Beyond Majority Vote: Multi-Perspective Adjudication for Medical Hallucination Detection
Abstract
AbstractUnderstanding the frequency of factual errors in chatbot-generated text and evaluating systems that detect these errors is critical for determining chatbot safety. Yet factual-error detection is often treated as a single-pass, single-annotator labeling problem. In long-form chatbot responses, factual errors can be subtle and embedded within mostly correct text.We develop a multi-perspective annotation study of medically relevant chatbot responses, combining first-pass annotation, LLM-as-a-Judge candidate discovery, and two forms of adjudication: domain-expert and evidence-based fact-checking. First-pass annotators frequently miss factual errors later validated by adjudicators. LLM-as-a-Judge improves candidate discovery, but is insufficient on its own: it misses factual errors that annotators catch. We also find disagreement among adjudicators, suggesting that adjudication over multiple candidate sources can improve benchmark completeness, but does not eliminate the need to apply judgment and expertise. Applied to an existing benchmark, this technique reveals a similar pattern of missed errors.Together, these results suggest that single-pass hallucination benchmarks achieve scale, but likely undercount errors. Multi-pass adjudication improves coverage, but benchmark results will still need to be interpreted in light of the judgment, expertise, and evidence used when deciding if an error is present.