Yeah, that's the stuff that more research needs to answer.
Of course, the neocortex has been receiving signals and "learning" for a long time before birth, so it's not exactly "blank" when the horse is born.
It's possible. Research is needed to answer these questions.
Much of the bulk configuration of the neocortex is clearly predefined in the genome. But how much of the interconnections that develop happen after the eyes start sending data to the cortex? I don't know when the eyes form and first start to send signals, but I know it's long before birth. If we could just cut up, dissect, and stick probes in a few thousand human babies we could answer more of these questions. :)
Without knowing the answer, we still have the issue of what we can do as programmers and robot builders to push the technology forward while we wait for the neuroscience to figure out more about real brains.
I believe there is still much to be learned about generic signal processing and learning systems. The type of system you can feed any sensory signal to, and it will figure out on it's own, what to do with the sensory information. I happen to believe the neocortex and most the supporting structure is just such a module in the brain, but whether I'm write or wrong about that is not all that important. What's important is whether there are better generic learning algorithms still possible to be developed. I think they are (and I think it's the key to creating human like behavior in robots), so that's what I'm exploring.
If your think the cortex is instead a lot of custom designed circuits, each for dealing with it's own special type of sensory data, then you can go about trying to duplicate such algorithms in code for processing data such as video. It's wise for people to be working on both approaches to see what we can find.