Back to AI News
arxiv

When Expert Disagreement Hurts: Prestige and Selectivity in LLM Decision Revision

This paper introduces a controlled two-pass audit to measure whether LLMs improve or harm their decisions when revising after disagreement, separating reversal counts from accuracy effects. Across five models and 1,989 evidence-grounded decisions in medicine, science, and law, no model met the selectivity bar under expert disagreement, and accuracy fell by up to 16.1 points. The expert label increased reversals for four of five models, but its accuracy impact ranged from a 5.2-point loss to a 2.4-point gain, showing reversal counts are a poor proxy for harm. The authors argue revision pipelines should be evaluated with paired, conditional metrics per model and protocol.

AI Brief

Provides concrete evidence that expert-labeled disagreement can degrade LLM accuracy despite increasing reversals, offering actionable evaluation guidance for revision pipelines.

arXiv:2611.16694v1 Announce Type: new Abstract: When language models revise a decision after another agent disagrees, neither reversal counts nor final accuracy shows whether they are fixing mistakes or abandoning correct answers. We introduce a controlled two-pass audit that holds the question, the evidence, and the model's initial answer fixed and varies only the revision context: neutral reflection, anonymous disagreement, or disagreement attributed to a domain expert. For binary decisions the accounting is exact: revision raises accuracy only if it is selective enough, reversing wrong answers more than $a/(1-a)$ times as often as correct ones, where $a$ is initial accuracy (nine times at 90%). Across five models and 1,989 evidence-grounded decisions in medicine, science, and law, no model clears this bar under expert disagreement, and accuracy falls by up to 16.1 points. The expert label, a prestige cue, adds reversals for four of the five models, yet its effect on accuracy ranges from a 5.2-point loss to a 2.4-point gain, so reversal counts are a poor proxy for harm. With identical arguments from both sources, the label still raises reversal for three proprietary models and in a pooled analysis of four newer models. Models also respond to stated source reliability. Revision pipelines should therefore be evaluated with paired, conditional metrics, per model and protocol.
Be the first to rate this article.

Comments & Questions (0)

Captcha
Please be respectful — let's keep the conversation friendly.

No comments yet

Be the first to comment!