Explanations of Symbolic Reasoning to Effect Patient Persuasion and Education

Artificial Intelligence (AI) models can issue smart, context-sensitive recommendations to help patients self-manage their illnesses, including medication regimens, dietary habits, physical activity, and avoiding flare-ups. Instead of merely positing an "edict," the AI model can also explain why the recommendation was issued: why one should stay indoors (e.g., increased risk of flare-ups), why further calorie intake should be avoided (e.g., met the daily limit), or why the care provider should be contacted (e.g., prescription change). The goal of these explanations is to achieve understanding and persuasion effects, which, in turn, targets education and long-term behavior change. Symbolic AI models facilitate explanations as they are able to offer logical proofs of inferences (or recommendations) from which explanations can be generated. We implemented a modular framework called XAIN (eXplanations for AI in Notation3) to explain symbolic reasoning inferences in a trace-based, contrastive, and counterfactual way. We applied this framework to explain recommendations for Chronic Obstructive Pulmonary Disease (COPD) patients to avoid flare-ups. For evaluation, we propose a questionnaire that captures understanding, persuasion, education, and behavior change, together with traditional XAI metrics including fidelity (soundness, completeness) and interpretability (parsimony, clarity).