Q-Pain

Q-Pain数据集由斯坦福大学创建，包含55个医疗问答对，旨在评估疼痛管理领域中医疗问答系统的社会偏见。数据集中的每个问题都包含一个详细的病人特定医疗场景，设计用于替换不同的种族和性别‘概况’，以识别在回答是否开药时的歧视。数据集的创建过程涉及临床专家设计与真实情况一致的医疗场景。该数据集主要应用于评估和改进医疗AI系统在疼痛管理决策中的公平性，以解决医疗AI应用部署前的安全问题。

The Q-Pain dataset was developed by Stanford University, consisting of 55 medical question-answer pairs, with the core goal of evaluating social biases in medical question-answering systems within the domain of pain management. Each question in the dataset features a detailed patient-specific medical scenario, which is designed to swap in distinct racial and gender "profiles" to detect discrimination when the system responds to whether to prescribe medication. The creation of this dataset involved clinical experts in designing scenarios that are consistent with real-world clinical situations. This dataset is primarily applied to evaluate and improve the fairness of medical AI systems in pain management decision-making, so as to address safety concerns prior to the deployment of medical AI applications.

搜集汇总

背景与挑战

背景概述

Q-Pain数据集由斯坦福大学创建，包含55个医疗问答对，旨在评估疼痛管理领域医疗问答系统的社会偏见。它通过设计不同种族和性别‘概况’替换的医疗场景，来识别开药决策中的歧视，并应用于改进医疗AI系统的公平性和安全性。