The AI for Games and Education session consisted of three presentations. The games were completely different, ranging from a topic out of the world of Games in combination with Combinatorial Game Theory, via Aggressive De-escalation, to Poly-Y. This was one of the three starting sessions of the 26th BNAIC. Immediately after the keynote lecture by Simon Colton.
The first presentation was titled “Combining Combinatorial Game Theory with an Alpha-Beta Solver for Domineering”, authored by Michael Barton and Jos Uiterwijk. Jos did the presentation.
The main topic was the combination of two research fields. From the perspective of Combinatorial Game Theory (CGT), the Domineering games of different size were investigated on subgame ordening. The authors used five concepts, viz. (1) unknown value, (2) hot (known value), (3) decreasing temperature, (4) infinitive subgames, and (5) number. Moreover, they exploited the standard move ordening with tie breaks. From CGT, the authors adopted the techniques on the exploration of subgames. From the Games World they took the α-β solver, negascout and the database concept. Emphasis was on pruning. The results were adequate, but much work has still to be performed. Plans and recommendations for future research were given. The article was a candidate for the Best Paper Award.
The second presentation was titled “Towards Aggression De-escalation Training with Virtual Agents: A Computational Model” by Tibor Bosse and Simon Provoost. Tibor Bosse gave the presentation.
The motivation of the research was in the observation that in many jobs we see aggressive confrontations. The authors gave some telling figures in this respect, such as in 10% of the cases, in which a dispute is at stake, there is escalation, and per year 50 drivers in the public transport in Amsterdam are facing aggressive behaviour. The authors distinguished two types of defense in relation to such behaviour: (1) reactive defense relying on empathic statements, and (2) proactive behaviour, i.e., being dominant (i.e., draw a line) and be not empathic. From these two types it is clear that decision making is most important in this research. The authors introduced a virtual instruction. Their model is based on the dynamics of the processes related to interpersonal aggression. In fact, there are two submoedels, viz. the aggressor model and the de-escalator model. The authors informed us on a number of simulation runs under different parameter settings. Both models were promising but are waiting to be integrated in one system and then should be validated.
The third presentation was titled “Monte-Carlo Tree Search for Poly-Y” by Lesley Wevers and Steven te Brinke. Lesley Wevers gave the presentation.
Poly-Y is a generalization of the game Y and that is a generalization of Hex. In short it is a connecting game. Poly-Y is played on a board with an odd number of sides greater or equal to three. The goal of the game is to capture the majority of the corners. A corner is captured by constructing a Y structure. The authors gave many examples in the their article. For reasons of equality there is a swap rule just as in Hex. The authors describe their application of MCTS to Poly-Y and refer to the literature for extensive descriptions. Subsequently they discuss the Monte-Carlo Playouts for Poly-Y with respect to their effectiveness. Then the pay attention to the opening analysis. This is an interesting topic, particularly in relation to the swap rule. For many openings they established win rates. As a consequence they tuned their strategy on their opening findings. “ If the opponent opnes with a move that has less than 50% win chance for us at the deepest level that we analyzed, we apply the swap rule.” Their experimental results were interesting. Finally, in the section future work they recommende some techniques used in the game Y by Saffidine and Cazenave. We look forward to see poly-Y participating in the next Computer Olympiad.
by Jaap van den Herik
