Beating people at Scrabble is already no contest for computer programs, which can easily memorise entire dictionaries. Now a Scrabble-playing program has gone one better by playing dirty.

Developed by Eyal Amir, an assistant professor and PhD candidate Mark Richards, from the Department of Computer Science at Illinois, the program is able to predict which letter tiles other players hold, and use this information to choose moves which block a high-scoring word that an opponent might otherwise have played. This aggressive gaming style gives it the edge over previous Scrabble programs, which focus solely on maximising their own scores.

To predict what tiles other players hold, Amir and Richards's program begins by eliminating those tiles that have already been played. It then narrows down the possibilities by assuming that the tiles left on an opponent's rack after they make a move do not include any letters that could have been used to form higher-scoring words than the word the opponent actually played. Adding in this "opponent modelling" greatly improved the program's game, allowing it to beat Quackle, one of the best conventional Scrabble programs, by five points on average.

Amir says the program can do more than simply beat its rivals. Because its play is more human-like than other Scrabble-bots, it could serve as a useful tool for training people to play against other people.

"The empirical results discussed in our paper suggest that opponent modeling adds considerable value to simulation," Amir concluded. "We do not expect that the value of information gained through opponent modeling will be the same in all situations. In particular, we expect the value to vary with the number of unseen tiles and with the number of tiles played by the opponent on his previous moves. Efforts are currently underway to analyze when the opponent modeling is most helpful."

Last year, Amir was chosen as one of the top young artificial intelligence researchers. He was featured in an article on the "AI 10 to Watch" in IEEE Intelligent Systems, a prominent publication that focuses on the field of artificial intelligence.