Comparative Explanations: Explanation Guided Decision Making for Human-in-the-Loop Preference Selection

AI systems can help humans make better trade-off decisions: MOLONE demonstrates that when AI systems must help humans choose between competing goals (like cost vs. quality), providing explanations about both what factors matter and how different options compare leads to faster, more confident decision-making compared to unexplained recommendations.

Current AI explanations often miss half the picture: Most explainable AI methods focus only on explaining which input features (data) drive decisions, but ignore explaining the output objectives themselves. This paper shows that explaining both what matters in the data and how different outcomes compare is crucial for helping humans understand complex trade-offs.

Local, comparative explanations are more useful than global ones: Rather than providing broad explanations about how an AI system works overall, MOLONE focuses on explaining specific comparisons between nearby options in the decision space. A user study confirmed this targeted approach helps humans make better choices more quickly in real-world scenarios.

Human-in-the-loop AI requires rethinking explanation design: The paper highlights that when humans are actively involved in guiding AI systems (rather than just receiving final recommendations), the explanation methods need to be fundamentally different—designed specifically to support preference elicitation rather than post-hoc justification of decisions.