My issue is this: for most questions I am interested in, ANNs simply aren't "biologically plausible" enough.

The claim that ANNs are in a sweet spot carries an implicit assumption about the set of questions that are interesting. Which is of course subjective!

@cian

They start with a list of desiderata for modeling aims (see page 433). You may not agree with the “Goldilocks” claim if you don’t agree with the desiderata.

@cian

I’d love to hear what you’d add to or remove from the list.

@ShahabBakht tricky q! Maybe:

- offer insight into why biological brains contain the mechanisms and organisation that they have (why so many genetically separate types, what does the cerrebellum do, etc)

- what forces drive the differences in cognitive processes between individuals within a species or differences between species

- what is the mapping from genotype to cognitive phenotype

- which computations are done via molecular signalling vs electrical circuit dynamics

@cian Biology and biological mechanisms seem to be crucial elements of the Qs on your list. I’m not certain ANNs can be helpful when it comes to this level of biological detail.
This is also related to one of the GACs on Friday: https://gac.ccneuro.org/gacs-by-year/2023-gacs/2023-1

The Qs on your list, to me, seem to call for a more Sherringtonian approach, while the common applications of ANNs in cog neuro today seem to mostly take a Hopfieldian view. Though, a lot of discussions and questions on Friday were around whether ANNs can reconcile these two views.