"
computers today don't create knowledge"
Whenever a computer program discovers a new prime number, or any of a zillion other things, a computer has created some knowledge. If you want to say that this is really a human operating through indirection, we need to again follow the chain all the way back through evolution and say that it was the fundamental physical laws which "created" the knowledge.
If I am misunderstanding your definition of "create knowledge", please clarify.
"human brains do evolution. they do replication and selection of ideas."
I'm not sure if I'm understanding you correctly here, but I'll respond to it as I interpreted it.
Let's be more precise, since "evolution" can pretty much apply to anything. e.g. The Grand Canyon "evolved" from a big chunk of rock. I assume that you're referring to the process of natural selection when you say evolution.
It's true that memes replicate and undergo a process analogous to natural selection in biology, memes float around the memepool, and their success at "infecting" minds (getting that mind's owner to agree with the meme) determines their fitness, which in turn allows it to reproduce, etc. But this is entirely separate from the process used within the mind/brain to determine whether or not to agree with the meme. Through some process brains arrive at decisions and opinions, and those decisions are what put selective pressure on the memes. But this says nothing about what's going on inside the brain to arrive at the decision, so we can't say that the brain operates by evolution. All we can say is that once the brain arrived at a decision, that decision added a bit of selective pressure on the memepool.
Saying that decisions themselves are produced by an evolutionary process is analogous to saying that death operates by evolution. Death is a biochemical process, and there's no natural selection going on inside the organism to cause it to die. We don't yet know what's going on inside a brain to arrive at decisions.
Anyway, having said that, there is a theory that brain development operates by something called "Neural Darwinism", but I don't know how valid it's considered to be. http://en.wikipedia.org/wiki/Neural_Darwinism
"Imperfect replication + selection -> complexity/knowledge"
Yes, but the converse is not true; complexity/knowledge do not imply imperfect replication and selection. Natural selection doesn't have a monopoly on all complexity. It's entirely conceivable that the brain uses some "algorithm" other than something analogous to natural selection.