I used to think that the benefit of using AI in medicine, specifically radiology, was plausible… There is SO much data in a CT scan or MRI that human eyes just can’t see, you lose soo much granularity converting a 3D scan to a 2D image on a screen so some bloodshot doctor can look at it…
AI should, in theory, be able to read a scan and find pinprick tumors or irregularities and at least tell the doctors where to look…
But now, watching LLM’s do their thing, I’m now convinced that there isn’t an AI around that is worth it’s weight in horse-shit.
AI can be a force multiplier for doctors. The catch is it needs to be doctor led. It’s quite trivial for a doctor to dismiss the obvious hallucinations. If it also points them towards more subtle signs they would potentially miss, then that’s a net win.
I think preventative medicine is where it would shine the most. There’s so much data that we can capture with little additional cost. It just wouldn’t be data that a human can reasonably make use of.
The problem is mainly who we can trust with this data.
Where AI could be useful, it needs significant investment, if paired with resources it can look for patterns in scans.
However the process an AI uses to make a decision is similar to encoding of an image, so I wonder how it compares in terms of resolution to high resolution pictures produced by MRIs
I used to think that the benefit of using AI in medicine, specifically radiology, was plausible… There is SO much data in a CT scan or MRI that human eyes just can’t see, you lose soo much granularity converting a 3D scan to a 2D image on a screen so some bloodshot doctor can look at it…
AI should, in theory, be able to read a scan and find pinprick tumors or irregularities and at least tell the doctors where to look…
But now, watching LLM’s do their thing, I’m now convinced that there isn’t an AI around that is worth it’s weight in horse-shit.
LLMs are not the same as neural network machine based image sorting. It’s been proven more accurate than pathologists.
I would still want a human being to verify the diagnosis before anything gets cut. ;-)
Pathology error rates are too high, everyone should be getting a second opinion and all pathology reports should be cross-checked.
Analytical AI is much more useful than generative AI and even if it were LLMs doing the work, with a doctor checking it can be very beneficial
And if they do misdiagnose one in five patients, well, they shouldn’t have gotten infected, it’s their fault.
AI can be a force multiplier for doctors. The catch is it needs to be doctor led. It’s quite trivial for a doctor to dismiss the obvious hallucinations. If it also points them towards more subtle signs they would potentially miss, then that’s a net win.
That’s why you do everything with human review, I said that in my comment, come on now
Doctors definitely need no help screwing up that much
Luckily it isn’t LLMs doing those tasks
I think preventative medicine is where it would shine the most. There’s so much data that we can capture with little additional cost. It just wouldn’t be data that a human can reasonably make use of.
The problem is mainly who we can trust with this data.
Oh absolutely. Like the “cologard” test… an early screening to identify a problem before it becomes a problem.
Where AI could be useful, it needs significant investment, if paired with resources it can look for patterns in scans.
However the process an AI uses to make a decision is similar to encoding of an image, so I wonder how it compares in terms of resolution to high resolution pictures produced by MRIs
Yeah, I think it should never be involved in decision-making… Pointing out a place for doctors to explore maybe…actually diagnosing? no.