3 ms·

> > It is mystical to hope for awareness, understanding, common sense, from symbols and data processing using parametric functions in higher dimensions. > but

> > It is mystical to hope for awareness, understanding, common sense, from symbols and data processing using parametric functions in higher dimensions. > but that is all our brains do, really. > How can we claim that a computer doing the same kinds of computations that we do is fundamentally different? This just reeks of human exceptionalism. Our brains and physiological mechanisms are far more complex than matrix multiplications sprinkled with some tan or max(0,x). I'm ok with saying that we are big machines etc, but the power of our primitive operations, the range in aspect and behaviors of our mechanisms is far far greater than what is currently done in NN. The complexity of what is learned in NN, seen as a computer program, seems to me to be at the level of a full-text-search index together with a probabilistic automaton. What i believe we are as humans (and other "very sentient" animals) is more on the level of linux: tons of subsystems each looking very different, tons of "drivers" for accomplishing very specific tasks rapidely.

sudosysgen 4y ago
Matrix multiplication with some non-linear functions, memory, and persistent internal state is Turing Complete, so there is no reason it couldn't eventually be used to implement an AI, as long as such a thing is at all possible.
- lapinot 4y ago
  Universal models of computations are still distinguishable by their complexity classes. Every universal model can simulate the others but these simulations may not be practical. The whole insane teraflop showoff of the big players of NN is fun and all (saddening actually) but at some point everyone has a limit. Then you have to think about how to get qualitative improvement, and since ML basically boils down to optimizing in computable function spaces just take a better computation model. If you're keen on gradient descent take some language of computable derivable functions (there are richer ones than finite space linear functions + some primitives). Cf Conal Elliott talking about how getting stuck on the idea of matrix operation graph is completely stupid (among other things): "A Functional Reboot for Deep Learning".

sudosysgen 4y ago

Matrix multiplication with some non-linear functions, memory, and persistent internal state is Turing Complete, so there is no reason it couldn't eventually be used to implement an AI, as long as such a thing is at all possible.

lapinot 4y ago
Universal models of computations are still distinguishable by their complexity classes. Every universal model can simulate the others but these simulations may not be practical. The whole insane teraflop showoff of the big players of NN is fun and all (saddening actually) but at some point everyone has a limit. Then you have to think about how to get qualitative improvement, and since ML basically boils down to optimizing in computable function spaces just take a better computation model. If you're keen on gradient descent take some language of computable derivable functions (there are richer ones than finite space linear functions + some primitives). Cf Conal Elliott talking about how getting stuck on the idea of matrix operation graph is completely stupid (among other things): "A Functional Reboot for Deep Learning".