Mentone Chess Club

Results of 100 AlphaZero v Stockfish 8 games

AlphaZero, the latest AI program developed by DeepMind has proved that it is capable of mastering chess.
Demis Hassabis, the founder and CEO of DeepMind and an expert chess player himself, presented further details of the system, called AlphaZero, at an AI conference in California.
The program often made moves that would seem unthinkable to a human chess player. “It doesn’t play like a human, and it doesn’t play like a program,” Hassabis said “It plays in a third, almost alien, way.”

[EDITORIAL]
From his statement, Mr Hassabis shows he is unfamiliar with Quantum Theory. “It plays in a third, almost alien, way.” is the third option of Quantum superposition theory.

      1. Plays like a Machine.
      2. Plays like a Human.
      3. Plays like both a Human and a Computer.
      4. Plays like neither a Human or a Computer.

AlphaZero has shown that machines can become superhuman without help and that the event horizon where this happens is in the near future. In 24 hours, AlphaZero taught itself to play chess well enough to beat one of the best existing chess programs around. Stockfish 8 beat all other standard chess engines in a tournament recently.

What is remarkable is that in one game it made seemingly crazy sacrifices like offering a bishop and queen to exploit a positional advantage that led to victory. Such sacrifices of high-value pieces are normally rare. AlphaZero benefits from not following the usual approach of assigning value to pieces in an attempt to minimize loss. Our human pre-conceptions about the value of chess pieces, limits our ability to play the best move even if that move IS to sacrifice the queen.

The best chess programs, developed and refined over decades, incorporate huge amounts of human intelligence and although in 1996 IBM’s Deep Blue beat then world champion Garry Kasperov, that program required careful hand-programming. AlphaZero is different in that it utilizes an interaction of two artificial neural networks (ANN), a "policy network" to define candidate moves, and a "value network" to evaluate positions. ANN is a computing system inspired by biological neural networks.

Such systems learn progressively from performing tasks and improve performance from all experiences previously gained. Very much how a human learns when not provided with task-specific instruction i.e. by utilizing an instinctive sense of board position and pattern matching. AlphaZero learned chess by ingesting thousands of example games and then practicing against a clone of itself. This approach is known as reinforcement learning and is modelled on biological methods of learning.

Here are links to few of the YouTube videos released by Google, with inciteful analysis and commentry by IM Daniel Rensch.

Game  3  Click to watch
Game  5  Click to watch
Game  8  Click to watch
Game  9  Click to watch
Game 10 Click to watch
ENJOY!