3 ms·

Hinton also said we should stop training radiologists 10 years ago because it would be solved by AI.

cubefox 4mo ago
That's not accurate, see https://www.lesswrong.com/posts/xE2HzcHWPFS9EsJqN/radiology-automation-does-not-generalize-to-other-jobs https://www.lesswrong.com/posts/xE2HzcHWPFS9EsJqN/radiology-...
- leoc 4mo ago
  He seems pretty unequivocal in the video: https://youtu.be/2HMPRXstSvQ https://youtu.be/2HMPRXstSvQ . He even points out that it’s something which he’s claimed previously, ruling out the idea that it was some kind of spur-of-the-moment conclusion.
  - cubefox 4mo ago
    Perhaps, but as the other article points out, the human requirement in radiology has a lot to do with legal and insurance requirements. This is similar to how a lawyer may not be better than an AI but the job is still exclusive to humans.
- root_axis 4mo ago
  Yes, Hinton's prediction was inaccurate.
ceejayoz 4mo ago
I was at a major radiology conference recently and the radiologists kinda seemed to agree. (The interventional radiologists are a little less nervous, for now.) Apparently 80% of FDA-approved AI "medical devices" (which covers software, not just hardware) are for radiology. (https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-enabled-medical-devices https://www.fda.gov/medical-devices/software-medical-device-...)
- root_axis 4mo ago
  This just underscores the point. Despite the advancement of "AI medical devices", the labor market numbers for radiology show Hinton was spectacularly incorrect. If we had followed his advice we would be experiencing a major crisis in medicine today.
  - ceejayoz 4mo ago
    But is it a "he was wrong" scenario, or is it a "he wasn't quite right on the timing" bit, ala https://en.wikipedia.org/wiki/Michael_Burry https://en.wikipedia.org/wiki/Michael_Burry ?
    - root_axis 4mo ago
      He is wrong by every conceivable measure. Not only was his timing wrong, where he specifically noted 10 years as the absolute maximum worst case scenario, but the trend has been accelerating in completely the opposite direction. To his credit, he actually made a falsifiable prediction, unlike most folks that rely on infinite timelines so they never have to be accountable for their prognostications.
tgv 4mo ago
The only way to never make wrong predictions is to not make any.
- root_axis 4mo ago
  And yet, here we are in a thread full of them.

That's not accurate, see https://www.lesswrong.com/posts/xE2HzcHWPFS9EsJqN/radiology-automation-does-not-generalize-to-other-jobs https://www.lesswrong.com/posts/xE2HzcHWPFS9EsJqN/radiology-...

leoc 4mo ago
He seems pretty unequivocal in the video: https://youtu.be/2HMPRXstSvQ https://youtu.be/2HMPRXstSvQ . He even points out that it’s something which he’s claimed previously, ruling out the idea that it was some kind of spur-of-the-moment conclusion.
- cubefox 4mo ago
  Perhaps, but as the other article points out, the human requirement in radiology has a lot to do with legal and insurance requirements. This is similar to how a lawyer may not be better than an AI but the job is still exclusive to humans.
root_axis 4mo ago
Yes, Hinton's prediction was inaccurate.

ceejayoz 4mo ago

I was at a major radiology conference recently and the radiologists kinda seemed to agree. (The interventional radiologists are a little less nervous, for now.) Apparently 80% of FDA-approved AI "medical devices" (which covers software, not just hardware) are for radiology. (https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-enabled-medical-devices https://www.fda.gov/medical-devices/software-medical-device-...)

root_axis 4mo ago
This just underscores the point. Despite the advancement of "AI medical devices", the labor market numbers for radiology show Hinton was spectacularly incorrect. If we had followed his advice we would be experiencing a major crisis in medicine today.
- ceejayoz 4mo ago
  But is it a "he was wrong" scenario, or is it a "he wasn't quite right on the timing" bit, ala https://en.wikipedia.org/wiki/Michael_Burry https://en.wikipedia.org/wiki/Michael_Burry ?
  - root_axis 4mo ago
    He is wrong by every conceivable measure. Not only was his timing wrong, where he specifically noted 10 years as the absolute maximum worst case scenario, but the trend has been accelerating in completely the opposite direction. To his credit, he actually made a falsifiable prediction, unlike most folks that rely on infinite timelines so they never have to be accountable for their prognostications.

tgv 4mo ago

The only way to never make wrong predictions is to not make any.

root_axis 4mo ago
And yet, here we are in a thread full of them.