Study Reveals Inconsistency and Potential Harm in AI Chatbot’s Self-Diagnosis Accuracy, Caution Urged
ChatGPT, an advanced AI-driven chatbot utilized in various healthcare settings, has raised concerns regarding its accuracy in self-diagnosis, potentially posing harm to patients. Researchers from Japan and the United States conducted a study to evaluate the precision of ChatGPT’s diagnoses and its recommendations for medical consultation. The findings indicated that further development is required to enhance the accuracy of ChatGPT’s diagnoses and the degree to which it suggests seeking medical help.
The study, led by Tokyo Medical and Dental University (TMDU), focused on assessing ChatGPT’s response to five common orthopedic diseases. Orthopedic complaints account for a significant portion of patient care, with up to 26% of individuals seeking medical attention for such issues. The research team submitted the same questions to ChatGPT over the course of five days, analyzing the reproducibility between different days and researchers. Moreover, they evaluated the strength of the chatbot’s recommendation for medical consultation.
Lead author Tomoyuki Kuroiwa highlights the inconsistent performance of ChatGPT in diagnosing the five conditions. While the chatbot achieved 100% accuracy in diagnosing carpal tunnel syndrome, it only scored 4% for cervical myelopathy. The reproducibility between days and researchers also varied, ranging from poor to almost perfect across the five conditions, despite inputting the same questions repeatedly.
The study also uncovered inconsistencies in ChatGPT’s recommendation for medical consultation. While nearly 80% of the chatbot’s answers advised seeking medical help, merely 12.8% met the strong recommendation criteria established by the study. Kuroiwa expresses concerns that patients may become confused or experience harm from misdiagnosis due to the lack of clear language used in the self-diagnosis process.
Senior author Koji Fujita emphasizes the importance of straightforward language alerting patients to seek confirmation of a disease from medical professionals. He highlights that ChatGPT’s current version lacks both accuracy and precision in assisting patients with self-diagnosis. Fujita further notes several limitations of the study, such as the use of simulated questions instead of those derived from actual patients, the focus on only five orthopedic diseases, and the reliance on ChatGPT alone. However, training ChatGPT on specific diseases of interest in the future could improve its diagnostic abilities.
While the potential of AI in self-diagnosis is promising, the study highlights the need for caution and further research. Clear communication and expert medical opinions remain crucial for the confirmation of diseases to minimize the risks associated with misdiagnosis. As the role of AI as a diagnostic tool develops, future studies will provide valuable insights into its efficacy.
In conclusion, the recent study sheds light on the inconsistency and potential consequences of relying solely on ChatGPT for self-diagnosis. Although AI chatbots hold promise in healthcare, improvements in accuracy and precision are necessary to ensure patient safety. Combining the power of technology with medical expertise will be key in leveraging AI for reliable and effective diagnostic purposes.
