Wednesday, July 30, 2014
Freestyle, Advanced and Centaur Chess are all similar in that humans are allowed to consult engines and make use of any technical or human assistance in selecting their moves. The main difference between these forms of chess and modern correspondence chess is the time limit.
In the first such tournament in 2005 Garry Kasparov was favored but it was won by two chess computer geeks from the US. The computer chess ‘specialists’ were able to make use of all available resources including hardware and software.
Centaurs, as the humans are known, use a program to explore the results of candidate moves. The human players are still fully in control of what moves are made. Proponents of this form of chess claim it increases the level of play, produces error-free games and spectators can gain insight into the thought processes of strong human players and the inner workings of strong programs. Almost all top level correspondence chess these days is played by Centaurs. If you are not a Centaur, you aren't playing at the top levels. Unfortunately engine users have filtered down into the lower ranks, but that is another issue. Engine use would not even be an issue if Centaurs would stay in their own realm and would stay off sites where they are not wanted.
In the competition between Centaurs and pure engines, Centaurs generally have a considerable advantage, especially when playing White. The conclusion is that the advantage of the Centaurs lies mainly in the exploitation of opening advantages. In one large event 21 of the 30 participants were using Stockfish, Houdini and Komodo. These engines, according to the participants, were chosen because Stockfish has a greater depth of calculation and because it is presumably tactically the strongest engine, Houdini is considered the most balanced and positionally a reliably engine, and Komodo because it has the best endgame knowledge. There is also a private engine called Deep Cryptic Cyclone 2.7 or Cryptic Heroes 1.1 which has been used by two participants from Abu Dhabi on pretty powerful hardware. Like Houdini and Stockfish this engine (the real names of the developers are not known) is similar to earlier versions of Rybka and Fruit. Supposedly the engines are run on two Xeon processors with 20 or 16 cores.
There are several reasons why Centaurs play such strong chess. Centaurs are good at selecting promising middle and endgame strategies which often begins in the opening. By using different engines, superior plans can be discovered. This is especially true if the Centaur knows which engine his opponent is using because they can try to exploit that particular engine’s weaknesses. Also, a lot depends on the hardware available to the Centaur. Of course humans are still prone to errors and make mistakes. In general, a superior hardware is an advantage, particularly when depth of calculation is required in critical positions. In correspondence play on sites like Lechenicher SchachServer and ICCF where engine use is allowed, even at lower levels of play, recognizing critical positions is important. For those who let their engines run a minute or two and then select whatever move it suggests, missing a critical point usually means losing the game because the cursory analysis will be faulty.
One of the top centaurs, a fellow named Nelson Hernandez, is not a rated player and claims not to have played an OTB game in over 20 years; he has stated that if he had a rating, he does not think it would be over 1100! In Hernandez’ view pre-game preparation is more important than the in-game performance, and this consists of deeply considering strategy like deciding how you are going to make decisions during the game, what repertoire of lines you will prepare, what you intend to do against specific opponents, what behavioral disciplines you intend to follow during the match, what opening books to use during the match. In top-level CC and Centaur play you usually see solid mainline openings being played because most of the top level OTB players play them because they are sound and follow the best principles of development. Thus, a top notch opening book is required in Centaur. Those crappy books containing games by low rated players that come standard with most programs simply won't do.
Another requirement is knowing which PV to follow in the multi-PV mode. The ‘PV’ is the Principal Variation and is the line the engine thinks is best. There are positions in which the engine needs help and knowing when that is does require some chess expertise. One top Centaur observed that an engine alone plays random openings, wastes time on forced moves, wants to play for the win, sometimes forcing the position and those traits mean it risks losing, especially when a draw would have been best, so relying entirely on an engine may be risky. At the highest levels of play the draw rate is very high because players make very few errors and rarely take risks. The result is these games are usually boring and rarely would anybody want to play over one.
For anyone interested in trying this kind of chess, testing engines or trying out opening experiments I can recommend Lechenicher SchachServer. Engine us is allowed, the interface is nice and it’s free.