We have written blogs recently that described how AI is being applied to medical issues, particularly with regard to radiology imaging. It seems now that in every issue of various internet medical journals there is another article about how AI is being utilized in many aspects of medical decision-making. With all of these advances, will AI give superpowers to doctors?
Maybe not, but this headline from a March 14, 2023 Medpage Today issue was still rather shocking: “Google AI Performs at ‘Expert’ Level on U.S. Medical Licensing Exam.” The article disclosed that Google’s AI model achieved 85% accuracy on a sample U.S. Medical Licensing Examination. The AI model achieved this high score on both multiple choice and open-ended questions, even giving explanations for its answers, and then evaluated its own responses.
One of the physicians who worked on the project said: “If AI can give caregivers back the gift of time, if AI can enable doctors and other caregivers more time with their patients and bring time and humanity to medicine, and if it can increase accessibility and availability for people, that’s our goal.” He also said this tool will be seen in the future as giving clinicians a kind of superpower.
Earlier tests of AI’s ability to pass the medical licensing exam produced good results, but this particular project used “large language models” that were focused on medical information. In effect, the project was able to use specific medical expertise to help determine exactly how to train the AI models. Dr. Vivek Natarajan, a research scientist at Google Health AI, noted that the model “really learned quickly about the nuances of the safety of the medical domain and aligned itself very fast.”
Is it only me, or is it strangely unsettling to read articles about how an AI bot can be “taught” to read sophisticated and complex medical articles, and then “learn quickly” how to use that knowledge to answer nuanced medical questions? Indeed, as noted in an earlier blog post, the “brave new world” is rapidly becoming our real world, and we had better be able to teach AI to help us before it figures out how to overtake us.