Yes – this game has some fixed, relative small set of rules so the RL could learn to play by playing millions of games at random but following the rules of the game. Confront this with dounting (infinite) number of situations which may approach one in a daily life.

OK, RL exists end results like the protein design or Go are impressive, but does exist a RL solving the benchmark problem?

Yeah RL exist since 80s (or in some form earlier) but have it solve the benchmark?

if only it would exist

That behavior is not very fascinating.

Well, i found fascinating that mere computer is able to “learn” the life-like moves, no matter which year it is. Does modern algorithms can do it better?

yep, this is it

This is not analog computing by no means