Neural networks may prove to be the better way to emulate real intelligence (up to this moment). However, just as natural intelligence, neural networks should evolve.... but how?

I'm not a specialist on NN, but from the experiments and research I've done, I think some work should be done in component-based NNs. I think the NN community focused too much on a sub-set of topologies that, while having interesting properties, do not represent the full potential of NNs. But, as I said, I'm really not up to date with all the research.

The answer of this cuestion is in the nature.

Exactly, it's in nature. Our purpous is to emulate it so we can understand it (not use it).

fortunately, we have some companies which are moving in this direction, by designing intelligent 'baby' AI's which will be trained just like humans so its mind can develop and 'evolve' like ours. We actually need some help with volunteers right now, so if you think you're well versed in C++(or Crystal Space) you can PM me for details