MAS and Social Simulation: A Suitable Commitment

The goal of this introduction is to point out several similarities and differences between the research fields of multi-agent systems and social simulation. We show that these fields are complementary in several aspects, thus each one can benefit from results that emerge from the other. We finish the introduction by presenting and classifying the contributions in this volume. 1 Multi-Agent Systems and Social Simulation: Objective Affinities The research fields of multi-agent systems and social simulation have some interesting points in common. We characterize each of the fields next, stressing their mutual influences in the last years. 1.1 Multi-Agent Systems The field of Multi Agent Systems (MAS) is a well-established research and applied branch of AI, which has taken its impetus from the problems encountered in the implementation of tasks on distributed computational units interacting with one another and with the external environment (Distributed AI). A report on the results achieved within DAI, and a synthesis of the reasons underlying the development of the MAS field, is beyond the scope of this introduction (for a quite comprehensive picture, see O’Hare and Jennings 1996). Suffice it to say that distributed AI systems soon revealed a need for autonomy. The more autonomous the local units of the system from a central one, the more efficient the task distribution and execution, and the lower the computational load of the overall system. This discovery stimulated AI researchers and designers to turn their attention to intriguing and apparently philosophical issues, such as how to conceive of an autonomous system and how to design it. In turn, the development of autonomous systems brought about another perhaps even trickier question, i.e. how to obtain coordination and cooperation among autonomous systems executing a common task? Application-oriented solutions to these questions have often been attempted (e.g. blackboard architectures, master-slave and benevolence assumptions; see Huhns, 1987). Nevertheless, in the last decade, the conceptual question of autonomy has increasingly become a focus of AI scientists’ attention. This is shown by several scientific events: from mainly European gatherings such as the early Modelling Autonomous Agents in a Multi-Agent World (MAAMAW) events which characterized the MAS field in its early days (see, Demazeau and Mueller, 1990, Demazeau and Werner, 1991), a larger community has grown (e.g. the International Conference on Multi-Agent Systems, ICMAS; for the last one, see Demazeau, 1998). The MAS field is increasingly characterized by the study, design and implementation of societies of artificial agents. Fruitful contributions are made by other AI sub-fields. Among these, one which deserves particular attention for its recent developments and increasing popularity is the Agent field with its highly reputed scientific events (the ATAL workshops, the Autonomous Agents Conference, etc.) and journals (the Journal of Intelligent Systems). If the AI, logic-based and cognitive science approaches have contributed considerably to developments of MAS, the social sciences have exerted relatively less influence. An exception to this rule is offered by economics and game theory, which have rapidly invaded the MAS field (for a critical review, see Castelfranchi and Conte, 1998). The hegemony of these fairly specific areas of the social sciences on MAS is essentially due to the attention paid by economists and game-theorists to the study of the evolution of cooperation from local interactions among self-interested agents, also the quintessential problem of MAS scientists. The role played by economics has prevented the MAS field itself from taking advantage of the whole range of theories, models, and conceptual instruments that abound in the social sciences and that have received a great impulse thanks to the spread of computer simulation. 1.2 Social Simulation The computer simulation of social phenomena is a promising field of research at the intersection between the social, mathematical and computer sciences. The use of computer simulation in the social sciences ranges from sociology to economics, from social psychology to organization theory and political science, and from demography to anthropology and archaeology. The use of computers in some social scientific areas can be traced back to the fifties (Halpin, 1998). In its early days, and up to the seventies, computer simulation was essentially used as a powerful implementation of mathematical modelling (Troitzsch, 1997). More recently, computer simulation is more often used in its own right, "as a means of manipulating the symbols of programming languages" (Troitzsch, 1997: 41). Nowadays, the computer simulation of social phenomena and processes can be considered a well established field of research, as is witnessed by a large numbers of publications and scientific events and its own journal, the Journal of Artificial Societies and Social Simulation (for a review, see Gilbert and Troitzsch, 1999). In particular, in the last two decades, the field of computer simulation has been able to benefit from a number of increasingly accessible facilities such as the development of high-level languages; the appearance of learning algorithms and systems; etc. DAI and the MAS have provided architectures and platforms for the implementation of relatively autonomous agents. This greatly contributed to the establishing of the agent-based computer simulation, an approach which has produced a vast body of simulation research, including rebuilding the Cellular Automata tradition, thanks to new technical and theoretical instruments (for a good example of simulation studies based on Cellular Automata modelling, see Hegselmann, 1996). The agent-based approach enhanced the potentialities of computer simulation as a tool for theorizing about social scientific issues. In particular, the notion of an extended (multiple) computational agent, implementing cognitive capabilities (cf. Doran 1998), is giving encouragement to the construction and exploration of artificial societies (Gilbert and Conte, 1995; Epstein and Axtell, 1996), since it facilitates the modelling of artificial societies of autonomous intelligent agents. If the MAS field can be characterized as the study of societies of artificial autonomous agents, agent-based social simulation can be defined as the study of artificial societies of autonomous agents. One could argue that the operation result should not be affected by the operators’ order. However, the two fields are far from self-sufficient, as the following discussion will try to show. In particular, we shall argue that: 1. despite their evident affinities, the two fields in question have suffered and still suffer from an inadequate interface; 2. their cross-fertilisation would encourage research in both fields and at the same time stimulate innovative research arising at the intersection between them. 2 MAS and Social Simulation: An Unwarranted Gap MAS and social simulation differ in terms of the formalism used (logicand AI-based in the MAS domain, and mathematically based in the social simulation domain). But they also differ in other, more substantial ways. 2.1 Background Theory Although decision and game theory have had a significant influence on both, theoretical differences between the two fields abound. MAS has inherited a large share of the AI and cognitive science conceptual and theoretical endowment, which entailed (a) long experience with the design and implementation of integrated architectures, rather than elementary automata; (b) a strong emphasis on the whole agent, rather than solely on its actions; (c) careful attention paid to the process of plan-construction, not just decision-making and choice; (d) familiarity with the normalization and implementation of agents mental, as well as their behavioral states; (e) a tendency to provide the social agent with specific capacities for actions answering social requests and tasks (e.g., obligations, commitment and responsibility, etc.), rather than modelling social processes as mere emerging properties of agents’ interaction. The area of social simulation, benefited from the social sciences to a far greater extent than MAS. Among others, the following factors contributed to the field’s progress: (a) a tendency to use computer simulation to test theoretical hypotheses, rather than the computational system’s efficiency; (b) more familiarity with the interpretation of real-life social phenomena; this in turn implied (c) the production of vast bodies of data relative to artificial large-scale populations. All these features converged to consolidate the scientific methodological reputation of computer simulation, and lessen the toy-world character of its applications. Arising at the intersection of several social sciences, the field of social simulation could profit from their most recent and significant advances, such as (d) the development of the paradigm of complexity, facilitated by a close interaction with the sciences of physical and biological systems; and, in particular, (e) the development of theories, models and techniques for implementing and exploring social dynamics and evolution: social learning (cf. Macy and Flache, 1995), evolutionary game theory (cf. Weibull, 1996), cultural evolution (cf., for one example, Reynolds, 1994) and memetics (see the Journal of Memetics), etc.