B. We set up a formal framework allowing to provide such answers in a systematic way. We instantiate that framework with the special case of questions about goal-conjunction achievability in oversubscription planning (where not all goals can be achieved and thus a trade-off needs to be found). We show that powerful question languages can be compiled into that special case. We finally show that, based on suitably assembling algorithm elements from previous work, the approach is feasible computationally compared to other established planning problems.