Chess has captured the imagination of humans for centuries due to its strategic beauty—an objective, board-based testament to the power of mortal intuition. Twenty-five years ago Wednesday, though, human superiority on a chessboard was seriously threatened for the first time.
At a nondescript convention center in Philadelphia, a meticulously constructed supercomputer called Deep Blue faced off against Garry Kasparov for the first in a series of six games. Kasparov was world chess champion at the time and widely considered to be one of the greatest players in the history of chess. He did not expect to lose. It was perhaps understandable; 1996 was an age of fairly primitive computer beings. Personal computers were only just becoming a more affordable commodity (35 percent of U.S. households owned a computer in 1997, compared with 15 percent in 1990), the USB had just been released, and it would be another five years until Windows XP made its way onto the market.
But Deep Blue was no run-of-the-mill computer. It was a behemoth built with the sole intention of being very good at chess. And it fulfilled that mission. On Feb. 10, 1996, the reigning world chess champion lost a game to a computer for the first time in history. Kasparov would win the 1996 match four games to two, but in May 1997, an upgraded Deep Blue would defeat Kasparov 3½–2½.
The ’96 match nonetheless demonstrated that the tide was starting to turn in the chess world, and the tide was deep, blue, and electronic. It introduced chess computers to the world, sparking conversations about a rise of automation in the famously romantic field.
Some version of computers had been playing chess even before the emergence of artificial intelligence as an official field in the 1950s. Alan Turing, the famous cryptographer, had developed a handwritten chess algorithm in 1950 called “Turochamp.” In 1957, Alex Bernstein, a researcher and chess enthusiast from the Bronx, created the first complete chess program with the help of a number of his IBM colleagues.
“Computer chess changed in the ’80s.” says Jonathan Schaeffer, president of the International Computer Games Association and professor of computer science at the University of Alberta. That decade, pioneering American computer scientist Ken Thompson released a paper proving something that now seems intuitive: If your computer was faster, your chess program would perform better. Programs could thus analyze more and more moves per second, increasing their chances of finding the best move possible.
Accordingly, computer chess “became about getting the fastest technology. When I started in the [computer chess] game, we were using a single computer. Then it became 16, then 210, and so on to chips and supercomputers,” says Schaeffer. In 1988, students at Carnegie Mellon University developed a sophisticated chess computer called Deep Thought. In January of that year, Deep Thought became the first computer to beat a grandmaster in a regular tournament game when it triumphed over Bent Larsen, a Danish GM. The next year, IBM hired three of those Carnegie students, Feng-hsiung Hsu, Thomas Anantharaman, and Murray Campbell, with the express aim of building a chess computer to rival the world champion; they would be joined by Chung Jen-Tan, Joseph Hoane Jr., and Jerry Brody later in the project. In October 1989, Kasparov played two games against Deep Thought, winning both of them with ease.
The first match demonstrated that the tide was starting to turn in the chess world.
“The loss to Kasparov in 1989 demonstrated the amount of work that needed to be done,” says Schaeffer, “so they took it to the extreme. They went off for seven years and built new computer chips that were faster, building a system that was scaled up to not just four computer chips, but 500. They added more knowledge to it as well as a book of openings, and eventually the brain of chess grandmaster Joel Benjamin helped provide expertise. This was a very long project involving many, many people, and significant financial expense, but it paid off for IBM in the form of media clamor.”
The 2,800-pound Deep Blue, complete with special-purpose chess computer chips, was the end product. It was capable of processing 200 million moves per second, or 199,999,997 more than Kasparov could manage, according to IBM. This produced a chess machine that was stronger than any of its automated predecessors, and the outside world was stunned at the eventual result—a human had been outdone by a machine in this game of intellect, wit, and judgment. At the 1997 match, Kasparov and Deep Blue would go toe-to-automated-toe in front of numerous television cameras and a large crowd.
But Kasparov’s loss was not as devastating as casual observers might have expected. Computers had beaten grandmasters before; it was inevitable that someone of Kasparov’s stature would fall too. And though Kasparov’s loss certainly came earlier than expected, the competitive chess world continued to go about its business relatively unfettered.
