Exploring patient trust in clinical advice from AI-driven LLMs like ChatGPT for self-diagnosis (2402.07920v1)

Abstract: Trustworthy clinical advice is crucial but burdensome when seeking health support from professionals. Inaccessibility and financial burdens present obstacles to obtaining professional clinical advice, even when healthcare is available. Consequently, individuals often resort to self-diagnosis, utilizing medical materials to validate the health conditions of their families and friends. However, the convenient method of self-diagnosis requires a commitment to learning and is often not effective, presenting risks when individuals seek self-care approaches or treatment strategies without professional guidance. AI, supported by LLMs (LLM), may become a powerful yet risky self-diagnosis tool for clinical advice due to the hallucination of LLM, where it produces inaccurate yet deceiving information. Thus, can we trust the clinical advice from AI-driven LLMs like ChatGPT like ChatGPT4 for self-diagnosis? We examined this issue through a think-aloud observation: a patient uses GPT4 for self-diagnosis and clinical advice while a doctor assesses ChatGPT responses with their own expertise. After that, we conducted a semi-structured interview with the patient to understand their trust in AI-driven LLMs for clinical advice. we have concluded that the confounding factors influencing a patient's trust revolve around their competency-evaluation. Essentially, trust is equated with efficacy, which is determined by whether decisions made based on the AI agent's clinical advice and suggestion will effectively achieve the patient health goals. Patients tend to trust doctors more than AI agents due to this strategy, believing that educated, authorized doctors can provide effective medical guidance. This competency-based trust also explains why patients often perceive more experienced doctors as more trustworthy compared to less experienced ones.