It's easy to overstate, but shouldn't be understated either with, as an example, solving problems with learning in AI providing insights into how dopamine works in brains.
There are obvious, huge differences between what goes on in a computer and what happens in a a brain. Neurons can't do back propagation is a glaring one. But they do do something that ends up being analogous to back propagation and you can't tell a priori whether some property of AI or neuroscience might be applicable to the other or not.
The best way to learn about AI isn't to learn neuroscience. it's to learn AI. But if I were an AI lab I'd still hire someone to read neuroscience papers and check to see whether they might have something useful in them.
https://www.technologyreview.com/2020/01/15/130868/deepmind-...
There are obvious, huge differences between what goes on in a computer and what happens in a a brain. Neurons can't do back propagation is a glaring one. But they do do something that ends up being analogous to back propagation and you can't tell a priori whether some property of AI or neuroscience might be applicable to the other or not.
The best way to learn about AI isn't to learn neuroscience. it's to learn AI. But if I were an AI lab I'd still hire someone to read neuroscience papers and check to see whether they might have something useful in them.