Evaluating contrastive explanations for AI planning with non-experts: a smart home battery scenario

Smart home systems with AI planning functionality have the potential to improve the lives of users. However, there is an emerging expectation that users should better understand and trust the decision-making of AI systems. In this paper, a smart home battery system is developed with a supplementary explanation module that allows non-expert users to intuitively visualise the planning process and to better understand its recommendations. The module relies on a notion of contrastive explanations, related to iterative planning, allowing users to ask contrastive questions based on state- and action-constraints that may or may not be satisfiable. The system is intended for an experimental study where participants interact with the planning system and complete a questionnaire, with the research objective being to evaluate the usefulness of the explanation module.