The Future of Artificial Intelligence: Can machines become sentient (self-aware)

Those things are calculations; we just don't access them at that level. By analogy, try calculating logarithms by playing Quake.

You realise that this, like neural nets, can achieve exactly nothing that conventional hardware can't, right?

No, it doesn't but that doesn't let you achieve a single thing that not doing it doesn't. Programming is very powerful. If nothing else you can exactly simulate your rewiring of the processor on a normal chip.

The CPU itself can't evolve, but you can write software on it that will achieve everything that evolving the CPU can. Everything.

This strikes me as little more than mysticism. Why should we believe that the particular structures of the brain are required for sentience? It seems to me the only reason for believing so is the idea that some "special woo" happens in the brain that magics up sentience - and that goes against everything we know about the brain.Neurons and connections, btw, are definetely not required because we know, for a fact, that neural networks can achieve exactly nothing that conventional hardware can't.

Whole brain modelling may be required to unpick sentience while we study it, but I see little reason to believe there's anything going on that is more than it appears. Emergence from the simple components is probably the answer; the requirement for particular wetware with particular molecules and structures of the brain is a claim without evidence. Once you get as far as Penrose's silly insistence that "quantum tubules" are the seat of consciousness you're just rebranding mysticism in a pretty scientific dress. As far as we understand the behaviour of neurons they do not a single thing that cannot be done with a normal computer (as can be proved mathematically). Further, everything we know how works in the brain operates by the complex interactions of high level components not the low level vageries of how it is constructed.