We can already generate big enough NNs that we can understand them in general (some areas get "specialised" in some way and more active in some conditions), and we know that the result is what we expect - but there's no way someone will take a look at the weights / resulting model and tell you what it does. I think we can create consciousness intentionally, but not understanding the process completely.
Selectively breeding cows for increased milk production allows evolution to find a way or ways of making that happen. The breeder doesn't need to worry about hormones or glands. Evolution doesn't need to worry about the breeder's motives, why are only the high milk producers producing offspring?
Sometimes, when I'm getting overlogical, I see machine intelligence as being pretty much our destiny (i.e. that singularity thing) and that organic life may be very much obsolete once it gets into its own.
As organic life was to the universe before it, machine life will be to us.
ADDED: but yes, extended phenotype and all that jazz.
The design of machines evolves all the time.
Of course, this only applies to intentional creation of consciousness. It says nothing about the distinct possibility that we could create a consciousness by accident.