From experience as a patient, I want doctors who can think from first principles, and have the data to assist. But truthfully "first principles" in biology is the domain of career scientists who study specific systems, not doctors whose domain is really praxis. Most doctors don't have the time or training to be anywhere near the vanguard, with the exception of hyper-specialists and scientist-clinicians. There needs to be a better method for doctors to leverage biological knowledge on a case-by-case basis — and to be able to use data *contextualized to that biological knowledge* to inform their diagnosis, treatment, and more generally clinical decision-making.
I do think that generative language models could turn the tide, and maybe even in the next few years. GPT-4 or GPT-5 may have enough capacity to make them trainable with little subsequent supervision needed. Generally, I've high hopes for that happening across many fields, with the models being domain-agnostic. That would speed us up tremendously.
From experience as a patient, I want doctors who can think from first principles, and have the data to assist. But truthfully "first principles" in biology is the domain of career scientists who study specific systems, not doctors whose domain is really praxis. Most doctors don't have the time or training to be anywhere near the vanguard, with the exception of hyper-specialists and scientist-clinicians. There needs to be a better method for doctors to leverage biological knowledge on a case-by-case basis — and to be able to use data *contextualized to that biological knowledge* to inform their diagnosis, treatment, and more generally clinical decision-making.
I do think that generative language models could turn the tide, and maybe even in the next few years. GPT-4 or GPT-5 may have enough capacity to make them trainable with little subsequent supervision needed. Generally, I've high hopes for that happening across many fields, with the models being domain-agnostic. That would speed us up tremendously.