The hard problem is "why do we have qualia?" and it is explained by neural nets. You input sensations and you get latent representations that are mapped into a space of possible representations. So all those ineffable perceptions are just values in this space of representations.

Thus, qualia appear by neural net processing of sensations, and are fed into a recurrent network that judges them moment to moment and updates its internal state, as well as perform actions. That is why we feel a stream of perceptions, not just separate moments.

Qualia appear to be irreducible because the mapping from perceptions to representations is nonlinear - even if we can duplicate it in neural nets, we still can't assign semantic value to each neuron and weight in the network. So it appears magical because it is of a level of complexity that exceeds the human working memory. But it is not so magical that we can't replicate it. We can recognize objects better than humans on certain datasets, and we have word embeddings built on huge corpora of text that are being used for translation. We can compute the "feel" of a word or image.