Reliability of LLMs as medical assistants for the general public: a randomized preregistered study
Надёжность больших языковых моделей в качестве медицинских помощников для широкой общественности: рандомизированное пререгистрированное исследование
2026-02-01
SCID: 54.1/6fmgwxsw
Discuss with AI
human user testinglarge language modelsmedical advicemedical decision-makingrandomized preregistered study
Figures from the paper
Abstract (AI)
Global healthcare providers are exploring the use of large language models (LLMs) to provide medical advice to the public. LLMs now achieve nearly perfect scores on medical licensing exams, but this does not necessarily translate to accurate performance in real-world settings. We tested whether LLMs can assist members of the public in identifying underlying conditions and choosing a course of action (disposition) in ten medical scenarios in a controlled study with 1,298 participants. Participants were randomly assigned to receive assistance from an LLM (GPT-4o, Llama 3, Command R+) or a source of their choice (control). Tested alone, LLMs complete the scenarios accurately, correctly identifying conditions in 94.9% of cases and disposition in 56.3% on average. However, participants using the same LLMs identified relevant conditions in fewer than 34.5% of cases and disposition in fewer than 44.2%, both no better than the control group. We identify user interactions as a challenge to the deployment of LLMs for medical advice. Standard benchmarks for medical knowledge and simulated patient interactions do not predict the failures we find with human participants. Moving forward, we recommend systematic human user testing to evaluate interactive capabilities before public deployments in healthcare.
Key Findings
1
In a randomized preregistered study of 1,298 participants, LLMs were evaluated for identifying medical conditions and recommending dispositions across ten scenarios.
2
LLM assistance performed no better than participants using a source of their choice, revealing user interaction as a major deployment challenge.
3
Medical knowledge benchmarks and simulated patient interactions failed to predict the models’ poorer performance with human participants; systematic human user testing is recommended before healthcare deployment.
4
Participants assisted by the same LLMs identified relevant conditions in fewer than 34.5% of cases and correct dispositions in fewer than 44.2%.
5
Tested independently, GPT-4o, Llama 3, and Command R+ identified underlying conditions in 94.9% of cases but selected the correct disposition in only 56.3%.
Research Object
LLM-assisted medical advice to members of the general public in ten medical scenarios
Research Subject
Reliability of LLMs and human–LLM interactions in identifying underlying conditions and selecting appropriate disposition
Publication Details
Publication Date
2026-02-01
Journal
Publisher
ISSN
Cited by
103
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest
References available in scid.ai4
Large language models encode clinical knowledge2023
Comparing Physician and Artificial Intelligence Chatbot Responses to Patient Questions Posted to a Public Social Media Forum2023
Benefits, Limits, and Risks of GPT-4 as an AI Chatbot for Medicine2023
AgentClinic: a multimodal benchmark for tool-using clinical AI agents2026