When IBM chess computer Deep Blue defeated the human world champion Garry Kasparov in 1997, it marked a seminal moment in that man was losing his superiority against machine.
Today, machines are catching up to the best players in the more complex shogi and go as computer programs that play board games are being developed as part of research into artificial intelligence.
That challenge loomed as more difficult for shogi and go than in chess because of the larger number of game scenarios computer programs are required to consider.
In chess, an average of 30 to 35 moves are possible in a single scenario, but in shogi there are approximately 80. When human players decide on their next moves, they whittle down their choices to the most promising options using their intuition and contemplate the possible scenarios that lie ahead.
As players gain more game experience, their intuition is honed and the correct move comes to them more quickly. A computer predicts every possible move including wasteful ones that its opponent could make to seize its pieces, which takes a massive amount of time.
To counter this, developers have created programs that evaluate the situation at each stage of a game and concentrate on determining the best moves. By analyzing the pros and cons of pieces and their movements, their judgments have gradually increased in accuracy. In January a program named Bonkras, which had previously won the World Computer Shogi Championship, defeated Kunio Yonenaga, Japan Shogi Association chairman and former meijin.
“It would be fair to say that computers now rank alongside shogi's top professionals,” says Takeshi Ito, an assistant professor of cognitive science at The University of Electro-Communications in Chofu, western Tokyo. “They might even surpass professionals in speed matches.”
In late 2011, Bonkras broke a points record that acts as a gauge of players' ability on Internet match site Shogi Club 24, which is also used by some professionals.
In the case of go, which uses a larger board than shogi, it is difficult to improve the playing ability of a computer program using a method similar to the one previously mentioned.
While the number of considerable scenarios in a game is 10 to the 120th power for chess and 10 to the 220th power for shogi, it is said to be 10 to the 360th power for go. This makes it difficult to quantify positional judgments. In 2005, the playing ability of shogi computers was close to that of a high-ranked amateur, but go computers had not even reached the level of 1st-dan amateur.
However, when a program utilizing the “Monte Carlo method” arrived on the scene in 2006, the playing ability of computers took a great leap forward.
The object of go is to capture territory, so if you follow the rules and keep adding stones, a game will reach a conclusion even if you simply keep adding pieces randomly. The Monte Carlo-method program persistently pushes onward and generates numerous endgames, which it uses to reverse-calculate moves with the highest probability of leading to victory. Its avoidance of making positional judgments for every possible scenario was revolutionary.
In March this year, a program named Zen, which is currently thought to be the most formidable in the world, beat 9th-dan professional go player Masaki Takemiya, 61, despite a four-move handicap, demonstrating its amateur rank-holder level skill. “The Monte Carlo method indeed proved to be effective,” says Ito. “And it will most likely grow stronger.”
So, what is the point of developing hard-to-beat computer programs like these in the first place?
“Their superior algorithms can have applications in other fields,” says Ito. He has been approached by a researcher developing a program to extrapolate the causal connection between genetics and illnesses, who wanted to utilize his thesis on shogi computer programs.
Computers are catching up to and overtaking human beings. Even so, it has taken decades for them to reach this point.
“Computers have finally reached this level thanks to the efforts of many developers, which actually goes to show how amazing humans are,” says shogi amateur 4th-dan Ito. “Shogi computers may have attained professional-level playing ability, but they still lag behind humans in terms of conceptual power and many other qualities.”
The developers of these programs are keenly aware of the depth that shogi and go possess.
UNDER PRESSURE IN A CLOSE BATTLE: AN INTERVIEW WITH RYUO TITLE HOLDER AKIRA WATANABE
The playing ability of shogi computers has already reached the level of professionals. I'm sure many view this as a threat to their livelihoods.
In 2007, I faced off against shogi program Bonanza. It had been described as possessing the ability of a 1st-dan from the professional training institute Shoreikai, so I intended to score a resounding victory.
As it turned out, after around 40 moves I came to the realization that it was quite a worthy opponent. Each of its moves were deliberate and none were questionable, right through to the endgame. In the end, it became a closely fought battle in which I was only one move away from defeat, which put me under pressure.
When the shogi program Bonkras played against former meijin Yonenaga, I acted as commentator. The human competitor's game fell apart to a point from which there was no chance of a comeback, and he was beaten. However, none of the moves made by the computer were impressive. If a human opponent were to play a solid game against it, there is no telling how the match would develop.
Computers have no feel for aesthetics and make moves that go against human theory without compunction. They play haphazard in the early stages of a game, making moves that a human would regard as inferior because they would weaken their own position, or choosing poor ones such as placing a gold general on the first row of the board when closing in on the opponent's king. However, this does not necessarily mean that they are wrong. There are ways of attacking that only computers can see.
Today, many professional players are deriving inspiration from the moves made by computers, and are studying them. I myself use a computer to manage game records, and I sometimes draw on the moves that they come up with. There is no longer any reason to resist such change.
That being said, we cannot rely entirely on computers. Endgame scenarios that hinge on whether the king can or cannot escape or defend itself have a clear objective, which plays to the strengths of computers. Even so, their positional judgments in other scenarios still lack precision. We can use computers for research, but an ability to select a good move from numerous options is necessary.
Next year, a second shogi master-versus-machine match will be held in which five professional players will take on five computer programs. No one will be surprised if newly promoted 4th-dan players (those who have just turned professional) lose. It will be intriguing to watch what happens when the programs come up against higher-ranked opponents.
A native of Katsushika ward, Tokyo. Made his professional debut in 2000 at the age of 15. Won his first title, Ryuo, in 2004, and has successfully defended it seven times to date, earning him the right to be called eisei ryuo (permanent Ryuo champion). Currently holds two titles: Ryuo and Oza.
- « Prev
- Next »