Special Issue on Multi-Agent Decision Making

Abstract

In general terms, decision making describes the process of taking decisions. In the multiagent setting, a key characteristic is that actors are self-interested agents and that decisions influence decisions of others. Classical decision making is concerned with how rational agents make optimal decisions; that is, how they maximize their expected utility. In distributed decision making complex problems are divided into smaller ones and decisions are made in a distributed manner, often at different points in time and space. Therefore, the design and the coordination of these decision subprocesses are of key interest. Another characteristic of classical distributed decision making is that often all involved decision makers are sharing—willingly or not—a global objective and that they are somewhat cooperative. For example, although a car manufacturer may have a decentralized production chain with rather independent decision makers, they all share the goal of manufacturing high-quality cars. In general, this is not the case for multi-agent decision making (MADM) which is concerned with self-interested agents pursuing their own objectives. Those objectives may be consistent but can just as well be completely contradicting; they can be publicly known or private. As a consequence, good decisions no longer only depend on the respective agent’s capabilities and the characteristics of the environment, but to a great extent on how other agents behave and on the interaction with them. In most settings these behaviors influence each other; that complicates matters further. Tools and techniques from classical decision making are no longer sufficient and new ones are needed. In the last decades, game theory, a mathematical theory to model and to reason about the behavior of rational decision makers that take into consideration the behaviour of their peers, has been successfully applied to such multi-agent settings. There are many more (classical) fields that have been shown to contribute to the understanding of this complex topic. This special issue covers some of them. MADM is a very broad area which is also reflected in the contributions of this special issue. The issue includes four technical contributions of which Bulling’s article A Survey of Multi-Agent Decision Making gives an overview of the area. In the article Strategic Argumentation in MultiAgent Systems, Thimm covers the field of strategic argumentation; that is, how agents put forward arguments in a strategically clever way—a neat combination of argumentation and game theory as well as mechanism design. Beliefs and knowledge of agents play a key role in argumentation, and in MADM in general. What if information in MAS is inconsistent? In the article Measuring Inconsistency in Multi-Agent Systems, Hunter et al. investigate how to measure inconsistency—a necessary step to resolve inconsistency—by using concepts from cooperative game theory. Finally, Bazzan considers the aspect of learning in MAS. In her article Beyond Reinforcement Learning and Local View in Multiagent Systems, she argues that multiN. Bulling (&) Department of Informatics, Clausthal University of Technology, Clausthal-Zellerfeld, Germany e-mail: bulling@in.tu-clausthal.de