What happens when ChatGPT’s second opinion overrules your doctor’s?

Anthony Pompliano’s tweet about a friend of his who decided to check his doctor’s diagnosis on ChatGPT — which then made a different, correct, diagnosis — has prompted many people to share similar experiences. The issue shows that when the technology is available, people are going to use it, regardless of the risks. A 2024 study from Live Science indicates ChatGPT’s medical diagnoses are accurate less than half the time, suggesting its success in these cases may be exceptional rather than typical.

This is essentially a logical continuation of the “Doctor Google” syndrome: entering one’s symptoms in the search engine to obtain a diagnosis and then discuss it with one’s bemused doctor. As generative algorithms replace traditional search engines, we can expect to hear many more such cases as Pompliano cites.

What’s the difference between a doctor’s diagnosis and a generative algorithm’s? In both cases there’s a diagnosis based on medical knowledge, but in the case of the algorithm, the diagnosis, depending on how well it has been trained, will tend to explore more possibilities, however remote. For example, in the case of a diagnosis that seems to make no sense, such as a patient in the UK showing symptoms of dengue, a tropical disease.