Sorry, but I completely disagree with your general take. An AGI is a generalized agent which is able to perform extreme information compression on any available set of inputs. Think Force = m*a and coming up with the laws of the universe. You can store each possible action in memory of all available results OR -- a generalized agent will find the most succinct and least complex form of modelling the state of a system. An AGI to me in other words is a form of intelligence which is able to extract information and compress it such a way to not allow the lossy compression or storage to hinder its ability to predict real outputs from a set of training data fed into it. By the way, I have some ideas on how to create one and I'll be sharing some of the key algorithms and data structures that I'll be using here for anyone interested:

https://github.com/photonlines/AGI-Algorithms-and-Prototypes...