The examples I’ve read about so far, from LLM’s being used to handle patient records to being used to take notes during patient visits, just seem particularly ripe for serious fuckups considering we know how often generative AI simply makes mistakes.
Not that it matters. It isn’t meant to be good. It’s meant to be bought.
I would kill for a doctor/therapist/dentist with no prejudices, biases, or religion. And while AI should be able to help us with that; it’ll never happen.
There’s this fairly common belief / misconception how AI is neutral, or more logical and less biased than a human, because it runs on a computer, which is a calculator / a machine… But pretty much the opposite of it is true. It’ll try to learn all biases, stereotypes, prejudices and shortcomings from the training data. And it’s really hard to mitigate for that, make it learn the correct pattern. Curate datasets which aren’t skewed in any way, reflect all cases and population groups…
We had all those articles. Go ask an AI to draw a picture of a doctor. And it’ll come up a white male. That’s how it works.
And what percentage of doctors are white male?
Good question? I’d say half of the general practicioners I know are female. It’s a bit of a different story with people working in hospitals, I think the doctors there are more male in my experience. But you’re guaranteed to meet women as well, if you spend any time at the hospital.
And they’re all mostly white I guess. But then I live in a region that is predominantly made up of white people, so idk what that’s supposed to tell us. It has little to do with the doctors themselves. I suppose if I were to live in South Korea, the doctors would predominantly be of a different ethnicity.
But I’m not sure what your question is aimed at. If we apply it to an AI model of reality… We have to then factor in all the biases and apply it correctly. Make sure it’s never used by general practicioners, just in hospitals. Make sure it’s only used in the west of Germany or wherever… And it must not learn to then go ahead and draw all doctors male, because that’s the best chance to score high on the reward function. Otherwise it just won’t depict reality and you’re left with an AI model that does wrong classifications. And then, even if it were, once you happen to look at one of the (rare) case where someone is female, it’ll be in some territory where the predictions are more volatile and not necessarily rooted in facts or the real causal relations.
What makes you think an AI would have no prejudices, biases, or religious slants even if it didn’t hallucinate everything it said?
The point is that there are a LOT of poor doctors that should be avoided and not everyone can afford second opinions.
The bar to remove a medical license is far too high.
