Not really, I think that you could however build one such thing with current tech, probably combining some LLM capabilities and some random forest or other such way to navigate the complexities of actual probabilistic diagnosis. Maybe use a transformer architecture but have symptoms in place of tokens and give the logits over possible diagnoses.
I know IBM has tried doing this, and supposedly always failed. I don’t know the details of their work, but I’m sort of perplexed about whether it really could have been so hard to produce something that at least performs at the level of a mediocre GP and knows when to say “I don’t know, refer to a specialist”. I worry that it might have been compared to a much higher bar than is sensible to use, and that much worse doctors than it retain their license just fine because no one tests them regularly against a diagnosis benchmark.
(anyway don’t worry about the miscommunication, I think the original point got a bit lost in the following comments and we drifted away from it)
Not really, I think that you could however build one such thing with current tech, probably combining some LLM capabilities and some random forest or other such way to navigate the complexities of actual probabilistic diagnosis. Maybe use a transformer architecture but have symptoms in place of tokens and give the logits over possible diagnoses.
I know IBM has tried doing this, and supposedly always failed. I don’t know the details of their work, but I’m sort of perplexed about whether it really could have been so hard to produce something that at least performs at the level of a mediocre GP and knows when to say “I don’t know, refer to a specialist”. I worry that it might have been compared to a much higher bar than is sensible to use, and that much worse doctors than it retain their license just fine because no one tests them regularly against a diagnosis benchmark.
(anyway don’t worry about the miscommunication, I think the original point got a bit lost in the following comments and we drifted away from it)