Wednesday, February 13, 2013

The Draw Death of Chess

Jose Raul Capablanca, chess prodigy and a world champion, proposed his chess variant sometime in the 1920's. He believed chess would be exhausted in the near future, that games between masters would always end in draws. Avoiding this perceived "draw death" of chess was his motivation for promoting a new version of the game. In "Capablanca Chess", the board is 8x10 (instead of the standard 8x8) squares because each side has additional pieces, the Chancellor, the Archbishop and two additional pawns. This configuration also necessitated some minor rules changes. Most players of that era did not share his concerns; "Capablanca Chess" failed to capture the imagination of the masses.

Seventy-odd years later, America's greatest player made a similar proposal. Bobby Fischer firmly believed the chess openings had been over-analyzed to the point that games were decided by opening preparation alone, that masters were no longer free to play creatively. The prevalence of chess engines and computer databases also caught Fischer's attention as he prepared for his return match with Boris Spassky. Instead of adding new pieces, however, in 1996 he proposed (with a few rules modifications) random placement of the standard pieces behind the wall of pawns in the initial position. The effect would be to negate all opening theory to date, throwing players onto their own resources from the very first move. The suggestion was not original, as "shuffle chess" had been considered before, but with Fischer's support the idea actually caught on and what started out as "Fischer Random Chess" is now played as "Chess960", including tournaments for computers and human players. The new name, "Chess960", is allegedly based on the number of possible starting positions.

                                  The World Correspondence Chess Championships

We must preface the following by pointing out the use of computers (aka chess "engines") to assist in selection of moves is currently legal in ICCF (International Correspondence Chess Federation) events.

By the 1980's, all pretense that development of chess-playing programs would lead to advances in artifical intelligence or human problem solving had been abandoned. Prestige and profit became the driving forces as stand alone micro devices, followed by software for the exploding pc market, flooded the commercial scene. The defeat of grandmasters (HITECH defeated GM Arnold Denker in 1988; DEEP THOUGHT defeated GM Bent Larsen in 1988 and GM Robert Byrne in 1989) by computers in exhibition matches was indeed newsworthy but did not have much effect at the (ICCF) championship level. Due to their lightning-fast calculations, the programs were formidable when the human opponent was under the time constraints of tournament competition. Of more interest to top correspondence players was openings research, formerly a tedious process involving notebooks and chess libraries, now streamlined by the database capabilities of ChessBase. Avoidance of ruinous errors, by "blunder checking" with an engine before transmitting the next move, became routine.

On average, in the first 15 world correspondence chess championship events (1950-1996) white won 37%, black won 24% while 39% of the games were drawn. These percentages are in line with established norms and confirm that having the first move gives white a slight advantage. However, a noticeable change occurred during the period 1996-2007 (ICCF World Championships 16 - 23.) White's average winning percentage dropped to 27%, black's to 12% and draws increased to 61%. Significant developments took place in the technology sector, innovations that would forever change the computing landscape. Microsoft's Windows operating system became available in the 1990's. In 2000, the first windows version of Fritz was released, overcoming the general public's aversion to DOS-based chess programs. Other computer game manufacturers rushed their windows chess programs to the retail market. The popularity of windows coupled with stronger chess engines meant the end of stand-alone, micro chess devices. Windows XP was released in 2001, the 32-bit system with 4 gigabytes of memory, the professional (64-bit) version with up to 128 gigabytes memory.

Although dual core processors were commercially available in 2005, Rybka's beta version was released as a single-processor program. By 2006, quad core processores were commercially available and Microsoft had released Vista in several iterations, with up to 128 gigabytes of memory in the 64-bit version. Rybka version 2.1 was released in 2007, both single and multi-processor versions. With multi-core speed and superior performance, Rybka surpassed Fritz as the engine of choice for serious analysis. Fritz/ChessBase maintained a loyal following due to database features.

Rybka continued to dominate, releasing version 3 (32-bit and 64-bit, single and multi-core versions) in 2008, Deep Rybka for multiple processors in 2009 and Rybka 4 in 2010. Fritz likewise continued to improve, releasing version 12 (new engine, improved database) in 2009 and Deep Fritz 13 (multi-core version) in 2012. Quad core and then multi-core processors became available, starting in 2008. Microsoft released Windows 7 in 2009 and Windows 8 in 2012, both systems capable of utilizing multi-core, multi-thread processing and 192 gigabytes of memory. It is the combination of processing speed and humongous amounts of memory that allows these programs to analyze chess games (thousands of positions per second) to greater depths in lesser amounts of time. This efficiency enables correspondence players to successfully engage in multiple, simultaneous top-level games, such as preliminary and championship rounds with 15-player sections.

And what of "Fischer Random Chess" or "Chess960"? Lest anyone think this variant will avoid the draw death, Rybka plays "Chess960" very well, winning the last three (2007-2009) "Chess960" World Championship computer events. Rybka's dominance in this variant led to cancellation of what had been an annual computer championship, perhaps a portent of things to come.

In what may be the final phase, beginning in 2008 (ICCF World Championships 24-27) the average percentage of draws exceeds 80% (the most recent - the 27th Championship - has over 90% draws!) We don't yet know the average percentage of white and black wins because the ICCF tournament secretary for the 25th Championship has hidden the game scores from public viewing until the event is completed (just one game remaining.) In the other three events, the average winning percentage for white is down to 13% and for black just 4%. Another telling statistic involves the "upset" win, something that occurs when a player in the lower half of a cross table scores a victory over a player in the upper half of the event. In the first 15 championships, upset wins were on average 10% of the games. In championships 16-23, upsets dropped to 4% of the games. Since 2008, events 24-27, upsets are less than 1%. There are a small number of  games still incomplete in the last three events, not enough to alter this trend.

Interested readers can find statistics, cross tables and pgn game downloads for all the ICCF World Championships on the CCLA web site: . Information is continually updated as final games are completed in championships 25-27. The accuracy of today's engine-assisted players is such that it's becoming nearly impossible to avoid a draw result. An impartial examination of the 27 ICCF World Championships (1950 - to date) must conclude the "draw death of chess" is now a distinct possibility. The future of correspondence chess is perfect play and it's going to be perfectly boring.