At the end of 2017, a revolution quietly arrived. AlphaZero, an artificial intelligence program developed by Google's DeepMind, defeated Stockfish, the world's most powerful chess program at the time. AlphaZero's 100-game record against Stockfish is 28 wins, 72 draws and 0 losses, arguably an overwhelming victory.
The following year, it proved its prowess again: 155-6 in 1,000 games against Stockfish, with the rest of the draw winning.
Usually, the fact that one chess program beats another is only circulated in the small circle of enthusiasts, but AlphaZero is no "ordinary" chess program.
You know, the moves of those previous programs required human chess players to first conceive chess paths, then walk out of these chess paths in the game, and then upload them to the network. In other words, these procedures rely on human experience, knowledge, and strategy. The main advantage of these early programs against human opponents was not their ingenuity, but their greater processing power, which allowed them to evaluate more chess options in a given amount of time. In contrast, AlphaZero does not rely on pre-programmed moves, combinations, or any strategy derived from human chess games.
AlphaZero's style is entirely a product of AI training: its builders simply give it a set of chess rules and instruct it to develop a strategy based on the rules to maximize its win-loss ratio.
After just 4 hours of self-play, AlphaZero became the most powerful chess program in the world. As of the time of writing, no one has been able to beat it.
