Unless you can demonstrate a significant functional difference between natural and artificial self-organising information processing systems, given the latter are modelled on the former, how can you confidently conclude that artifical systems are incapable of these things?
They likey are not be capable at this stage of their development, no more than a worm's cortex is at creating literature. Yet, the same basic neural cells grouped into vastly larger clusters ulitmately are.
Furthermore, there's no reason to assume that future neural net machines will be entirely semiconductor based. Scientists have already demonstrated neural cells from some snail or worm (can't recall which) being used to perform binary addition.