What this analogy is saying is that if an image is generic and derivative enough (or massively overrepresented in the training data) it may be possible to recon

What this analogy is saying is that if an image is generic and derivative enough (or massively overrepresented in the training data) it may be possible to reconstruct a very close approximation from the model. If the training data is unbiased, I question the validity of copyright claims on an image that is sufficiently derivative that it can be reproduced in this manner.