I read a most interesting blog post called “Chess Programs Are Not Smart” on zenpawn’s blog. This post was an attempt to generate discuss by providing a guest article written by Thomas Hall. I must confess that I used to play a lot of chess in the old days. Now that I am a software developer, chess programs are especially an interesting topic to me. First I plan to review what the author of the guest article stated. Then I will chime in with my own two cents on the subject.
Hall agrees that chess programs of today are indeed very successful. Two years ago one of them beat the human world champion at the time. However the problem of making a good chess program is difficult due to the number of possible moves in chess. No matter how great the hardware, the computer is not able to consider all future possibilities for any move. Therefore the computer must make some crucial decisions to arrive with a move in a timely fashion. The open and end game are especially difficult for a computer to analyze. Therefore most good chess programs rely on opening books and end game tables to determine their move. These techniques are wrote memorization and not any real insightful intelligence.
My own take mimics the initial statements by Hall. I would also add that the goal of good chess programs is to win. In that regard, they are most successful. Their goal is not to further the research of computer artificial intelligence in games. If that were the purpose, then I would imagine the programs would not be very good at winning. I am a coder. Zenpawn (the owner of the original blog) is also a software developer. My thought is that since we can code, we should ourselves write up some AI programs to get out computers to actually learn to be good at chess. I mean can’t we teach the computer the rules, let it play against itself for a long time, and make it learn?
Yes tasks like this are easier said than done. However it would prove fruitful in the sense that we would actually be trying to advance the world of computer artificial intelligence. The point here is that we should not look to commercial successful chess programs to do that. They are too busy beating the pants off grandmasters to bother with AI research. Cheers. And to close with a cheesy line, checkmate.
Mysterious Double Instance Hampering Performance - I study the existing code base. Confer with a colleague. Then I determine the optimal plan to change the functionality to load only a slice of all the dat...