Though last Sunday’s Tokyo assembly elections garnered the most media attention, another contest came in a close second, even if only two people were involved. Fourteen-year-old Sota Fujii’s record-setting winning streak of 29 games of shogi was finally broken on July 2 when he lost a match to 22-year-old Yuki Sasaki.
Fujii has turned into a media superstar in the past year because of his youth and exceptional ability in a game that non-enthusiasts may find too cerebral to appreciate. The speed of Fujii’s ascension to headline status has been purposely accelerated by the media, which treats him as not just a prodigy, but as the vanguard figure of a pastime in which the media has a stake.
Press photos of Fujii’s matches show enormous assemblies of reporters, video crews and photographers hovering over the kneeling opponents. Such attention may seem ridiculous to some people owing to the solemnity surrounding shogi, which is played much like chess, but if Fujii succeeds in attracting new fans, then the media is all for it.
That’s because all the national dailies and some broadcasters cover shogi regularly and in detail. In fact, most major shogi tournaments are sponsored by media outlets. The Ryuo Sen championship, toward which Fujii was aiming when he lost last week, is the biggest in terms of prize money, and is sponsored by the Yomiuri Shimbun. NHK also has a tournament and airs a popular shogi instructional program several times a week.
The Fujii fuss, however, is about more than his prodigal skills. Fujii ushers an old game with a stuffy image into the present by accommodating the 21st century’s most fickle god: artificial intelligence. Much has been made in the past few weeks of Fujii’s style of play, which is described as being counter-intuitive and abnormally aggressive. What almost all the critics agree on is that he honed this style through self-training that involved the use of dedicated shogi software incorporating AI.
But before Fujii’s revolutionary strategic merits could be celebrated, AI needed to be accepted, and a scandal last July put such technology into focus. One of the top players in the game, Hiroyuki Miura, was accused by his opponent of cheating after he won a match. Miura repeatedly left the room during play and was suspected of consulting his phone when he did so. The Japan Shogi Association (JSA) suspended him as they investigated the charges.
As outlined by Toru Takeda in the Nov. 22 online version of Asahi Shimbun, the JSA checked the moves Miura had made in previous games against moves made by popular shogi software to see if there was a pattern. In four of his victories there was a 90 percent rate of coincidence. Miura’s smartphone was also checked by a third party, which found no shogi app. Moreover, there was no communications activity recorded for the phone on the day of the contested match because it had been shut off the whole time.
Miura was officially exonerated on May 24, at the height of the media’s “Sota fever,” but that doesn’t mean Miura was not using shogi software to change his game strategy. In November last year, Takeda theorized that, given the prevalence of the software and the amount of progress programmers had made in improving its AI functions, it’s impossible to believe that there is a professional shogi player who has not yet taken advantage of the technology. Miura, he surmised, had become what chess grandmaster Garry Kasparov once called a “centaur” — half man, half computerized beast. By studying the way shogi programs played, Miura had likely appropriated the AI function’s own learning curve. He didn’t have to check the software to determine moves — it was already in his nervous system. Miura is, in fact, one of the pros who battled computerized shogi programs in past years. In 2013, he played against shogi software developed by the University of Tokyo and lost.
The evolution of shogi software was covered in a recent NHK documentary about AI. Amahiko Sato, one of the game’s highest ranked players, has played the shogi robot Ponanza several times without a victory. The robot’s programmer told NHK that he input 20 years of moves by various professionals into the program and it has since been playing itself. Since computers decide at a speed that is exponentially faster than humans, the software has played itself about 7 million times, learning more with each game.
“It’s like using a shovel to compete with a bulldozer,” Yoshiharu Habu, Japan’s top shogi player, commented to NHK after describing Ponanza’s moves as “unbelievable.”
Fujii is simply the human manifestation of this evolution, and what’s disconcerting for the shogi establishment is that he didn’t reach that position because of a mentor. As with most skills in Japan, shogi hopefuls usually learn by sitting at the feet of masters and copying their technique in a rote fashion until they’ve developed it into something successful and idiosyncratic. Fujii leapfrogged the mentor phase thanks to shogi software.
An article in the June 27 Asahi Shimbun identified Shota Chida as the player who turned Fujii on to AI a year ago, just before Fujii turned pro. On the NHK program Habu noticed something significant as a result: Fujii’s moves became faster and more decisive. He achieved victory with fewer moves by abandoning the conventional strategy of building a defense before going on the offensive. Fujii constantly looks for openings in his opponent’s game and immediately strikes when he sees one, which is the main characteristic of AI shogi.
Fujii’s defeat obviously means that his type of play is no longer confounding. Masataka Sugimoto, his shogi teacher, told Tokyo Shimbun that he doesn’t think Fujii “uses software as a weapon,” since he now faces players who also practiced with AI. But that doesn’t mean his game play hasn’t been changed by AI. Before the Miura scandal, pros who used software were considered the board-game equivalents of athletes who took performance-enhancing drugs. Now they’re the norm, and the media couldn’t be happier.