Will there be an AI that diagnoses patients better than an average doctor prior to 2030?
41
88
1k
2029
82%
chance

This question will resolve to yes if at some point before 2030 there is an AI whose diagnosis are in average more accurate than those made by human doctors.

Get Ṁ600 play money
Sort by:

How general does the domain have to be?

AFAIK there are already AI doctors that can outperform humans at determining whether a tumor is malignant - I'm guessing you're asking something more like whether there's an AI that can outperform the average human GP at their entire job, but I'm also guessing you don't necessarily expect the AI to be able to verbally converse with patients over audio in real time to collect information for the early stages of the diagnosis.

Would a text-channel-only AI that outperforms human doctors only at the task of text-channel-only diagnosis count? Presumably an AI that only beats GPs and can't refer you to a specialist the way a GP would if you had a weird problem, counts?

(Important disclaimer: I have no idea what effect being able to see and talk to their patients actually has on human GP diagnostic performance.)

Do you know where your performance benchmarks for existing GPs are coming from, or are you just waiting for someone to publish test results saying their AI outperformed a sample of human doctors?

@adamzeelen Agreed. More clarity is needed here before I invest. Depending on the resolution criteria this may already be a YES.

predicts YES

@adamzeelen I was thinking in an AI that diagnoses through text and outperforms the average of human doctors who are able to see and talk to their patients. And I'm expecting a scientific test that proves this, I assume that if an AI is able to do this at some point there will be at least someone who writes about it.

@LuisV : the problem is how to define "outperform". A relevant metric would be: compare clinical outcomes of randomly chosen patients diagnosed by human doctors vs. those diagnosed by AI. But the chances of AI achieving superiority by that metric by 2030 are basically zero, not because of AI, but because it would take a very long time (not to mention money) to design and run an extraordinarily complex and ethically challenging study like that. There is literally no time to do this by 2030, even if the superior AI was available today (which it isn't).

@LjubomirButurovic agreed on the technical legal points

Otoh if there is a medical ai to system running in international waters which all rich people and even middle class people fly to visit and consult, and are happy about it, that'd be very convincing to me

More related questions