Amusingly this is why people say LLMs will beat doctors. It’s because the 90% of cases is so easy that a motivated guy with Google can get there and a smart NP can get there too.
It isn’t that it’s easy to do all a doctor does. But their training and knowledge shines in the 99th percentile case except they never exercise it there so you can usually get there with Google.
“Oh but an LLM will guess the common case and never think of the rare!”
Yeah but so will a doctor given 10 minutes on it. They’re not exactly going to House MD you. You’re gonna die.
I discovered a friend’s chronic medical issue that two VA specialists and a PCP couldn’t figure out using an LLM health project that had been posted here. Works when it works, n=1. Certainly, don’t trust the robot, but it doesn’t hurt to rubber duck debug with it to find blind spots. Fancy search engine sometimes is right (although it can lie too!).
Language models are really good at free association tasks, such as semantic fuzzy search. Next token prediction is among the worst possible ways to use them (although if there's no other obvious way of getting the information out of the model, it works in a pinch).
I think a great use case for AI is to act as triage for a new case so that it can send you to the right specialist and have them evaluate you. It could potentially remove the need to see a GP for a referral to a specialist, thus freeing the GP up to spend more time treating others.
You don't need a GP for referral to a specialist currently. You just have to pay. It's what I do.
But I understand what you're saying. Insurance gates these but could do so with their own tech rather than relying on the third party. Could help with keeping loss ratios at the minimum.
>Insurance gates these but could do so with their own tech rather than relying on the third party. Could help with keeping loss ratios at the minimum.
I work in insurance. In my experience, the fact that you have to go to the doctor for a referral discourages people from getting said referral.
So the tradeoff is that you would get fewer referral-specific visits (i.e. person going to their GP to get a specialist referral) at the likely expense of more specialist visits.
In my experience (arthritis), specialists typically won't take direct appointments without a referral. They don't want to triage the 90%, and most offices are booked weeks-to-months in advance...
I've seen a lot of doctors who insist patients must be making up things when they say "but XYZ", and my question becomes...so what?
If someone credibly lies to you and gets codeine or ritalin or something once or twice...that's not really significant, in terms of negative outcomes.
If someone lies to reach a medical professional, then you treat them like any other bad customer interaction and stop doing business with them after some point.
You're going to have to choose, denying people in need directly, or doing it indirectly because other people game the process and get in the way of those in need.
The length of the queue doesn't determine the length of the priority queue for urgent issues necessarily, and more accessible preventative care could hypothetically go a long way toward reducing the demand on first-contact touchpoints in the US's healthcare.
It isn’t that it’s easy to do all a doctor does. But their training and knowledge shines in the 99th percentile case except they never exercise it there so you can usually get there with Google.
“Oh but an LLM will guess the common case and never think of the rare!”
Yeah but so will a doctor given 10 minutes on it. They’re not exactly going to House MD you. You’re gonna die.