Artificial intelligence has been beating us at chess for years. Now he is also teaching us to improve it

  • 7

May 11, 1997. The date is written in fire in the memory of Garry Kasparov, that of the IBM developers, the chroniclers and, in general, that of any chess fan or even the history of the 20th century. After a long tug-of-war, a round-trip competition whose first games dated back to the beginning of 1996, Deep Blue, a powerful supercomputer designed by Armonk’s multinational, definitively doubled the pulse of the Russian Grand Master and defeated him in a field in which he had been practically unbeatable: the board of white and black squares.

Beyond the game, chess, strategies with queens and affixes, Deep Blue’s victory showed us the potential of machines to beat even the best humans in their field. The IBM supercomputer did it in the late 1990s and it’s still doing it today, a quarter of a century later, GT Sophy on the PS4 with the Gran Turismo. And that to cite just one example.

The fact is that the machines —and nowadays artificial intelligence in a special way— may not only serve to get the colors out of us and make the great flesh and blood champions bite the dust. Perhaps they can also help us to compete better. Or even, twisting the loop a little more, to improve the rules and the general conception of our own games.

A gifted student… with a silicon brain

Can the student teach the teacher not only how to play, but how to improve the game?

AlphaZero, the program developed by deep mind —a laboratory linked to Alphabet Inc.— to master chess, shogi and go, three of the most complex and demanding board and strategy games for the human brain.

As between, he partially reveals his own name, AlphaZero differs from other previous versions of artificial intelligence for its ability to learn how to play “from scratch”, based solely on knowledge of the rules. It is not “fed” with huge amounts of games and examples inserted by its creators. No. Without human supervision, by repeating its own “auto-plays” over and over again, AlphaZero develops its own game strategies.

Over the last few years that has worked like a charm for him to reign over the board; but now the experts have gone one step further: they propose harnessing the power of its silicon brain to achieve ‘better’ chess. “AlphaZero’s ability to continually improve its understanding and achieve superhuman playing strength in classical chess and Go lends itself to the question of evaluating chess variants and possible variants of other board games in the future,” says a group of researchers in the magazine Communications of the ACM.

In the battle between artificial intelligence and humans, we are losing.  At least in the Gran Turismo

“With only the implementation of the rules it is possible to effectively simulate decades of human experience in one day, opening a window into the high-level gameplay of each variant. Computer chess comes full circle, from the early days of man and machine colliding to a present of man-machine collaboration, where artificial intelligence can empower players to explore what chess is. and what it could become”, reason the authors in the publication of the Association for Computing Machinery.

This is not, they insist, to question the classical chess model — “it is still fascinating and unlikely to go out of style”, defend—, but to look for alternative variants to achieve a “more creative” game. It is also nothing new, nor a turn of the wheel in the history of chess itself. As the authors remind us, the game we know today is the result of the adjustments applied since the 6th century and there are figures that we have fully internalized today that are relatively frequent. castlingfor example, was applied in its current form in the 17th century.

Convertkit D7jpea8l6ra Unsplash

“What would chess have been like if castling hadn’t been built into the rules? Without resorting to repeating history, we reinvented chess and addressed those questions with AlphaZero.”

The authors formulate, in particular, nine “changes”, some already raised between the courts themselves over the years. the table of Communications of the ACM with the variants you can consult it under these lines, but in general three modifications are proposed: dispensing in part or totally with the castling rule; a number of alterations to the moves pawns can make, including backward and sideways moves; the “self-capture” option, which would allow the judges to get hold of their own pieces; and an alteration that would consider a game ending in a draw as a victory for the attacking player.

T1

The goal is to combine the power of AlphaZero and human curiosity itself to “reimagine what chess would have been like if history had taken a slightly different course.”

“When the statistical properties of high-level AlphaZero games are compared to those of classical chess, they appear a series of more decisive variantswithout affecting the diversity of plausible options available to a player”, reflects the article, which concludes that, both statistically and even aesthetically, “some variants would give rise to games that are at least as attractive as classical chess “, the authors reflect.

Some of his proposals are in fact more than that.

A first tournament without castling was already held in Chennai at the beginning of 2020 and there are portals where you can play with some of the modifications they suggest. In their case, for each variant they used the AlphaZero model in 10,000 autoplay games at one second per move and 1,000 games at one minute. His tests yielded interesting results. For example, those of the second type required between 62 and 76 movements, which doesn’t make a big difference with respect to the 68 that is usually registered, on average, in the games that are played with the classic rules.

Reimagining Chess with AlphaZero desde CACM on Vimeo.

Whether you are more or less in favor of touching the canon of chess, what seems undeniable is that computer programs are already influencing the game. It is not necessary to go to the example of Kasparov and Deep Blue. the article of Communications of the ACM It reflects, for example, how the number of “decisive games” in super tournaments has decreased and that players take longer to go from preparing at home to executing original moves on the board.

An infinite game of chess with AI engines: this is the fascinating experiment that you can see on YouTube

The idea would not differ much either from the philosophy of random chess raised by another Grand Master and former world champion, bobby fischerwho proposed a fortuitous start of the games to prevent dominance in opening strategies from marking the rest of the game.

The objective: achieve “more creative” games.

Although in this case it means letting the teacher be advised by a pupil with a silicon brain.

Images | Jani Kaasinen (Unsplash) and ConvertKit (Unplash)

May 11, 1997. The date is written in fire in the memory of Garry Kasparov, that of the IBM developers,…

May 11, 1997. The date is written in fire in the memory of Garry Kasparov, that of the IBM developers,…

Leave a Reply

Your email address will not be published.