Radiologists vs AI

[Jon]

The good news is after they’re pwned by the computer a few times they can probably learn to take discrepancies a bit more seriously.

[Doctor Phibes]

[George K]

I've seen reports of stuff like this in the past - before the term "AI" became popular. I suppose it's inevitable that machine learning can spot things that the human eye can't see or ignore.

This may be one of the reasons that Interventional Radiology appears to be getting more popular - it's literally "hands on" as opposed to "eyes on."

[Klaus]

@George-K I would imagine that a major benefit of AI-based CT scan diagnosis would be that one could feed it way more images than a human radiologist could ever handle. Like, say, a CT with ten images per millimeter, from toe to hair. Maybe for something like all-purpose cancer screening.

[Axtremus]

One barrier to human-AI collaboration is the inability of AI to “explain” its inferences. Given an input, an AI model can tell you this input maps to that output, but it cannot explain to you “why” it mapped the input to that particular output. “Explainable AI” is still a big, unresolved problem.

[Klaus]

I would say "explainable machine learning" is a big unresolved problem. Or, even more specifically, "explainable deep neural networks".

Other forms of AI are perfectly explainable. For instance, SAT solvers usually generate a concrete counterexample if they reject a formula.

[Aqua Letifer]

@Doctor-Phibes said in Radiologists vs AI:

@Jon said in Radiologists vs AI:

The good news is after they’re pwned by the computer a few times they can probably learn to take discrepancies a bit more seriously.

I'd guess that the next generation of pretty much everybody is going to be a lot more adept at working with AI

I called the radiology shit about 8 months ago. Just sayin'.