Both ANNs and brains are (1) highly distributed and parallel (2) make small changes (3) efficient. Ideas like generalization hold for all systems with these three properties. In that sense, @tyrell_turing is obviously right. Maybe @GaryMarcus uses narrow def of abstraction.https://twitter.com/GaryMarcus/status/1200088188934451200 …
-
It's not vague at all, lemme make it concrete for you: 1) Neurons do something very close to linear integration followed by a non-linearity. 2) They do this in a parallel manner, leading to distributed representations. 3) ANNs capture this in a simple model. Period.
“knowing” (1) has not carried us very far in neuroscience, and I think it is entirely possible that (1) is in some way misguided; we don’t even know with certainty that the individual neuron is the best level of analysis.
