How do you build an artificial general intelligence? No one really knows. Or more precisely, no one has proven that they know. The idea that a narrow AI, what we currently have, will somehow lead to an AGI is a guess, and to my mind, not a very compelling one.
Building a clock doesn’t get you any closer to time travel any more than a neural net gets you closer to AGI. I believe we cannot program an AGI. This seems obvious to me for the simple reason that we don’t understand general intelligence. However, we may be able to build a system that can evolve it, because we may understand those basic principles well enough.