Recall the idea: mental computation involves DNA/RNA manipulations. The idea is that these complex molecules have the structure to support classical computations, as opposed to neural nets, and are ideal candidates for memory storage, again as opposed to nets. The Gallsitel-King conjecture is that brain computations exploit this structure (e.g.
here). One place quite ripe for this kind of process is long term memory. There is progressively more evidence that the genome plays an important role in this area (see
here). Here are two more reports (
here and
here) on recent memory research highlighting the role of genes in fixing and extinguishing memories.
Again, I can't stress this enough, neural networks are capable of universal computation just like a Turing machine is (ignoring memory limitations). If the main argument is that neural nets just can't do what DNA/RNA manipulation would allow, then this is just not true because they can.
ReplyDeleteThat being said, there is real evidence of DNA/RNA being relevant to neural computations and memory storage.
I don't believe the claim G&K make is that NNs cannot do what classical architectures can, but they do so in a very resource profligate manner. And they do this in a resource prfligate manner because they do not distinguish memory from CPU and don't have ways of implementing classical notions of memory which are required for coding notions like variables, binding to a variable etc. So, if I gave the impression that their argument was that NNs cannot compute what TMs do, then this was wrong. The argument is more interesting: a kind of poverty of resources argument: once you see the scope of the required computations, NNs will use up all the brain for the simplest ones.
Delete