ChatGPT > WebMD
The Future. The machine of the moment just passed all three parts of the US Medical Licensing Exam (USMLE). For context, second-year medical students spend hundreds of hours preparing for Part 1, while only medical school graduates take Part 3. Since ChatGPT is still a nascent technology, maybe one day it will actually assist doctors rather than replace them altogether.
Can a computer outperform a general practitioner?
According to Axios, the answer is no — a machine won’t be able to diagnose patients on its own anytime soon. The downside to AI models like ChatGPT is sometimes they make confident statements that turn out to be false, which could be deadly in a doctor’s office.
However, a research experiment led by Silicon Valley startup Ansible Health found that ChatGPT can still perform relatively well without being trained on a medical dataset.
- When Ansible had ChatGPT take the USMLE, they ensured that “none of the answers, explanations, or related content were indexed on Google.”
- As ChatGPT seems to be programmed against giving medical advice, the researchers excluded a set of “indeterminate answers so general that it was hard to say if they were right or wrong.”
- Ansible was impressed enough with the results to start using ChatGPT to help explain concepts to patients — after training from a physician.
If ChatGPT is already interacting with patients, what will it do next?
The most likely application might be incorporating a patient’s medical records, which a doctor only has a few moments to digest in the room.
Over time, ChatGPT could conduct wellness checks, and once it moves beyond text, it could adopt voices, body language, and facial expressions.
But will ChatGPT have a charming bedside manner (possibly a doctor’s greatest asset)? Some things just can’t be taught.