No—it’s just an observation from my experience (CS degree in the 90′s).
Just to be clear, he is making a clear conceptual mistake that indicates he does not understand universal computability:
… the reason for this is simulating the neural activity on a Von Neumann (or related computer) architecture does not reproduce the causal structure of neural interactions in wetware. Using a different computer architecture may avert this problem …
If there is some other weird computer architecture that can reproduce the causal structure of neural interactions in wetware, then a universal computer (such as a Von Neumann machine) can also reproduce the causal structure of neural interactions simply by simulating the weird computer. This really is theory of computation 101.
“He does not understand universal computability” seems an overstatement, universal computability doesn’t logically imply functionalism, although I agree that it tends to imply that definitions of consciousness which are not invariant under simulation have little epistemic usefulness.
There is a tiny chance that when he said “does not reproduce the causal structure of neural interactions”, what he actually meant was “would simulate the neural interactions extremely slowly”, but if that was the case, he really could have said it better.
My priors are that when people without formal computer science education talk about brains and computers, they usually believe that parallelism is the magicalpower that gives you much more than merely an increase in speed.
In practice it’s just a matter of computational power. His statement makes it fairly clear that he doesn’t understand this distinction.
Circuit level simulations of advanced microchips certainly exist—this is not just theory. Yes they are super expensive when run on standard CPUs (real-time simulation of an iphone CPU naively would require on the order of an exaflop). However, low level circuit binary logic ops are much simpler than the 32⁄64 bit ops that CPUs implement, and there are more advanced simulation algorithms. Companies such as Cadence provide general purpose binary logic emulators that actually work, in practice for reasonable cost, not just theory.
No—it’s just an observation from my experience (CS degree in the 90′s).
Just to be clear, he is making a clear conceptual mistake that indicates he does not understand universal computability:
If there is some other weird computer architecture that can reproduce the causal structure of neural interactions in wetware, then a universal computer (such as a Von Neumann machine) can also reproduce the causal structure of neural interactions simply by simulating the weird computer. This really is theory of computation 101.
“He does not understand universal computability” seems an overstatement, universal computability doesn’t logically imply functionalism, although I agree that it tends to imply that definitions of consciousness which are not invariant under simulation have little epistemic usefulness.
In theory there is no difference between theory and practice. In practice there is.
A physical Turing machine can simulate an iPhone, in theory. Would you like to try to build one? :-D
The only problems would be speed and memory.
There is a tiny chance that when he said “does not reproduce the causal structure of neural interactions”, what he actually meant was “would simulate the neural interactions extremely slowly”, but if that was the case, he really could have said it better.
My priors are that when people without formal computer science education talk about brains and computers, they usually believe that parallelism is the magical power that gives you much more than merely an increase in speed.
In practice it’s just a matter of computational power. His statement makes it fairly clear that he doesn’t understand this distinction.
Circuit level simulations of advanced microchips certainly exist—this is not just theory. Yes they are super expensive when run on standard CPUs (real-time simulation of an iphone CPU naively would require on the order of an exaflop). However, low level circuit binary logic ops are much simpler than the 32⁄64 bit ops that CPUs implement, and there are more advanced simulation algorithms. Companies such as Cadence provide general purpose binary logic emulators that actually work, in practice for reasonable cost, not just theory.