Comic Strips

iirc the reason it isn't used still is because even with it being trained by highly skilled professionals, it had some pretty bad biases with race and gender, and was only as accurate as it was with white, male patients.

Plus the publicly released results were fairly cherry picked for their quality.

That's because the medical one (particularly good ar spotti g cancerous cell clusters) was a pattern and image recognition ai not a plagiarism machine spewing out fresh word salad.

LLMs are not AI

To expand on this a bit AI in medicine is getting super good at cancer screening in specific use cases.

People now heavily associate it with LLMs hallucinating and speaking out of their ass but forget about how AI completely destroys people at chess. AI is already getting better than top physics models at weather predicting, hurricane paths, protein folding and a lot of other use cases.

AI's uses in specific well defined problems with a specific outcome can potentially become way more accurate than any human can. It's not so much about removing humans but handing humans tools to make medicine both more effective and efficient at the same time.

My favourite story about it was that one time when neural network trained on x-rays to recognise tumors I think, was performing amazingly at study, better than any human could.
Later it turned out that the network trained on real life x-rays with confirmed cases, and it was looking for penmarks. Penmarks mean the photo was studied by several doctors, which mean it's more likely to be the case that needed second opinion, which more often than not means there is a tumour. Which obviously means that if the case wasn't studied by humans before, the machine performed worse than random chance.
That's the problem with neural networks, it's incredibly hard to figure out what exactly is happening under the hood, and you can never be sure about anything.
And I'm not even talking about LLM, those are completely different level of bullshit

The important thing to know here is that those AI were trained by very experienced radiologists who are physicians that specialize in reading imaging. The AI's wouldn't have this capability if the humans didn't train them.

Also, the imaging that AI performs well with is fairly specific, and there are many kinds of imaging techniques and diagnostic applications that the AI is still very bad at.

One of the large issues was while they had very good rates of correct diagnosis, they also had higher false positive rates. A false cancer diagnosis can seriously hurt people for example

Yeah this is one of the few tasks that AI is really good at. It's not perfect and it should always have a human doctor to double check the findings, but diagnostics is something AI can greatly assist with.

Also, there'll be no one to blame for mistakes! Failures are just software errors and can be shrugged off! Increase profits and pay less for insurance! What's not to like?

Imagine an episode of House, but everyone except House is an AI. And he's getting more and more frustrated by them spewing nonsense after nonsense, while they get more and more appeasing.

"You idiot AI, it is not lupus! It is never lupus!"

"I am very sorry, you are right. The condition referred to Lupus does obviously not exist, and I am sorry that I wasted your time with this incorrect suggestion. Further analysis of the patient's condition leads me to suspect it is lupus."

They don't use the generative models for this. The AI's that do this kind of work are trained on carefully curated data and have a very narrow scope that they are good at.

Image categorisation AI, or convolutional neural networks, have been in use since well before LLMs and other generative AI. Some medical imaging machines use this technology to highlight features such as specific organs in a scan. CNNs could likely be trained to be extremely proficient and reading X-rays, CT, MRI scans, but these are generally the less operator dependant types of scan, though they can get complicated. An ultrasound for example is highly dependent on the skill of the operator and in certain circumstances things can be made to look worse or better than they are.

I don't know why the technology hasn't become more widespread in the domain. Probably because radiologists are paid really well and have a vested interest in preventing it... they're not going to want to tag the images for their replacement. It's probably also because medical data is hard to get permission for, to ethically train such a model you would need to ask every patient in for every type of scan it their images can be used for medical research which is just another form/hurdle to jump over for everyone.

It's certainly not as bad as the problems generative AI tend to have, but it's still difficult to avoid strange and/or subtle biases.

Very promising technology, but likely to be good at diagnosing problems in Californian students and very hit-and-miss with demographics which don't tend to sign up for studies in silicon valley

Basically AI is generally a decent answer to the needle in a haystack problem. Sure, a human with infinite time and attention can find the needle and perhaps more accurately than an AI could, but practically speaking if there's just 10 needles in a haystack it's considered a lost cause to find any of them.

With AI it might find in that same stack 30 needles, of which only 7 of them are the needles, which means the AI finds more wrong answers than right, but ultimately you do end up finding 7 needles when you would have missed all 10 before, coming out ahead.

So long as you don't let an AI rule out review of a scan that a human really would have reviewed, it seems a win to potentially have more overall scans get a decent review and maybe catch things earlier in otherwise impractical preventative scans

That's correct — and you're right to point out this common reply by AI chat boxes. Let's breakdown why that happens:

📝 LLMs are predictive models: When a specific pattern shows up a lot in the training data set — like your example reply, the LLM will be more likely to reply in a similar way in the future, just like when people walk through a patch of grass and create a visible path. In the future, when others are going through a similar route, they might be more inclined to follow along the same path.

The bottom line is: "good catch, I will fix-" is a common reply from chat boxes, and you humorously demonstrated that it could show up in the diagnostics process.

The loss is the jobs we lost along the way.

Booring. Find a new joke.