Medical Interview AI with an Automatic Feedback Mechanism towards ChatGPT's Over-Detailing Responses

Abstract

Objective Structured Clinical Examination (OSCE) is a common test for medical students, assessing their clinical skills and attitudes through clinical interviews with Standardized Patients (SP). However, High-cost training is required for SP to act based on scenarios, and it doesn’t always standardize them effectively. Addressing this, we explore using artificial intelligence in medical interviews. This study focuses on using ChatGPT, a large language model, to simulate SP responses in OSCE. However, ChatGPT often gives overly detailed responses, unsuitable for OSCE that test information elicitation skills. To address this limitation, we propose a mechanism that identifies and corrects such excessive responses from ChatGPT and integrates this mechanism into the ChatGPT-based SP, ensuring that the responses are relevant and aligned with the objectives of OSCE.