Intel / article

XAI-Arena: Can LLMs Assess the Quality of XAI Explanations?

Source: arxiv.org
Author: Yanfei Hu Fleischhauer, Alona Zharova, Nadja Klein, Stefan Feuerriegel
Date: Published Sep 10, 2026

Context

XAI-Arena asks language models to compare explanations of machine-learning decisions along dimensions such as clarity, task adequacy and faithfulness. It also varies the intended audience, recognizing that different users need different kinds of explanation.

The authors report a Spearman rank correlation of 0.693 between model-generated and human ratings, which they describe as a strong positive association in their human validation. It supports using model judges to scale comparative assessment, while agreement with human ratings alone does not establish that an explanation accurately represents the underlying decision process. The reported result concerns assessment agreement.

Terms in this piece · Glossary

LLM-as-judge — Using one model to score another's output against a rubric, so quality can be measured at a scale human grading cannot reach.
calibration — How well a model's confidence matches reality — a calibrated model saying "90% sure" is right about 90% of the time.
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.

Key quotes

“Human validation shows a strong positive association between LLM-generated and human ratings (Spearman's rho=.693, p<.001).”

“Together, LLM-based evaluations can capture systematic differences in XAI explanation quality and provide a scalable and reproducible framework for comparative assessment of XAI explanations.”

Read the source arxiv.org