astonished that anyone knowledgeable could claim that neural nets are (obviously) an “abstraction of neural processing” when we don’t yet know how brains work. if you don’t know how Y works you can’t really speak with certainty about whether X is an abstraction of Y. Period.https://twitter.com/tyrell_turing/status/1200072223299657728 …
-
Don't we know at least a little (if not everything) about how brains work? That there are many many neurons that communicate, etc. And is it not fair to say ANNs are a very simple abstraction from that current understanding of brains (there are many neurons, etc.)?
Maybe, but maybe not. We can't even model the worm, and have had wiring diagram since 1986. Maybe we are on right track, & maybe we aren't. Maybe it's ok to simplify everything dendrites do, and treat all cell types equally, and maybe we miss something fundamental by doing so.
-
-
I think that from a behavioral standpoint, we can infact model many aspects of the worm.
-
from a more detailed standpoint, I think there simply too many unknowns just now. As far as I know, we don't even have good neuron or synapse model of most neuron/synapses for the worm. a wiring diagram is not very useful without those, from a modeling perspective at least.
- 7 more replies
New conversation -
-
-
Not just worm. We can't even model a cell. The last time I looked, a neuron was also a cell.
-
(Sorry for the intrusion) Just an addendum: it is open to debate whether we need to model a whole cell (mitochondria, Golgi app., RER/SER, mRNA, DNA, enzymes, lysosomes/peroxisomes, various membranes...) in order to have a fully working model of a neuron. Actually, avoiding
- 3 more replies
New conversation -
-
-
The fact that we might be missing something fundamental is also highlighted by the ludicrously low power that the brain operates on, compared to what ANNs need.
-
Also the brain has this remarkable power to generalize, which it possibly does by applying topology to what it perceives. ANNs by their current implementations fail spectacularly at this.
- 2 more replies
New conversation -
-
-
But as far as I've seen in this convo the claim is not that ANNs are a super useful abstraction that will yield the same behavior as brains (what you seem to getting at in second part of tweet), just that they are AN abstraction from brains.
-
issue is whether they are a correct/useful abstraction from a proper understanding of something we don’t understand. i just don’t see how methodologically we could be CERTAIN about that in our current state of ignorance.
- 8 more replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.