LOFT06: Logic and the Foundations of Game and Decision Theory

点击翻译按钮获取中文摘要

We propose to host the conference on Logic and theFoundations of Game and Decision Theory (LOFT) in the UK, 2006.This highly ranked international conference brings togetherresearchers in computer science (with an interest in interaction and communication, particularly in game-likedistributed systems, i.e., multi-agent systems -- MASs),economics, game theory, logic and knowledge representation.Key issues of LOFT are:1 Interaction: Where game theory inspired MASsregarding competitive and two-person games, MAShelps to further the theory of cooperative games and games inwhich coalitions play a role. Dynamics is important: there is nointeraction in a static world, and many questions in games dealwith equilibrium.2 Information: in realistic problems agents have imperfectinformation and memory. They face a trade-off between actingupon the given knowledge, or first acquiring more of it. Again,dynamics is important: players typically revise their beliefsduring a game.3 Agency: Reasoning about players is easier whenassigning them certain informational, motivational and evenemotional attitudes: agents must have beliefs and goals in orderto be able to act (the pre-dominant assumption in game theory isthat of rationality: agents maximise expected payoff). On top of these themes, one of LOFT's defining values is itsinter-disciplinarity. Where the initial aim of LOFT was to bringtogether Economists and Computer Scientists, now there is alarge group of people integrating and further developing`LOFT-issues': researchers in Game Theory, Logic, ComputerScience, Philosophy, Cognitive Psychology, Epistemology,Artificial Intelligence and MAS.Although the UK has outstanding figures and groups inmany of those areas, the interest in the combination of theLOFT-themes emerged relatively late, and, in our opinion, onlyafter the research in MAS in the UK had established itsworld-leading level. In agent research, it is now common toconceive of a MAS as a computational game, and issues in MAS,like partial information, bounded computational power, iterativeinteraction, and coalition formation have provided `classic'game theory with new view-points and problems. Here is anexample: Where Game Theory takes a game (or an interactivescenario) for granted, and then tries to single out certainpatterns of behaviour (solution concepts), the MAS community,with a drive to built agent systems, has also addressed what itcalls Mechanism Design: can we, given some desirable behaviourof the overall system, design the agents and the protocol insuch a way that this behaviour will be generated by rationalaction of the agents?Moreover, MASs provide rich application domains ofnon-classical Game Theory, like e-auctions, trading andbidding, e-voting, argumentation and dialogues, coalitionformation and social laws. The area that investigates formal andcomputational models for such complex domains is also calledSocial Software. Last but not least, the Computer Sciencecontext of much of the MAS research has provided Game Theoristswith rich models of computation, branching time models, modelsfor interaction, and fixpoint logics.This merge of interest groups is reinforced by the interesttaken up by Logicians for both Game Theory and MAS. Since thefirst LOFTs, logicians have applied their tools to formallydescribe properties in games. The recent logical formalisationsof Aumann's claim that common knowledge of rationality explainsbackward induction is only one of many examples demonstratingthe added value of logical approaches. Logic helps to specifyrich interaction environments, but also to verify complexbehaviour in MASs. Finally, Logic is arguably the mostappropriate tool to formalise reasoning, and (counterfactual)reasoning is the key to making one's decisions and understandingother's behaviour in any interaction.