Human levels of sample efficiency. Can you elaborate?

Human can read all the rules IRL, and after 2 matches already have meaningful gameplay. Does the AGI has to be able to read a game wiki and be able after couple matches catch the concepts of the game?

Currently computers can learn a game faster in terms of time, but they complete millions of matches during learning, being 6 orders of magnitude behind human in terms of matches.