Anticipation in Coordination

Abstract

The games defined in formal game theory (like e.g. the Prisoner’s Dilemma game) are widely used to model social interactions (Colman, 2003). Recently, several influential research efforts (e.g. Axelrod (1984) and Epstein and Axtell (1996)), based on Multi-Agent Simulations (MAS), have been carried out successfully in order to explain (and even try to influence) such important aspects of societies like cooperation and competition. The typical framework of such approaches consists of the use of simple agents interacting with an environment via simple rules or game playing. Although the phenomena arising in such environments are important enough to deserve detailed investigation, we have adopted a different approach here. We have been interested in cognitively plausible agents whose performance can be compared against experimental data from human participants.

The use of cognitively sophisticated agents can be regarded as a development of the opposition of standard game theory and the bounded rationality framework (Colman, 2003). In standard game theory, players are described as perfectly rational and possessing perfect information about the game including knowledge about the possible moves and payoffs, and opponents. On the other hand, the bounded rationality view on cognition states that people are almost never perfectly rational (Colman, 2003) due to limitations in perception, time, thinking, and memory. Moreover, people tend to minimize the cognitive effort while making decisions. Finally, the results of experiments involving games demonstrate that people rarely play as prescribed by the normative game theory. One such famous example is the Prisoner’s Dilemma (PD) game which will be dealt with in this Chapter.

The influence of cognitive constraints and mechanisms on decision making in the Iterated Prisoner’s Dilemma Game (IPDG) and thus on the simulations describing social interactions has been studied for instance in a series of investigations (see Hristova and Grinberg (2004) and Lalev and Grinberg (2007) and the references therein). Moreover, the use of of cognitively plausible agents can insure that the information gained by using them in simulations of complex social interactions will take into account specific cognitive mechanisms which are essential for the explanation of the phenomena observed.

One such cognitive mechanism is anticipation and its role in explaining cooperation and coordination will be the focus of our results and discussions in this Chapter. Special attention will be devoted to the use in MAS of the anticipation model proposed by Lalev and Grinberg (2007), where the role of anticipation on cooperation in IPDG has been investigated. The detailed analysis of the model features and the comparison with previous experiments with human participants demonstrated the importance of prediction for adequate description of the behavioral data on cooperation. These results were obtained in the experiments and in the theoretical frameworks by using individual playing against a tit-for-tat opponent focusing on individual decision making. Here, we want to present results which demonstrate the role of anticipation in small societies of agents. The key characteristics monitored will be cooperation and coordination as related to the essence of social interaction as discussed in other chapters of these book.