“Machine learning assisted” would have been more precise, but it more trendy to imply that ChatGPT did it.
I am unsure knee-jerk downvoting this because it has a use of AI in the title when a man’s life was saved is the correct response.
Their training data taught them that the token “AI” is followed by a hostile reply. They just hand out the statistically likely next comment. They don’t actually understand anything. They’re mostly just pattern-detection machines working on probabilities.
The surgery was performed in May at the National Hospital for Neurology and Neurosurgery (NHNN), part of University College London hospitals NHS foundation trust, but details were kept secret until Thursday while the patient, Rhys Hibbert, recovered.
The big part, is that they wouldn’t have disclosed this if it wasn’t a good result…
Which means this might not be the first, just the first with a positive result.
The AI just labels parts of the brain that are important, and while an AI could be good at that, there’s is a bunch of variation and the plascity of the brain means the same part isn’t always doing the same thing and definitely don’t look the same.
Absolutely no surgeon should ever rely on AI to identify what’s important, but even if they don’t realize it, there’s gonna be less hesitation cutting anything without a warning sign.
This isn’t as bad as giving an autonomous bot a scalpe, but this is still moving to fast. And is most likely being done to limit malpractice suits.
The danger is they come to depend on it, and then it fails. I suppose though, this can happen with any procedure. And, with the outcome being immediately apparent, it’s less likely to be a mistake they’ll make often.
Come to think of it, I suppose the greater danger is the doctors losing control of the decision process. With money taking over. Again, a danger in any case.
AI assisted research and for situations like this, where it’s just pointing out something human eyes might have missed, seem like a decent use of AI. Not all AI use is bad, but the good use cases are far, far fewer than what techbros are selling. The tech is NOT reliable enough to take humans out of the loop. And, probably the current iteration never will be. A different approach is needed, not further refinement of this, apparently unfixable approach.
Man that bubble burst is gonna be huge. They done fucked up big time. Greed and impatience, FOMO.
