Rigorous Interpretation Is a Form of Evaluation

TL;DR

The paper argues that interpretability in machine learning should be viewed as a rigorous, scientific form of evaluation that can diagnose, detect, and predict model issues beyond traditional performance metrics.

Contribution

It proposes a framework for interpretability to serve as a scientific evaluation tool, emphasizing falsifiability, reproducibility, and predictive power.

Findings

Interpretability can diagnose root causes of unwanted behavior.

It can detect subtle faulty mechanisms invalidating outputs.

It can predict issues before they occur by understanding model weaknesses.

Abstract

Current machine learning models are evaluated through behavioral snapshots, with benchmark accuracies, win rates and outcome-based metrics. Model explanations and evaluations, however, are fundamentally intertwined: understanding why a model produces a behavior can be as important as measuring what it produces. If we trusted interpretability, we argue that it can serve not merely as diagnostics but as a richer and more principled form of model evaluation beyond surface-level performance metrics. We explore three ways interpretability can function evaluatively: (1) fixing problems by identifying the root causes of unwanted behavior, (2) detecting subtly faulty mechanisms that invalidate model outputs, and (3) predicting potential issues before they arise by fully understanding the model's weaknesses. To fulfill its evaluative potential, we argue that interpretability methods must generate claims that are falsifiable, reproducible, and predictive -- that is, interpretability must meet scientific standards.