KI/RuG symposium Artificial Intelligence in the Wild


Hans in the Wild

(Invited Lecture)

Hans van Ditmarsch

Department of Computer Science

University of Otago

Abstract

In this talk I will present some reality checks, and questions on the relevance of specific logical methods, that I encountered when modelling the dynamics of games and other multi-agent system dynamics.
The first case study is the game of Pit. Pit is a multi-player card game that simulates the commodities trading market, and where actions consist of bidding and of swapping cards. Its dynamical game features can be described in the extension of a standard language for dynamic epistemics. This formalization is then used to outline the game theory for a simplification of the Pit game, which uncovers some interesting equilibria. It is quite hard to compute optimal strategies for generalizations of this very simple game, let alone for the 'real' Pit game. Does this help human players to win Pit?
The second case study is the 'murder mystery' game of Cluedo. It turns out that the complexity of models describing the information in the game gets smaller - but only barely so - as a result of subsequent game actions. If the game had been designed differently, for example, if apart from a murder weapon, murder room, and a murder suspect, one would have had to determine as well the motive for the murder, the complexity might have increased instead. Would that have made the game less fun? Or less playable? Did the inventor of Cluedo know that, back in the 1940s? How do I become a better Cluedo player anyway?
The third case study, time permitting, is about the use of a logic for the dynamics of knowledge and belief, as expressed by preferences, when modelling conversational implicatures and presuppositions in dialogue, and also, in a somewhat different setting of security protocols, intentions. Does this increase our insight into communicative processes? How does this compare with probabilistic approaches? Can it be used in dialogue systems?