Gary Marcus

I am amazed at how many ppl resist this basic fact! Folks: you may believe ANNs miss critical biological details, cool, that's a legit position to take. But why pretend that ANNs are not an abstraction of neural processing? That is simply not a tenable position, frankly.

started a new thread because my jaw dropped at the hubris of this remark.

mere approximation doesn’t mean you have really captured what that component is, let alone how system as a whole works. NOBODY really understands how you get from neural nonlinearities to cognition without clear account of that, rest is straw-grasping

You're trying to have a debate with me that I'm not interested in, bc I am not taking the position you think I am. I am not claiming that I can say with certainty that ANNs are a *good* model of the brain. I am claiming that they capture *some* aspects of neural processing.

it may turn out that we are fundamentally wrong in our early 21st century thinking about the brain, and eg that most of the action is at the dendrites and that we have misunderstood what neurons (and other constituents of the brain) do. see comment re in vitro vs in vivo

But I agree that we may massively underestimate the computational power of dendrites. @IlennaJ and @aha_momentum in my lab are going crazy about that idea ;)

@GaryMarcus I think you're not understanding @tyrell_turing . I saw another post today that said "all neurology is inferior to ANN models," and THAT point of view is genuinely worth ridiculing.