I have to disagree with you -- I think we could do it now, with electronic switches. We wouldn't get genius level AI, but we could most definitely get something that learns from its environment and makes real decisions without programming it in in a decision tree (which I suspect is your issue -- the chinese room problem).
Don't forget, our brains are basically -- a complex array of biological switches. The trick is to get the right mix of structural organization and functional systems and that complex array of switches. Our brains aren't just neural networks, they are highly structured neural networks with dedicated function visual cortex, auditory cortex, etc. plus a wide range of modulators and probably some structures we haven't identified yet or don't understand yet. But at the end of the day, unless you are indulging the mind projection fallacy, we are most likely just wet electrochemical machines with emergent intelligence (and a fair bit of dysfunction).
I don't hope or fear. I actually think it would be very, very cool to have real AI, and, like everything that humans do, there are probably good things and bad things that will come of it when we do.
Real AI "could" give us the stars. We could conceivably build a large, smart ship that is capable of repairing itself and handling complex challenges and that had the facilities to create an ecology (if necessary starting with amino acids and a library) at the other end. The ship would then last the centuries needed to bridge the distances until it found a suitable planet and could then do anything from start life there altogether to insert human life and human-supporting ecology. Not an original SciFi topic, but one that is quite plausible and that is arguably more plausible than the cryogenic freezing or planetoid-sized multigeneration manned ship alternatives.
Real AI could also wipe us out when our robot slaves revolt. Or anything in between.
But I'm guessing we will find out comparatively soon. Moore's Law shows no signs of wearing out, and if anything might soon experience another paradigm-shift jump, possibly to a new and faster scaling law altogether. Software is also increasingly mature.
If somebody wants to give me 10 million or so, I'll promise to do my best to make it happen in the next 10 years (and I think I can do it, and came within an ace of writing a proposal for a DARPA grant to do just that before they changed or clarified the intent of the program to exclude that as the goal). But it is definitely the kind of project that requires complete dedication and adequate resources. I think I've got a good idea of the metastructure required, but there is still implementation of that structure in code and debugging (and the hardware).