Explanations are a Means to an End: Decision Theoretic Explanation Evaluation

Explanations of model behavior are commonly evaluated via proxy properties weakly tied to the purposes explanations serve in practice. We contribute a decision theoretic framework that treats explanations as information signals valued by the expected improvement they enable on a specified decision task. This approach yields three distinct estimands: (i) a theoretical benchmark that upper-bounds achievable performance by any agent with the explanation, (ii) a human-complementary value that quantifies the theoretically attainable value that is not already captured by a baseline human decision policy, and (iii) a behavioral value representing the causal effect of providing the explanation to human decision-makers. We instantiate these definitions in a practical validation workflow, and apply them to assess explanation potential and interpret behavioral effects in human–AI decision support and mechanistic interpretability.

Explanations are intended to help people make better decisions, but they are often evaluated using indirect criteria, such as whether they seem understandable or accurately reflect the model. This paper argues that explanations should instead be evaluated by whether they improve decisions in a specific task. We propose a framework that asks three questions: whether the task contains useful information an explanation could help reveal, whether that information goes beyond what people already use, and whether giving the explanation actually improves people’s decisions. This approach helps researchers identify when explanations are worth testing and diagnose why they succeed or fail.