• PhD Student:
  • Alix Goudyme
  • Funding : Artois
  • PhD defended on :
  • Jun 16, 2022 • thesis room, Faculty of Science Jean Perrin, Lens

Description

The concept of intention and its link with what an agent knows is an important concept in Philosophy, Logic and Artificial Intelligence [4,5] and can be found in many applications such as negotiation, law or argumentation. In this thesis, we are particularly interested in the concept of intention in the context of games. The game universe provides a controlled environment but nevertheless rich enough to test the different methods that can be developed. We restrict the object of study to games with incomplete information and potentially asymmetric (games where the players do not have the same goals or the same rules). In these games, taking into account players’ beliefs is crucial to be able to play well.

A game in which players must take into account the beliefs of other players is called an epistemic game. Known examples of this type of game are Cluedo, 7 Families, Mr. Jack (asymmetric), and the cooperative game Hanabi. The representation of such games can be done using generic languages such as those proposed in [1,2]. The moves chosen by some players can reveal certain intentions. For example, in a card game, if an opponent makes a move, it means that he most certainly has a certain card and that he has a certain goal. Or, in a competitive game, he intends to make his opponent think he has that card in order to make him make a bad decision. Or in a cooperative game such as Hanabi [3], to let the opponent know that he has the card.

It is potentially very costly to detect the intent of other players. In a live game, the time for each player to make a decision is often very short and it can be tricky to take into account both the beliefs and intentions of other players. On the other hand, it is much more feasible to study the play of different players offline and dissect what a player’s intentions were at a given moment of the game. This can be used to question some bad moves or to explain the best ones. The logical aspect allows a better “explicability” of the moves of the different players.

Our goals will be to :

  • propose a logical modeling of the intention in the framework of epistemic games,

  • to apply this modeling to different well known games allowing to explain the course of games,

  • generalize these methods to generic game representation languages.

References

[1] Michael Thielscher, GDL-III: A Description Language for Epistemic General Game Playing in Proc. of the 26th International Joint Conference on Artificial Intelligence, IJCAI-17, p1276-1282, ijcai2017

[2] Guifei Jiang, Dongmo Zhang, Laurent Perrussel, Heng Zhang: Epistemic GDL: A Logic for Representing and Reasoning about Imperfect Information Games, in Proceedings of the 25th International Joint Conference on Artificial Intelligence, IJCAI 2016: 1138-1144

[3] Markus Eger, Chris Martens, Marcela Alfaro Cordoba: An intentional AI for Hanabi in Proc. of IEEE’s 2017 Conference on Computational Intelligence in Games.

[4] M. E. Bratman. Intention, Plans, and Practical Reason. CSLI Publications. 1987.

[5] A. S. Rao and M. P. Georgeff. BDI-agents: From Theory to Practice, In Proceedings of the First International Conference on Multiagent Systems (ICMAS'95), San Francisco, 1995.