Well, if you're looking for understanding that's good enough we can program it, then yeah, we don't understand it. Also I think there was a misunderstanding on my part, in that we know quite a bit about human intelligence (that is, g), but not so much about intelligence-in-general. I'd rather clear up any miscommunication before throwing citations at you -- is your disagreement that g is not a well-established thing which maps very well to any tests of general intelligence you can conceive, or that g is a useless predictor, or that intelligence as measured by g is non-hereditary, or that g isn't useful because it's not a specific module of the brain we can simulate in code or hot-swap with that part of a donor's brain?

People may still quibble, but I was under the impression the standard acceptable definition for intelligence (at least among AI folks) is Legg and Hutter's informal "the ability to efficiently achieve goals in a wide range of domains". (Which has been formalized with AIXI.) I'm also aware of several groups in the AI field who believe they can actually construct a mind, in fact that this is the safest way forward, and that hoping it just "emerges" is folly as old as the first neural net proponents...