Re: They learn by making mistakes and correcting their errors.
I would be more inclined to say:-
They FUNCTION by making mistakes and WEIGHTING their errors.
Many of us here will have written programs that "learn" how to win at Tic Tac Toe. Trivial perhaps, but anchors one's imagination as to how AI "works".
If Deep Mind had been forced to play chess where White starts by moving the Kings rook pawn, then black does the same, then every combination from there, then it would never have got to the nitty gritty of beating Kasparov, there would be just too many meaningless moves in its database. So it was no doubt primed with thousands of classic openings which set it on the correct learning "tree structure".
If Kasparov had gamed the system by starting with a non-standard opening, that would have forced Deep Mind to throw away its entire Opening Move catalogue and force it to play from scratch, which would have put it at a big disadvantage.