Can GPT-4 replace doctors? This AI is able to give correct diagnosis to patient in few seconds

The GPT-4 may be used in the medical field in the future. Dr. Isaac Kohane, a specialist in information technology and medicine at Harvard University, and his colleagues have found that this artificial intelligence (AI) has outperformed many doctors in making a diagnosis. It even passed the exam necessary to get a doctor's license.

The GPT-4 answered medical licensure exam questions correctly 90% of the time, and its answers outperformed not only the ChatGPT based on GPT-3 and GPT-3.5, but also some practicing physicians who already have a license to practice medicine.

The GPT-4 has shown that it not only knows many medical facts, but can also be an excellent translator. This AI understands very well a patient's medical history written in Portuguese. It was able to transform that terminology-filled technical text into something that even a sixth grader could understand.

Also, this AI offered useful tips to doctors on how to properly communicate with patients.

In addition, the GPT-4 performed well at processing reports on extensive research, summing up their content instantly.

This AI has presented its answers in such a way that at first glance it may seem that real intelligence is talking to the researchers, just like human intelligence. But if being watchful, one can notice that this AI is still limited by templates in its judgments and does not know how to fully justify anything by understanding cause and effect relationships.

The researchers gave this AI real patient data to see if it could handle it like a doctor with years of experience and years of research would. And, to their surprise, this AI was able to diagnose a rare disease in the patient.

Dr. Kohane recounted this experience in the book AI Revolution in Medicine, which he co-authored with freelance journalist Carey Goldberg and Microsoft Vice President of Research Peter Lee.

As per Kohane, on the one hand, it makes sense to be happy that such a useful tool will be available to millions of people in the foreseeable future, but on the other hand, it is still quite difficult to imagine how to make this platform safe for users because the GPT-4 does not always provide correct answers, and Kohane's book cites many examples of its errors.

According to the authors of the book, to minimize the risk of major errors, one can spend several sessions with the GPT-4 so that this AI can re-read or check its own work with a fresh eye. Also, it would be good to entrust the inspection of its work to a specialist who understands the problem.