They calculated that computers will be able to simulate the human brain as far as raw computing power is concerned by about 2045. About 150 years from now - since a single pc will be as 'smart' as a human - they have the computing power of the entire human race. Now, two problems with that. First off, as you mentioned, a computer doesn't work like a human brain. All a computer does is add ones and zeroes. It does it super-duper-fast though. Why is the human brain so fast? Pattern recognition, mostly. And our brain works with relations, instead of absolute numbers. That is why in total darkness you cannot say how large an object is (if that object were illuminated by itself). You know an object's size by relating it to other objects in the vicinity (spelling?). So, in order to simulate human like intelligence you have to 'hack' that intelligence with raw processing power. It takes a huge amount. But thankfully we will probably be able to see that in our lifetime. But that does not mean we can make AI that is equally smart. Processing power does not equal intelligence! You'd still have to program the computer to do smart things, which is essentially what the 'intelligence' part in AI means. And you could have a million supercomputers at your disposal, if you can't program them to be intelligent you won't get far. I'm sure x4000 knows most of this already.
The second thing is: by the time we have computers that can sort of match human smartness, we'd still want photorealistic graphics, and effectively we still don't have the processing power for AI. Because we wanna use it all for better looking games. AI is sort of a lost child in computer games, as audio is too. Graphics and gameplay probably take about 80% of dev resources, because it's what all the gamers want. So, in effect, even when we have those jawdroppingly fast computers in 50 years, we still have relatively dumb AI as we don't want to spend cpu cycles on it.
All of this is assuming that Moore's Law holds steady, and right now it doesn't look like that will happen -- we are instead shifting outward to the side, adding multiple cores and such. This is useful to a point, and the human brain is certainly massively parallel, but right now we don't really have good ways of writing coherent software logic that spans multiple cores. Most software (like mine) that can be split across multiple core is basically only split by feature -- AI on one, music on another, graphics on the graphics CPU, etc. Right now there isn't a game around that splits its core simulation across multiple cores (not that I'm aware of, anyway). This problem is SO hard, in fact that, IMB has a $1 million USD prize for the programmer(s) who come up with a way to compile single-core style code so that it can be run on multiple cores. That's kind of one of the holy grails of current computing, and I'm not sure it's even possible (but that's totally outside my area).
But yes, even with processing power at our disposal the software issue is still very much an issue. I think we are just hitting a point where the software is getting more development attention because we finally have the extra processing power to do something more advanced. Before it was always the hardware that was the limitation, but now that that limitation is easing, the software has new options that are just starting to be explored. And, people can complain all they want about the AI in any given modern game, but the fact remains that the AI is much better than what was in games 10 or especially 20 years ago. The field is advancing, but it's doing so slowly over time.
One big challenge is that a lot of the researchers that are looking into AI are focused more on true sentience, simulating actual brains, etc -- and that's the sort of thing that requires supercomputers. Game programmers are never trying to actually create true intelligence, they are only trying to mimic it, and so that comes with a whole set of challenges all its own. AI is improving overall, slowly but surely. Hopefully we'll see a marked jump upwards in quality in the next few years (now that the hardware issue is mitigated for a short while), but we'll just see.