Monday, November 7, 2016

Fritz 4

     While browsing through Alex Yermolinsky's Road to Chess Improvement (published in 1999) there is a section at the end of the book where he presented some 10-minute games he played against Fritz 4 that I found interesting, especially when comparing his GM commentary on Fritz 4's play and the evaluations of today's Stockfish engine. 
     Fritz is a German chess program and published by ChessBase; the latest version, now based on Rybka, is Fritz 15.  The engine was produced in the early 1980s. In the early 1990s the Fritz chess programs were called Knightstalker in the US.  In 1995, Fritz 3 won the World Computer Chess Championship in Hong Kong, surprisingly beating a prototype version of Deep Blue. This was the first time a program running on a stock PC defeated the supercomputers and mainframes that had previously dominated this event. 
     Yermolinsky gave the following weapons at the human's disposal when working out the best move in a complicated middlegame position: opening theory, positional understanding, endgame theory, horizon effect and intuition. There are also numerous psychological factors involved when humans play. 
     But, already by by the time of Fritz 4, the day when engines could be defeated by positional play based on general principals and waiting for them to self-destruct were gone.  By the mid-1990s engines had modern opening books and they were making decent moves while waiting for tactics to show up and then humans faltered.  When Yermolinsky wrote his book he was already thinking that the time had come that when a man beat a machine it should be considered an upset. 
     When playing against Fritz 4 Yermolinsky figured he would lose long games, so his strategy was create complications early, usually by playing to its greed. Offer it a Pawn for the initiative and positional compensation like weak squares, or its willingness to accept a passive position for a small material gain, or its neglect of King safety, etc. In those days engines had little understanding of having compensation for material. 
     Even as late as 2004 when Correspondence GM Robin Smith published Modern Chess Analysis wherein he told how to beat chess engines he wrote, "When the going gets tactical, the computers get going." And, he stated that they had three strengths relative to people: calculation, calculation and calculation, observing that within their search horizon, often six full moves (!), they didn't miss a trick. Engines no longer self-destructed if you could avoid tactical errors, but they could be outsmarted positionally if humans had enough finesse and knew how to play to their weaknesses. Needless to say, progress was so swift that the book was quickly outdated and today, a dozen years later, his advice is pretty much useless.
     Here is one of Yermoinsky's better games against Fritz 4.
 

No comments:

Post a Comment