Here is a statement from Anders Kierulf, the developer of SmartGo, about SmartGo and its performance at the 2000 US Computer Go Championship.
"SmartGo is being developed by Anders Kierulf at Smart Go, Inc. The program is written in C++, and currently consists of about 50000 lines of code, plus 25000 lines for the integrated GoTools code. It's based on selective whole-board lookahead, looking about 5 moves ahead in the opening and 3 moves during middle and end game. At each position, it chooses 10 to 15 moves to look at using a variety of techniques, including pattern recognition and tactical search. It incorporates GoTools, a sophisticated analyzer for enclosed life and death problems developed by Thomas Wolf, and reasonably good tactics for capturing blocks of stones. For more details on the program, see www.smartgo.com.
Our overall approach to computer Go emphasizes the importance of search to make the program scale well with increasing computing power. We expect that Go programs are at least 15 to 20 years away from beating a professional player, so that should give us at least a factor of 1000 in computing power. It also means that computer Go is a long term project, and thus investments in code quality should pay off. But that remains to be proven.
Both games played by SmartGo were quite interesting. Against The Many Faces of Go, SmartGo managed to kill a group (after a series of mistakes by both programs). SmartGo correctly ignored several moves where Many Faces added stones to the dead group. Unfortunately, SmartGo's poor understanding of connections let the group come back to life late in the game. Against Poka, SmartGo passed once too often and let Poka live inside its territory, thus winning by only 70 rather than 170 points. As yet, SmartGo has no concept of playing safe when it's ahead.
We're encouraged that SmartGo played as well as it did. There are still a lot of weaknesses in the program, in particular connections, group strength assessment, and integration of GoTools. SmartGo is currently being worked on by one person full time, and we hope to address some of these weaknesses before next year's tournament."
The 2000 North American Computer Go Tournament
Other computer Go Tournament results