A paper describing AlphaZero, Google’s AI chess engine, has finally appeared. I thought this was fascinating, until it got to the math. I wonder if the program’s “random walk” through the game space to achieve “superhuman performance” has any applications to evolutionary concepts?
The game of chess is the longest-studied domain in the history of artificial intelligence. The strongest programs are based on a combination of sophisticated search techniques, domain-specific adaptations, and handcrafted evaluation functions that have been refined by human experts over several decades. By contrast, the AlphaGo Zero program recently achieved superhuman performance in the game of Go by reinforcement learning from self-play. In this paper, we generalize this approach into a single AlphaZero algorithm that can achieve superhuman performance in many challenging games. Starting from random play and given no domain knowledge except the game rules, AlphaZero convincingly defeated a world champion program in the games of chess and shogi (Japanese chess), as well as Go.
The rules for chess are pretty specific. I guess maybe what you were thinking is that the other player’s moves define a changing landscape and the program tries to optimize its moves to achieve the goal, winning.
One thing is that evolution does not have a definite way to define winning (an ultimate end of the game), other than total extinction. You are only and always a provisional winner until you are not.
P.S. I am a go player. I have a copy of lela zero & lizzie (a version for home computers). And I watch the Michael Redmond analyses of self-play games, alphago master against zero.
Michael Redmond is an long time American professional go player. He moved to Japan to become a part of their professional go program and is now 9p (highest rank) and plays in many of the professional title tournaments.