I buy it to an extent. But what gives me pause: When we compare the early stages of the visual system of animals we see striking similarities to neural networks we train on visual data. The detectors in the artificial neural network converge to detect the same kinds of features found in the biological network. In other words the structure that we find in the networks comes from the data.

We can also see this when brain areas are recruited for tasks they were not specifically designed for. For instance visual processing areas can be recruiited for other tasks in the blind.

So while there is specialisation in brain tissue, rather than there being circuits designed at a low level for certain cognitive tasks, it is more that the connectivity and kinds of neurons within an area are tuned to excel at a specific class of problems.

It is softer and more diffuse organ than a "grammar gland". Rather it is a kind of soil within the mind that permits the growth of certain kinds of cognitive structures. Rather than being computer like and existing prior to the input of data, it is merely a capability expressed by the tuning of an overall system.

Chomsky's universals in this view become rather less important. It is merely the statement that brain tissue is tuned in such a way that it encourages, among other things, certain specific kinds of linguistic structure.

It's similar to noticing that the human hand is good stirring soup and then describing it as a soup stirring organ. Yes the hand has properties that allow it to be easily used as a soup stirring organ, but those same properties make it useful for some other things as well.