The paper argues that current human-centred explainable AI (XAI) research has a timing problem: it typically evaluates explanations only after they have been produced. That post-hoc approach, the authors contend, offers little actionable guidance for making explanations more accessible in the first place.
To address this, the study suggests shifting part of the evaluation earlier—into feature selection. By integrating human-based feature understandability into the selection process, the method aims to build tabular explanations from features that are inherently easier for people to grasp, rather than trying to retrofit clarity after the fact.
Because the source is a single arXiv abstract, the details of the proposed method and its evaluation are not yet available in this summary. The contribution, as stated, is a conceptual reorientation: treat understandability as a selection criterion, not just an outcome metric.