Artificial intelligence fails in primary diagnosis in over 80% of cases, study shows

Artificial intelligence fails in primary diagnosis in over 80% of cases, study shows. This was reported by Euronews. It appears that artificial intelligence systems perform well in providing a final diagnosis when they have complete information, but encounter serious difficulties in the initial stages of the diagnostic process when data is limited, a new study shows.

“These systems are very good at naming a final diagnosis when the information is complete, but they experience difficulties at the beginning of a case when the available data is scarce,” notes one of the researchers.

The analysis shows that all AI systems studied fail to offer an adequate differential diagnosis in over 80% of cases. This is a key stage in medical practice, during which doctors consider various possible diseases based on symptoms.

When providing a final diagnosis, the results are significantly better—the success rate varies between approximately 60% and over 90%, depending on the specific system.

Most of these solutions show higher accuracy when laboratory results and imaging studies are added to the textual information.

Among the best-performing systems are a group including Grok 4, GPT-5, GPT-4.5, Claude 4.5 Opus, Gemini 3.0 Flash, and Gemini 3.0 Pro.

Despite improvements in newer versions and progress in solutions optimized for logical reasoning, researchers emphasize that currently available artificial intelligence systems still do not reach the level necessary for safe application in clinical practice.

“Our results confirm that the use of artificial intelligence in healthcare requires mandatory human involvement and very strict control,” states researcher Suchi.

The same opinion is expressed by Dr. Susana Manso García of the Spanish Society of Family and Community Medicine, who was not involved in the study.

“The study itself emphasizes that these technologies should not be used for clinical decision-making without supervision. Although artificial intelligence is a promising tool, the clinical judgment of the physician remains indispensable,” she points out.

“The recommendation to the public is to use these technologies with caution and to always consult a specialist for health problems,” the expert adds.