Ok, maybe this is why UG always sounds so wrongheaded to me when explained in layman's terms:
Our knowledge of English is not purely derived from the sentences we've heard.
If you were to take the sentences in isolation, devoid of meaning and context, I suppose you
might end up being circular. But new sentences are being thrown at you all the time (the training data is ~unlimited). Even a computer can recognize common patterns unsupervised, then apply those patterns to learning other patterns on subsequent data, and finally classify examples has never seen before -- this seems weak.
And doubly so if you consider that it isn't just a sentence we've heard, but a sentence about a real noun and a real verb that exists in space (at least initially). If I pick up a ball and say "ball" consistently, bam, you now have some kind of tagged input to work with.
We must have had some model of a language in common
when we started, and we fit the parameters to what we
heard as children.
Even taken the first sentence as given, why does that imply a UG? Perhaps the model is built from observed reality first, then language is built using that model? Or, maybe, to argue Chomsky's side, the "UG" is an extremely simple bias...
Now, Chomsky has probably addressed these things, but to a layman, the typical justifications seem a little weak.