Is ChatGPT coming for doctors next?

In addition to revolutionizing our ability to write, search, and learn, the conversational AI bot ChatGPT is also having an impact on the way we conduct web searches. Is ChatGPT coming for doctors next

We’re having a moment with chatbot ChatGPT, which aims to change the way we create content, conduct web searches, and obtain information.

How did ChatGPT do recently? It was almost able to pass the US Medical Licensing Exam (USMLE), which is very challenging, usually taking 300 to 400 hours to prepare for, and includes everything from basic science to bioethics.

ChatGPT’s excellent performance in handling the USMLE demonstrates the potential for these AI bots to become helpful in medical education and even diagnosis in the future.

” The researchers report in the PLOS Digital Health journal that ChatGPT performed well on all three tests without any special training or reinforcement.

Is ChatGPT coming for doctors next

“In addition, ChatGPT demonstrated high levels of concordance and insight in its explanations.”

LLMs, or artificial intelligence, are large language models like ChatGPT. In a similar fashion to your phone’s predictive text feature, these LLMs are designed specifically for written communication. They can predict how words should fit together in a phrase based on thousands of sample texts and sophisticated algorithms.

Even though that’s a bit of an oversimplification, the point is: ChatGPT doesn’t actually know much about anything, but it can create phrases that seem reasonable on almost any topic by studying a lot of web content.

It is nonetheless important to emphasize the phrase “plausible-sounding”. Different wording may make the AI appear eerily intelligent or draw ridiculous conclusions, depending on the likelihood of the different wording.

The study

USMLE sample questions were used by researchers from Ansible Health to test ChatGPT, after making sure the answers couldn’t be found on Google. They could then rest assured that ChatGPT was providing fresh answers based on the data they had taught it.

A total score of 52.4% to 75% was obtained by ChatGPT when put to the test. (Usually, 60% is the pass mark). It was defined as a major insight in 88.9% of the responses when it was something novel, non-obvious, and clinically relevant.

“This marks a major milestone in clinical AI maturation for clinical AI to pass one of the hardest expert exams without any human assistance,” the study authors said.

Aside from being remarkably consistent in its responses, ChatGPT even showed the ability to justify its responses. In addition, it had a higher accuracy rate than PubMedGPT, a bot trained solely on medical literature.

It is important to remember that ChatGPT was trained on inaccurate data. Upon direct questioning, the bot will acknowledge that more work is needed to improve the reliability of LLMs. Science Alert’s David Nield believes it will never take the place of medical professionals in the near future.

The analysis of internet knowledge still has a lot of room for improvement, especially as these AI bots continue to improve. Rather than replacing humans, they may become critical aids in the field of medicine.

“Research suggests large language models may improve clinical decision-making and medical education,” wrote the researchers.