Sure, it's called hard work and dedication. You could read "From NAND to Tetris" for starters. After that you should have the basics down.

Java is actually pretty fast if you don't do stupid things. It's just that Java programmers tend to produce pretty complex OO code with lots of small objects that kill performance. But still, compared to other "modern" languages like Python or Ruby, Java is very fast.

I believe your argument is wrong. First, the halting problem and similar undecidable problems are irrelevant to the AI discussion. By general intelligence we usually mean intelligence that is like that of humans, i.e. general enough to solve a variety of problems. Humans can't solve the halting problem, so it doesn't matter if computers can't either.

As for your second argument. First, in many cases, a speedup alone is good enough to make a qualitative difference. Take for example programs that play chess. Clearly I could take to source code to the latest Fritz and its databases and play just as well as the program. However, I wouldn't be able to participate in any competition, because every move would take me years. Second, arguments using complexity classes like P are not really appropriate here, as P deals with decision problems. In AI we're more interested in heuristic solutions to optimization problems. Here it is rather simple to give algorithms that can produce smarter versions of themselves. Evolutionary algorithms do this all the time.