Google’s medical AI, Med-PaLM, has just passed a health competition in the United States which allows you to practice as a doctor.
AI and health are linked by a link that was very quickly explored by many players in the sector in order to make the best use of the exceptional data processing capacities of the technology. But the usefulness of artificial intelligence in the field could go even further in the future when we see what Google’s is already capable of.
“I passed medicine! »
Since the emergence of ChatGPT, then Google Bard, a test exercise for these chatbots has been particularly popular: giving them school exams. We had thus been able to see ChatGPT pass several impressive tests, until simply obtaining the bar in the United States.
But it may be a new stage that has just passed today. Google’s medical AI, Med-PaLM, currently being tested in hospitals, has also just passed the US Medical Licensing Examination (USMLE), an exam that allows students to obtain their license to practice medicine. In order to be accepted, it is necessary to obtain more than 60% of correct answers. And Med-PaLM was credited with a score of 67.6%!
Not bad, but could do better
At first glance, the result is exceptional, especially since according to Google, it is the very first AI to be able to boast of this success. Yet, according to the findings of the study itself, Med-PaLM results are encouraging, but remain lower than those of clinicians “. In another study, which remained at the preprint stage, the improved version Med-PaLM 2 would have obtained 86.5% of correct answers.
But this exemplary schooling is obviously not enough to entrust the life of a patient to artificial intelligence. As Professor James Davenport of the University of Bath reminds us, we must not confuse ” medical matters and medicine proper, which includes the diagnosis and treatment of a real health problem “. Except that if the AI already has enough correct data, won’t it be able to quickly learn to use it wisely?
Source : The Straits Time
6