Warning: Scientist rant ahead! Maybe we'll just have some "bad" AI like we have some "bad" doctors? The scary part of the medscape article was the last source where the student doctor decried the lack of study in their coursework on how to use AI in their medical practice. Although the specialized medical AIs may do well in controlled patient populations and study settings when directly supervised, it does not seem that we are at the "just let the AI be the doc" stage. Even for entering the orders or health information, you'd have to check that it was correct and at some point it is just easier to do it yourself. I've tried, within the last month, having AI 1) summarize a journal article, 2) list the GPCRs that do not have seven transmembrane domains, and 3) write an email requesting reagents from the author of a journal article. Outcomes: 1) Summarized, but missed both the importance of the article to the field and the key elements that made the paper a major advance in medicine--in other words, it failed to integrate the new knowledge to the existing body of work. Also failed to find any weaknesses or flaws even when directly asked to analyze for this aspect. Every study has some weakness or flaw. 2) Hallucinated and made false statements while listing a reference that directly contradicted the AI's conclusion. Failed to find breakthrough in field (from eleven years ago in one of the top science journals) that found that plants have this feature. 3) Wrote an email, but failed to list the reagents correctly. Using this email would have required extensive editing and hunting in the paper and supplementary information on the journal's website for the correct name of each reagent.
Yes, AI is shockingly better than it was even one year ago. Does that mean it is "good enough" to determine your treatment? Who could your family sue if you were completely incapacitated as a result of the treatment? Would AI learn from being sued and the associated "costs"?
Thanks, my rant made me feel better about the importance of human minds in combining both breadth and depth of knowledge and applying them to complex endeavors.