Friday, July 27, 2018

Why Very Advanced Chess Programs are Calvinist

Sure. Alpha-Zero or the next gen Alpha Zero could beat the World Champion with an average player as partner sharing moves in certain conditions. As a player that finished 3rd in blitz chess at Chess.com - Play Chess Online - Free Games out of 193 players yesterday with a 1320 rating and a slow wireless connection, and with a 1550 rating at Lichess after learning the game since 2012, and graduating from a programming course in 1981 I have confidence that computers and computer programming can beat human players with the right programming. A Super GM at a simul can beat even fifty players sometimes- Alpha Zero should have no trouble partnering with a patzer. Maybe its like asking if a fat auto-bound office worker could win the Olympic Marathon with very fast high-tech ford electrical power legs if he walked without help every mile or two for a minute.

Alpha-Zero would need to study the partner’s chess games and learn how the player moves. It would need to know what openings the player knows, how far into the middle game and the sort of end game techniques the player has. If the 1400–1600 player has played 10,000–20,000 blitz games at Chess.com - Play Chess Online - Free Games and Lichess, and if the two partners had a way of communicating and sharing before the match so they would be on the same sheet of music regarding how to play (such as Bogo Indian or London System, Italian Game or Spanish, Berlin Defense or hack attack etc), I think they could together come up with a winning plan. Alpha-Zero could anticipate the chaos of inaccurate moves and blunders the partner uses foreknowing a litany of corrections or move vectors from the occasional or frequent inaccuracy.

The pair could practice the tandem chess against Stockfish 8 or 9 eventually and see how they do. They could play against PlayMagnus. Alpha Zero needed just four hours to learn how to play chess well enough to beat the best chess engine in the world although some seem to think that Stockfish might have thrown in the towel intentionally. Alpha Zero would know every game and move the World Champion has ever played and have several options paradigm to use for every move the champ made with options paradigm extending ahead for thousands of moves though games seldom reach 100. It would have the moves the patzer makes in subsets for every possible game the champion might play, and plan vectors and variable winning responses to any theoretical adversarial move.

Alpha-Zero learns too fast to have trouble playing and winning against a human-even the WC, even with half of its petaflops tied behind its back. A very advanced chess playing algorithm creates a deterministic board environment. It is Calvinist.

No comments: