SPAIN — Aranzazu Vinas, Helena Matute, and Fernando Blanco published findings in the journal PLOS Digital Health on July 9, 2026. Their study revealed that physicians tended to trust artificial intelligence (AI) treatment recommendations, even when patient outcomes contradicted those recommendations.

The study analyzed data from 223 physicians who participated anonymously in online experiments. These experiments involved physicians making decisions about treating hypothetical patients for a rare disease, using a hypothetical treatment described as still under development and not yet proven. Physicians were informed that an AI system identified which patients were more or less likely to benefit from the treatment. After choosing which patients to treat, participants were presented with data on patient recovery and then rated their perceptions of the AI's reliability.

In one experiment, the hypothetical treatment was equally and moderately effective for all patients, despite AI recommendations suggesting otherwise. A second experiment featured a hypothetical treatment that was equally ineffective for all patients; here too, the actual effectiveness did not align with the AI recommendations. In both experimental conditions, the physicians generally rated the AI system as reliable.

Physicians did not use the patient recovery data to conclude that the AI recommendations were incorrect. In the second experiment, physicians did not realize that the treatment was entirely ineffective. Aranzazu Vinas, lead author of the study, stated, "In both experiments, physicians mostly trusted the AI's classifications and had trouble learning from the feedback." She added, "Furthermore, in the second experiment, professionals did not notice that the treatment was completely ineffective."

Helena Matute, a co-author, explained the implications of the findings. Matute said, "People tend to say that there is always a human controlling the algorithm, but our experiments show that doctors (as well as anyone else) have problems in learning from the available evidence when it contradicts the suggestions of an algorithm." Fernando Blanco, also a co-author, stated, "It is important to investigate the errors that humans (including doctors) make when working with algorithms, in order to learn how to minimize the problems that arise from them." Vinas is affiliated with the University of the Basque Country.