5 ms·

this is why people should better study neuroscience, psychology if they want to advance research in AI. also things related to graph topology in neural network

by jokoon 2y ago

this is why people should better study neuroscience, psychology if they want to advance research in AI. also things related to graph topology in neural networks maybe, but probably not related to artificial NN. I was given this video, which I found was pretty interesting: https://www.youtube.com/watch?v=nkdZRBFtqSs https://www.youtube.com/watch?v=nkdZRBFtqSs (How Developers might stop worrying about AI taking software jobs and Learn to Profit from LLMs - YouTube)

renewiltord 2y ago
There are loads of psychologists and neuroscientists today. Has any of them in the last few years produced anything advancing AI? The proof of the pudding is in the eating so if they have at a higher rate than just straight CS/Mathematics and related then there’s probably some truth to it.
chmod775 2y ago
I can't seem to figure out the connection between this comment and the article at hand, except that they're both about AI.
dartos 2y ago
I don’t think psychology will have any bearing on AI. I doubt neuroscience will either, but I’m not as sure on that. The more impressive AI systems we have moved further away from the neuron analogy that came from perceptions. The whole “intelligence” and “neural” part of AI is a red herring imo. Really poor ambiguous word choice for a specific, technical idea.
- sva_ 2y ago
  > I doubt neuroscience will either, but I’m not as sure on that The stuff on spiking networks and neuromorphic computing is definitely interesting and inspired by neuroscience, but it currently seems mostly like vaporware
  - dartos 2y ago
    Yep, I’ve heard about spiking networks, but haven’t read into them much yet.
- fastball 2y ago
  *perceptrons
  - dartos 2y ago
    Darn autocorrect. Thank you.
    - actionfromafar 2y ago
      Haha, I didn't get it when I read "perceptions". Thought ... of what? :-D
- nradov 2y ago
  The question is whether current AI technologies represent any progress towards a true human equivalent artificial general intelligence. Most likely not, but no one knows for sure. If the answer turns out to be no then real progress will likely require theoretical insights from psychology, neuroscience, and other fields.
  - dartos 2y ago
    Fwiw, I don’t think we’re any closer to general intelligence then we were 5 years ago. Other than that, I agree, especially since you added “and other fields.” Psychology might eventually give us a useful definition of “intelligence,” so that’d be something. Obviously all research can influence other areas of research.
- Symmetry 2y ago
  It's easy to overstate, but shouldn't be understated either with, as an example, solving problems with learning in AI providing insights into how dopamine works in brains. https://www.technologyreview.com/2020/01/15/130868/deepmind-ai-reiforcement-learning-reveals-dopamine-neurons-in-brain/ https://www.technologyreview.com/2020/01/15/130868/deepmind-... There are obvious, huge differences between what goes on in a computer and what happens in a a brain. Neurons can't do back propagation is a glaring one. But they do do something that ends up being analogous to back propagation and you can't tell a priori whether some property of AI or neuroscience might be applicable to the other or not. The best way to learn about AI isn't to learn neuroscience. it's to learn AI. But if I were an AI lab I'd still hire someone to read neuroscience papers and check to see whether they might have something useful in them.

by jokoon 2y ago

renewiltord 2y ago

There are loads of psychologists and neuroscientists today. Has any of them in the last few years produced anything advancing AI? The proof of the pudding is in the eating so if they have at a higher rate than just straight CS/Mathematics and related then there’s probably some truth to it.

chmod775 2y ago

I can't seem to figure out the connection between this comment and the article at hand, except that they're both about AI.

dartos 2y ago

I don’t think psychology will have any bearing on AI. I doubt neuroscience will either, but I’m not as sure on that. The more impressive AI systems we have moved further away from the neuron analogy that came from perceptions. The whole “intelligence” and “neural” part of AI is a red herring imo. Really poor ambiguous word choice for a specific, technical idea.

sva_ 2y ago
> I doubt neuroscience will either, but I’m not as sure on that The stuff on spiking networks and neuromorphic computing is definitely interesting and inspired by neuroscience, but it currently seems mostly like vaporware
- dartos 2y ago
  Yep, I’ve heard about spiking networks, but haven’t read into them much yet.
fastball 2y ago
*perceptrons
- dartos 2y ago
  Darn autocorrect. Thank you.
  - actionfromafar 2y ago
    Haha, I didn't get it when I read "perceptions". Thought ... of what? :-D
nradov 2y ago
The question is whether current AI technologies represent any progress towards a true human equivalent artificial general intelligence. Most likely not, but no one knows for sure. If the answer turns out to be no then real progress will likely require theoretical insights from psychology, neuroscience, and other fields.
- dartos 2y ago
  Fwiw, I don’t think we’re any closer to general intelligence then we were 5 years ago. Other than that, I agree, especially since you added “and other fields.” Psychology might eventually give us a useful definition of “intelligence,” so that’d be something. Obviously all research can influence other areas of research.
Symmetry 2y ago
It's easy to overstate, but shouldn't be understated either with, as an example, solving problems with learning in AI providing insights into how dopamine works in brains. https://www.technologyreview.com/2020/01/15/130868/deepmind-ai-reiforcement-learning-reveals-dopamine-neurons-in-brain/ https://www.technologyreview.com/2020/01/15/130868/deepmind-... There are obvious, huge differences between what goes on in a computer and what happens in a a brain. Neurons can't do back propagation is a glaring one. But they do do something that ends up being analogous to back propagation and you can't tell a priori whether some property of AI or neuroscience might be applicable to the other or not. The best way to learn about AI isn't to learn neuroscience. it's to learn AI. But if I were an AI lab I'd still hire someone to read neuroscience papers and check to see whether they might have something useful in them.