An AGI is a system that can improve itself and is virtually unbound by a domain

up:: AI Alignment MOC Although Narrow AI powerfully sort patterns in limited domains like the AlphaZero Chess AI, their strength is only limited to one domain. DeepMind TK’s AlphaZero can’t operate a coffee machine and neither can IBM’s Watson steer a self-driving car.

Human Brains work in a lot of different domains, even adapt to new one’s. A general AI is able to do the same, not being bound by a specific domain or virtually not being bound by a specific domain if that domain is big enough. An AI whose domain is “The World” is pretty much a General Intelligence.