[Comp-neuro] Hilbert's questions
billl at neurosim.downstate.edu
Tue Aug 12 22:25:54 CEST 2008
> I wouldn't expect to convince someone like you
I fear there are many who will be hard to convince
Your models, which I'm very impressed with and have enjoyed reading about, are nonetheless from
my perspective based on views of underlying neural processing that are untenable. I also often
use I&F neurons and believe that one can learn from them. However, they fall so far short of
neuronal function that they are like modeling mitochondria with internal combustion.
By extension, there is much that can be learned using comp cog sci models but they are to my
mind at best a very gross approximation and at worst subject to the massive
Kuhnian-Ptolemy->Copernicus-ish revision referenced by Jim.
> the models can account for a huge range of biological and behavioral
> data, and critically make predictions that are being tested and validated
A number of the issues that come up have been referenced in the preceding discussions but I
would just note that if recordings are made using assumptions such as eg rate coding and
CA3 homogeneity then models that are designed based on these assumptions may be confirmed
by additional experiments without either being trivial or being really "right." One could of
course argue endlessly, and likely acrimoniously, about this without making much progress.
I have for some years felt that Hubel and Wiesel were "wrong" and ended up doing neuroscience a
disservice with the force of their brilliance. I realize that this is not a popular viewpoint
since it moves us back to square -1.
More information about the Comp-neuro