analiza aiqbench
This study demonstrates that integrating domain-specific specialist agents with a two-phase consistency verification process significantly improves uncertainty calibration in medical multiple-choice tasks. By decoupling consistency checks from ensemble aggregation, the framework achieves substantial reductions in expected calibration error while simultaneously boosting discrimination metrics on high-disagreement datasets. These findings offer a concrete pathway for deploying more reliable LLMs in clinical settings where accurate confidence signals are critical for deferral decisions.
Multi-Agent Reasoning with Consistency Verification Improves Uncertainty Calibration in Medical MCQA
arXiv:2603.24481v2 Announce Type: replace-cross Abstract: Miscalibrated confidence scores are a practical obstacle to deploying AI in clinical settings. A model that is always overconfident offers no useful signal for deferral. We present a multi-agent framework that combines domain-specific specialist agents with Two-Phase Verification (Wu et al., 2024) and S-Score Weighted Fusion to improve both calibration and discrimination in medical multiple-choice question answering. Four specialist agents (respiratory, cardiology, neurology, gastroenterology) generate independent diagnoses using Qwen2.5-7B-Instruct. Each diagnosis undergoes a two-phase self-verification process that measures internal consistency and produces a Specialist Confidence Score (S-score). The S-scores drive a weighted fusion strategy that selects the final answer and calibrates the reported confidence. We evaluate on high-disagreement subsets of MedQA-USMLE and MedMCQA (100 and 250 questions). All results are specific to this filtered regime. On MedQA-250, the full system achieves ECE = 0.091 (74.4% reduction over the single-specialist baseline) and AUROC = 0.630 (+0.056) at 59.2% accuracy. Calibration gains of 49-74% hold across all four settings. Ablation analysis reveals that Two-Phase Verification drives ECE reduction while multi-agent reasoning drives AUROC improvement, suggesting that consistency checking and ensemble aggregation address different failure modes of LLM uncertainty. Whether the resulting confidence signal is sufficient to support clinical deferral decisions in practice remains a direction for future investigation.
Abstract:Miscalibrated confidence scores are a practical obstacle to deploying AI in clinical settings. A model that is always overconfident offers no useful signal for deferral. We present a multi-agent framework that combines domain-specific specialist agents with Two-Phase Verification (Wu et al., 2024) and S-Score Weighted Fusion to improve both calibration and discrimination in medical multiple-choice question answering. Four specialist agents (respiratory, cardiology, neurology, gastroenterology) generate independent diagnoses using Qwen2.5-7B-Instruct. Each diagnosis undergoes a two-phase self-verification process that measures internal consistency and produces a Specialist Confidence Score (S-score). The S-scores drive a weighted fusion strategy that selects the final answer and calibrates the reported confidence. We evaluate on high-disagreement subsets of MedQA-USMLE and MedMCQA (100 and 250 questions). All results are specific to this filtered regime. On MedQA-250, the full system achieves ECE = 0.091 (74.4% reduction over the single-specialist baseline) and AUROC = 0.630 (+0.056) at 59.2% accuracy. Calibration gains of 49-74% hold across all four settings. Ablation analysis reveals that Two-Phase Verification drives ECE reduction while multi-agent reasoning drives AUROC improvement, suggesting that consistency checking and ensemble aggregation address different failure modes of LLM uncertainty. Whether the resulting confidence signal is sufficient to support clinical deferral decisions in practice remains a direction for future investigation.
20 pages, 6 figures. Preprint under review
Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
arXiv:2603.24481 [cs.AI]
(or arXiv:2603.24481v2 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2603.24481 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 312 cuvinte
Subiecte conexe
As Expected? Efficient LLM Planning via Expectation-Observation Consistency Verification
arXiv · 3 nov. 2026
Retrosynthesis of Synthetic Media for Explainable AI Provenance Forensics
arXiv · 3 sept. 2026
Action-Consistent Decentralized Belief Space Planning with Inconsistent Beliefs and Limited Data Sharing: Framework and Simplification Algorithms with Formal Guarantees
arXiv · 3 nov. 2026
VeriGeo: Controllable Geometry Question Generation with Numerical and Analytical Verification
arXiv · 15 iun. 2026