This is an excellent question whose answer is only obvious at first glance, asked by Scott Alexander on Astral Codex Ten. The correlation between the number of neurons and intelligence holds for biological brains and for AI, if we take the number of parameters to be equivalent to the number of neurons.
Part of the answer is that minds with more neurons can encode more patterns. Minds with fewer neurons need to compress the patterns more, and that compression loss means that it’s harder to accurately do pattern matching.
As Alexander points out, that may not be the full answer. Why does practice improve performance at pattern matching, such as is needed to solve Raven’s progressive matrices, only a little, and never to the same level than a genius who didn’t practice at all?
One possibility is that you have some kind of “pattern matching region” taking up some very specific percent of your brain, and the bigger the brain, the bigger the “pattern matching region”. […] But this doesn’t really seem right – endless practice on thousands of Raven’s style patterns helps a little, but a true genius will still beat you. But by the end of your practice, you will have far more patterns stored than the genius does. So what do they have that you don’t?
Unfortunately, I don’t think the attempt at explanation that follows in Alexander’s blog is fully satisfying either.
Nehaveigur 