“I don’t think it affected chess players too much,” says Matthew Sadler, chess grandmaster and co-author of Game Changer, a book about modern chess engine AlphaZero, “Firstly, Kasparov was probably stronger than Deep Blue at the time, despite the loss. Secondly, it didn’t really inspire any chess players with its play.”
It helped that Deep Blue, at the time, was the exception rather than the rule—machines of its strength weren’t widely available. In 2006, though, a chess computer called Deep Fritz beat then–world champion Vladimir Kramnik. “I think that’s really when chess players sort of thought, Oh, my goodness, the machines really are getting stronger than us,” says Sadler, “when they were beating us not on supercomputers, but on relative commodity hardware.”
The change here wasn’t just that a computer could win, but that a computer could help human players win if incorporated into their training regimes effectively. Computers were adept at judging the quality of moves and positions accurately, particularly during opening sequences. Some found this easier than others. Sadler says: “I think a lot of competitive players took a while to adjust to the new reality. For example, if you weren’t really computer-literate, and all of a sudden you found yourself in a world where having a computer really makes a difference, that’s a difficult thing.”
Despite initial resistance from certain parts of the community, the advantages that computers afforded chess players eventually made them impossible to ignore. Sam Shankland gained his international master title in 2008, right around when computers started to become a necessity. “There was some backlash, but honestly, those people are mostly gone now,” Shankland, now a grandmaster and 2018 U.S. chess champion, says. “They either got tired of losing and quit chess or they got tired of losing and adapted.”
The sheer wealth of knowledge chess players now had access to meant that determination was increasingly rewarded. “I think that chess is essentially a subset of talent and hard work,” says Shankland, “and as training resources like computers become better and more accessible, talent tends to become less important compared to hard work—which suits a workhorse like myself.”
Such accessibility has also led to chess, once reserved for rich families who could afford tutors and other training, to become a markedly more democratized pursuit. “Take India, for example,” says Shankland. “Apart from Vishy [Anand], they weren’t a particularly strong chess nation historically. Now, they’re clearly the fastest-growing country in the world in terms of rising stars, and I think a lot of that is down to training resources becoming more widely available.”
The availability of advanced chess analysis at the flick of a smartphone has caused a bizarre balance of power in the media and a certain trepidation among top-level players, as Peter Heine Nielsen, coach of current world champion Magnus Carlsen, points out:
When I started working with Vishy Anand, at a postgame press conference the players would explain the games, and everybody would look at them with excitement and think, “Wow, these guys are clever.” Now, the player in the press conference is a bit nervous because they have only calculated themselves, while all the journalists have been using advanced technology. So they are afraid to say, “I thought this wasn’t a strong move” in case they’re wrong.
So sometimes before a press conference I speak to Magnus and tell him the computer said this or that, just so he knows. The spectator-player dynamic has changed a lot—some of the mystery has gone.
However, while certain “human” aspects of the games have disappeared, recent developments have caused professional players to rethink what they know about their beloved board game. In 2017, a team of scientists at Google-owned DeepMind created AlphaZero, a self-learning “neural network” program that surpassed the strongest chess program after just four hours of playing against itself.
“Before the computer boom, and before the neural network boom, we were thinking quite dogmatically,” says Nielsen. “After both occurred, we were forced to rewrite our own solutions. It led to the game becoming more exciting.” Moreover, the two strongest chess engines—Leela (which is based on AlphaZero) and Stockfish—are available online, which signifies a remarkably more distributive and collaborative approach to chess innovation than that which was pioneered by Deep Blue, a closed circuit.
Despite all their progress, there are still some goals to which innovators in the chess world can aspire. “The next step is for engines to explain what they’re doing,” says Sadler, “so that the average player can understand why an engine says, ‘No, trading that piece is a bad idea.’ ” The relationship remains one of reciprocity.
One thing is certain: Chess programs will remain the most important piece of a professional player’s preparatory arsenal. “Not using a computer to do chess would be like not using a calculator to do math,” says Nielsen, “I like it—but it doesn’t matter if I like it or not. It’s the right way to do it.”