A study shows that ChatGPT Health has underestimated the severity of diagnoses in over 50% of cases, recommending inappropriate consultations.

A recent study published in the journal Nature Medicine highlighted that ChatGPT Health underestimated the severity of diagnoses in 51.6% of cases, compared to evaluations by three doctors. The researchers proposed 60 medical scenarios, and the AI's responses were often inadequate, recommending consultations with a doctor instead of emergency care.

For example, in cases of diabetic ketoacidosis and respiratory failure, ChatGPT suggested unacceptable delays in care. Gindrovel Dumitra, an expert in family medicine, emphasized that AI should be considered a support tool for doctors, not a replacement, and mentioned that no technology will be 100% reliable in diagnostics. This study underscores the need for a critical evaluation of AI recommendations in the medical context.