On agent cooperation: the relevance of cognitive plausibility for multiagent simulation models of organizations

Human organizations and computational multiagent systems both are social systems because they are both made up of a large number of interacting parts. Since human organizations are arrangements of distributed real intelligence, any DAI model is in some sense a model of an organization. This resemblance is based on the characteristic of complex systems, that the whole is more than the sum of the parts in the important pragmatic sense that, given the properties of the parts and the laws governing their interaction, it is not a simple matter to infer the properties of the whole. Within DAI research there are two distinct lines of research: Distributed Problem Solving (DPS) and multiagent systems. The multiagent systems approach exploits the full potential of the DAI paradigm since the agents applied are more socio-realistic, in that they might compete and disagree with other agents, and will generally act in their own best interest. Increasing the autonomy of the individual agents places more emphasis on the fact that the overall system and its properties emerge from the interplay between the agents, and places more emphasis on social interaction and coordination patterns. Hence, coordination among the agents is necessary in order to achieve coherent collective behavior because of the higher autonomy levels and the bottom-up design process. The bottom-up design process entails that interaction (cooperation, coordination) strategies are not (a priori) incorporated as an integral part in the design of the system. High levels of autonomy and a bottom-up design principle are the main characteristics that make multiagent systems useful in modeling coordinated problem solving. Being models of coordinated problem solving, multiagent systems stress the distinction between the micro and macro levels and require some connection between both levels. The connection is that the rational actions of many individual agents produce macro social effects, mediated by social behavior patterns and coordination knowledge. This approach is in line with important sociological and organization theories such as the theory of structuration and the ACTS theory. The theory of structuration views the relation between the system and the actions of the individuals as circular. The system is embedded in the actions and structures the actions, which, in turn reproduce the structuring properties of the system. This bi-directional connection is in line with the principles of methodological individualism since this principle does not need to deny that there may be significant lawful connections between the properties of the “whole”, but must insist that all such properties are either definable through, or connected by, laws of coexistence with properties of the “parts”. Multiagent systems are seen as emergent systems. Emergence is a construct to explain salient high-level phenomena, which are produced as a result of the non-trivial combination of simple low-level mechanisms and properties of the constituent agents, and which are of a different nature than the properties of the components alone. However, in line with the principle of methodological individualism, it is always possible to describe these salient high-level phenomena in terms of observable lowlevel properties, since high-level properties are not directly observable. Hence, multiagent computer models are tools for computationally inferring the properties of the whole, given the properties of the parts and the laws of their interaction, in a way that is consistent with methodological individualism