Original title: Accountable AI with Grounded, Faithful, Consistent, Actionable Rationales: A Case Study in Clinical Trial Matching with VERDICT
Zikai Zhou, Yufei Jin, Yilin Xu, Yu-Chiang Wang +2 · September 2026
Doctors need to know why an artificial intelligence system makes a choice, especially when matching patients to clinical trials. Standard large language models can pick trials fairly well, but they often change their rules randomly and give explanations that do not match how they actually decided. To fix this, researchers built VERDICT. It translates a matching task, its rules, and its constraints into formal logic equations. Specialized math software called solvers then solves these equations to make the final choice. Because math rules guide the outcome, the system treats identical facts the exact same way every time. The researchers found that VERDICT beat other systems in accuracy, applied its rules with complete consistency, and gave explanations that clinicians preferred because they clearly showed key assumptions and pivotal facts.
Imagine you're grading essays with rules to deduct points for typos, but sometimes you forget and give full marks anyway. The inconsistency means similar essays get different grades, and students can't trust your feedback. In this paper, AI often applies rules inconsistently to medical decisions. VERDICT creates a strict grading system that always follows the rules, so the same patient details get the same trial recommendation.
For patients and doctors, this means AI decisions in clinical trials can be trusted, examined, and challenged, ensuring treatments are based on consistent, justified policies.
The abstract presents this as a case study in clinical trial matching, so its effectiveness in other domains or with different data is not established. It evaluates on specific datasets, limiting generalizability.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Accountable AI with Grounded, Faithful, Consistent, Actionable Rationales: A Case Study in Clinical Trial Matching with VERDICT — Zikai Zhou, Yufei Jin, Yilin Xu et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.