4 ms·

> Nonetheless, Hinton and a few others immediately took up the challenge of working on biologically plausible variations of backpropagation. Trying to prove th

> Nonetheless, Hinton and a few others immediately took up the challenge of working on biologically plausible variations of backpropagation. Trying to prove the plausibility of a theory is one approach to science I guess... The researchers have already concluded that brains are simply information processing machines and that AI techniques are a sufficiently representative model to use to learn what brains are like. I don't see how this research could give us clues to anything other than what is already presumed to be true by the researchers.

mjburgess 6y ago
You're downvoted but this is correct. It is much like when the analogy was "springs and cogs", and an academic department created in that era "cog-nitive" science, would be the attempt to rotate enough gears in the right way. Many presumptions are being made here in "computational cognitive science" which preclude including many relevant features of animal learning and animal biology. Their whole world view is that "patterns of electrical signals in neurons" is where learning takes place. This is very likely to be false: it fails, for example, to note that the brain grows. Organic growth isn't even scoped here. A brain is a time-evolving dynamic system, whose architecture is at every level dynamic. (& Not least, embedded in a motor system which has a profound effect on its structure ).
- eli_gottlieb 6y ago
  >Many presumptions are being made here in "computational cognitive science" which preclude including many relevant features of animal learning and animal biology. This post doesn't actually seem to be citing computational cog-sci, which is usually a bit better about these things. Instead it's addressing the field of biologically plausible (ie: with Hebbian learning rules) deep learning. > (& Not least, embedded in a motor system which has a profound effect on its structure ). Sure, but that would expose how weak so much of the present AI work actually is when it comes to studying the motor system.
- visarga 6y ago
  > Organic growth isn't even scoped here. Other things current AI's are lacking besides growth: embodiment + the social and physical environment, ability to make interventions in the environment, self reproduction, learning from reward signals, autonomy, adaptation, radical open-endedness. "Patterns of electrical signals in neurons" are just part of the picture. Yes, learning happens there, but learning is fed by signals from the body and environment. It would be silly to focus on the neurons while ignoring the actual content, then start wondering where meaning comes from, and if syntax is enough. Meaning doesn't come from mere neurons, it comes from being an embodied agent.
- erikerikson 6y ago
  > Their whole world view is that "patterns of electrical signals in neurons" is where learning takes place Actually, the mechanisms are chemical processes involving trophic factors (i.e. inputs to those processes) and alteration of the physical structures the signals are transmitted with. You say "the brain grows" but the alteration of its structure to strengthen our weaken transmission and connections in response to signals is how it grows usefully. Which was present in the work described by the article.
  - mjburgess 6y ago
    An infant brain and an adult brain do not differ by their "strength of connection and trasmission". It is a hypothesis, and I think a false one, that this is identical to learning. The dynamical number and arrangement of neuronal tissues is not incidental. The micro (sub-neurone), medial (neurone) and macro structure (morphology) of the brain is time-varying, not merely its connection patterns (conditioned on fixed micro/medial/macro).

mjburgess 6y ago

You're downvoted but this is correct. It is much like when the analogy was "springs and cogs", and an academic department created in that era "cog-nitive" science, would be the attempt to rotate enough gears in the right way. Many presumptions are being made here in "computational cognitive science" which preclude including many relevant features of animal learning and animal biology. Their whole world view is that "patterns of electrical signals in neurons" is where learning takes place. This is very likely to be false: it fails, for example, to note that the brain grows. Organic growth isn't even scoped here. A brain is a time-evolving dynamic system, whose architecture is at every level dynamic. (& Not least, embedded in a motor system which has a profound effect on its structure ).

eli_gottlieb 6y ago
>Many presumptions are being made here in "computational cognitive science" which preclude including many relevant features of animal learning and animal biology. This post doesn't actually seem to be citing computational cog-sci, which is usually a bit better about these things. Instead it's addressing the field of biologically plausible (ie: with Hebbian learning rules) deep learning. > (& Not least, embedded in a motor system which has a profound effect on its structure ). Sure, but that would expose how weak so much of the present AI work actually is when it comes to studying the motor system.
visarga 6y ago
> Organic growth isn't even scoped here. Other things current AI's are lacking besides growth: embodiment + the social and physical environment, ability to make interventions in the environment, self reproduction, learning from reward signals, autonomy, adaptation, radical open-endedness. "Patterns of electrical signals in neurons" are just part of the picture. Yes, learning happens there, but learning is fed by signals from the body and environment. It would be silly to focus on the neurons while ignoring the actual content, then start wondering where meaning comes from, and if syntax is enough. Meaning doesn't come from mere neurons, it comes from being an embodied agent.
erikerikson 6y ago
> Their whole world view is that "patterns of electrical signals in neurons" is where learning takes place Actually, the mechanisms are chemical processes involving trophic factors (i.e. inputs to those processes) and alteration of the physical structures the signals are transmitted with. You say "the brain grows" but the alteration of its structure to strengthen our weaken transmission and connections in response to signals is how it grows usefully. Which was present in the work described by the article.
- mjburgess 6y ago
  An infant brain and an adult brain do not differ by their "strength of connection and trasmission". It is a hypothesis, and I think a false one, that this is identical to learning. The dynamical number and arrangement of neuronal tissues is not incidental. The micro (sub-neurone), medial (neurone) and macro structure (morphology) of the brain is time-varying, not merely its connection patterns (conditioned on fixed micro/medial/macro).