Artificial intelligence makes 80% errors in initial patient diagnosis
A study published in JAMA Network Open revealed that artificial intelligence models from OpenAI, Google, and DeepSeek frequently err due to insufficient data when making initial diagnoses. An analysis of 21 models showed an error rate exceeding 80% in differential diagnosis.
Researchers highlight that AI models struggle to generate comprehensive lists of diagnostic hypotheses and tend to prematurely narrow diagnostic considerations, leading to high error rates.
With complete information, model accuracy improves significantly: errors drop below 40%, and the best systems achieve about 90% accuracy. Clinical epidemiologist Sanjay Kinra from the London School of Hygiene notes that, although promising, these technologies are not yet a replacement for doctors.
Overall, the study underscores the need to enhance algorithms and provide higher quality, more complete data to optimize AI performance in medical practice.
It is expected that with further development and integration, AI usage in medicine will improve, but treatment decisions should still remain the responsibility of human professionals.
