Three research scientists at DeepMind Technologies teamed up with former world chess champion Vladimir Kramnik to “explore what variations of chess would look like at superhuman level,” according to their new article in Communications of the ACM. Their paper argues that using neural networks and advanced reinforcement learning algorithms can not only surpass all human knowledge of chess, but also “allow us to reimagine the game as we know it….”
“For example, the ‘castling’ move was only introduced in its current form in the 17th century. What would chess have been like had castling not been incorporated into the rules?”
AfterAlphaZero was trained to play 9 different “variants” of chess, it then played 11,000 games against itself, while the researchers assessed things like the number of stalemates and how often the special new moves were actually used. The variations tested:
– Castling is no longer allowed
– Castling is only allowed after the 10th move
– Pawns can only move one square
– Stalemates are a win for the attacking side (rather than a draw)
– Pawns have the option of moving two squares on any turn (and can also be captured en passant if they do)
– Pawns have the option of moving two squares — but only when they’re in the second or third row of squares. (After which they can be captured en passant )
– Pawns can move backwards (except from their starting square).
– Pawns can also move sideways by one square.
– It’s possible to capture your own pieces.
“The findings of our quantitative and qualitative analysis demonstrate the rich possibilities that lie beyond the rules of modern chess.”
AlphaZero’s ability to continually improve its understanding of the game, and reach superhuman playing strength in classical chess and Go, lends itself to the question of assessing chess variants and potential variants of other board games in the future. Provided only with the implementation of the rules, it is possible to effectively simulate decades of human experience in a day, opening a window into top-level play of each variant. In doing so, computer chess completes the circle, from the early days of pitting man vs. machine to a collaborative present of man with machine, where AI can empower players to explore what chess is and what it could become….
The combination of human curiosity and a powerful reinforcement learning system allowed us to reimagine what chess would have looked like if history had taken a slightly different course. When the statistical properties of top-level AlphaZero games are compared to classical chess, a number of more decisive variants appear, without impacting the diversity of plausible options available to a player….
Taken together, the statistical properties and aesthetics provide evidence that some variants would lead to games that are at least as engaging as classical chess.
“Chess’s role in artificial intelligence research is far from over…” their article concludes, arguing that AI “can provide the evidence to take reimagining to reality.”
Read more of this story at Slashdot.