PREF-XAI Personalizes Black-Box Model Explanations Using User Preference Elicitation

arXiv·medium signal

Greco, Karolczak, and Slowiński shift explainable AI from model-centric explanations to user-centric ones by incorporating user preferences into rule-based explanations of black-box ML models. PREF-XAI elicits what aspects of model behavior each user cares about and tailors explanations accordingly. Addresses the real-world problem that a data scientist, a compliance officer, and an end user need fundamentally different explanations of the same model.