Explainable AI often aims for technical transparency, but that does not always mean users can make sense of the output. A new paper from arXiv proposes MIND-X, a systematic design approach that instead starts from the user's mental model of the task. The goal is to generate explanations that align with how people already think, rather than forcing them to interpret model-centric reasoning.
The authors illustrate the approach in the domain of fake review identification, where users must decide whether a review is trustworthy. MIND-X is presented as a general framework, but the example shows how it can be applied to a concrete, high-stakes classification problem. The paper argues that by leveraging mental models, explanations can better support human sense-making and decision-making.
This work highlights a growing shift in XAI research: moving the evaluation criterion from model fidelity to user comprehension. While the paper is an early-stage proposal, it offers a structured way to design explanations that are both interpretable and useful in practice.