Greek Computerized Backgammon Program Wins First Prize in AI Competition

During November 18 – 26 in Tillburk, Holland, the 16th annual competition for computerized table games was held, also known as the Olympic Games of game programs, with 59 entries in 18 games.

The best programs of the world featuring table games such as chess, backgammon, and others competed for the title of “Olympic champion”.

These ‘Table Olympic Games’ have been carried out since 1989, in parallel with the Conference for Advances in Computer Games, and the aim is to foster growth and research in the artificial intelligence sector of the games.

In this year’s competition, and more specifically in the sport of backgammon (the corresponding game in Greek is called “doors”), one entry, a program entitled Palamidis, was developed by the Artificial Intelligence Team of the Department of Applied Information Technology of the University Macedonia in Thessaloniki.  The program was designed by Nikolaos Papachristou, doctoral candidate, under the supervision of Professor Dr. Ioannis Refanidis.

Palamidis was able to overcome its opponents and won the “gold” medal in its first attendance in the competition. According to Greek mythology, the invention of dice is attributed to the legendary inventor Palamidis, as are various other table games, as well as the invention of meters, stations, and the division of the year into hours, days and months.

Palamidis is a program that allows users to play the Greek traditional games of backgammon, including “Doors”, “Plakoto” and “Feyga”. In the competition however, only the game “Doors” was featured, since the other games are not popular outside of Greece.

The heart of the program is an artificial neural network, which has learned to play backgammon by playing millions of times against itself and continuously improving its output using a technique of reinforced learning.

Palamidis uses the neural network in order to mark all possible movements that result from rolling dice, taking into consideration all the likely tosses and the possible next movements of the opponent. Finally, it selects the move that maximizes the probabilities of success in the game.