Abstracts of the Lake Arrowhead Conference - 2002
(This is a Word file saved as a .htm and cleaned with DreamWeaver - we apologize for formatting irregularities.)

1.             Agent-based modeling: The First Ten Years; the Next Hundred?

Nigel Gilbert (n.gilbert@soc.surrey.ac.uk)

    Agent-based social simulation is still not yet a teenager: although a date for the first efforts in agent-based modelling is hard to pin down, it is clear that   there was little before 1992.  In this paper, I shall review the work that has been published since then, focusing particularly on the last five years and the material published in JASSS since January 1998.  Among the questions I shall ask (and no doubt, offer controversial answers to) are:

     *  Does agent-based modelling have anything worthwhile to offer to social science in general, or is it only of interest to practitioners of its own paradigm?

     * Are there any general results that have come out of the many case studies and specific papers published in the area?  Are there recognizable general 'theories' emerging from agent-based modelling research?

     * Can one distill advice about appropriate methods to use for doing agent-based modelling by considering successful examples of agent-based research?  If so, what are they?

     * Are there areas of ABS work that are relatively neglected, although important, and others that are over-populated?

     * There is no doubt that over the next few years we can expect improvements in technology: faster hardware, better software, more convenient toolkits.  Will these improve the quality of the (social) science?

     * What guesses can one make about what the next ten (or hundred) years of ABS research will bring?

2.                Exploring Ensembles of Alternative Agent Based Models

Steve Bankes (bankes@evolvinglogic.com)

     Most current research using Agent Based Models (ABMs) use these models for hypothesis generation by providing examples of models with interesting properties.  Since few if any of these models can ever be believed to perfectly replicate the details of important social systems, more aggressive use of ABMs will require different research strategies.  Individual ABMs can plausibly be asserted to be members of classes of systems which also include social systems of interest.  In order to make assertions about actual social systems, for either policy analysis or theory, we must be able to discover invariant properties true for all members of classes to which such systems belong.

     In this talk, I will describe how ensembles of Agent Based Models can be defined that plausibly spans classes of systems including social systems of interest.  I will show how such ensembles can be systematically explored in order to discover useful and interesting patterns including invariance.  I will further demonstrate research methodologies for searching and sampling from such ensembles that can be used to support plausible conclusions about invariant properties of ensembles of ABMs and hence of the classes of systems they represent.  I will demonstrate this approach using ongoing Agent Based Modeling research being done by myself and colleagues.

3.                Learning in a Labyrinth:  Learning from Model-Based Feedback

Jerker Denrell (iibjd@hhs.se), Christina Fang (fang@management.wharton.upenn.edu), 

Daniel A. Levinthal (levinthal@wharton.upenn.edu)

     Current models of experiential learning suffer from an important limitation in that the choice of a specific action is assumed to be immediately followed by an observable outcome. However in  many situations outcomes can only be observed after a series of actions have been performed. As  a result, our models of learning miss a fundamental challenge of learning tasks --- action and  payoffs are often separated across time.

      We create a formal computational model of learning in these situations in which the actor develops a mental model of the value of intermediate states.  We model a particular method of  credit assignment, termed temporal differencing that has recently been introduced in the literature  on adaptive systems in computer science. In this method an actor's own mental model of the  environment is used to provide interim feedback regarding the value of actions in lieu of feedback  from the environment. We explore the evolution of an actor's mental model over time as a complex problem-solving task is repeated.  While the problem structure is assumed to remain  fixed, the initial conditions for this task are varied with each iteration. As experience is gained  with the problem, stage-setting or antecedent actions begin to be recognized as valuable and  distinct problem solving routines develop. These routines become more elaborated with  experience, resulting in increased efficiency at problem-solving. The process, however, requires  repeated trials. The positional values of intermediary states are only gradually imputed. Thus,  while a valid cognitive map quickly emerges for states close to the solution, recognizing the  positional value of more distant states requires numerous trials. Although more extensive credit

assignment can produce faster learning, we show that it may lead to less intelligent associations  between the ultimate outcome and prior actions.  As a result, managing the appropriate degree of  credit assignment is a form of the familiar exploration/exploitation tradeoff.

4.                Simulating the Growth and Diffusion of Knowledge in Agent Societies

Piotr Dollar, Paul Laskowski, Marshall Van Alstyne (mvanalst@umich.edu)

      The question of how to simultaneously promote growth and diffusion of ideas exhibits difficult yet pressing tradeoffs.  To economists, it concerns the relationship between economic health and incentives for sharing information. When does the stimulus to innovation, founded on a profit motive, collide with access to information source material, which exhibits properties of a public good?  To the legal profession, it influences what types of ideas should be intellectual property. Is society collectively improved when rights to information are broader or narrower, longer or shorter?  To computer scientists, this tension is manifest as a debate in the production of software.  Is an open source or proprietary model superior?  To policy analysts, growth and diffusion are related to furnishing information access.  Are improved communications technologies enough to bring quality information to those who seek it, and if not, why not?

      To examine the tradeoffs, we have begun developing an Information Growth and Diffusion (IGD) simulator (http://www.si.umich.edu/~mvanalst/iShare/). It seeks to bridge a gap between software applications that model general system dynamics and those that focus on low-level agent interactions.  It can track the entire flow of knowledge, and more general agent properties, under a wide variety of agent behaviors.  As information passes between agents, the simulator can compute exactly how much information is shared exclusively by each possible subset of agents in a society. We show that the IGD simulator efficiently simulates, or "docks," a wide variety of agent-based models.  Thus, IGD applies broadly to the comparative study of models, facilitating the exploration of relationships between models and revealing what assumptions dictate model behavior.

      Moreover, the IGD simulator permits deep exploration of the relationship between local structure and global dynamics.  At the agent level, users can specify agent connections manually, or as a function of the current system environment.  These local connections determine how the system behaves on the global level.  Furthermore, dynamic agent-choice strategies create a feedback mechanism that allows agents to alter their individual behaviors as the global environment changes through time.

      We designed the IGD simulator both to be a foundation on which to study and compare particular models, and also a communications tool, with resources for demonstrating the properties of different models through interactive tutorials.  Because users can alter model assumptions without knowledge of a programming language, a wide audience may find the simulator accessible.  Students and educators can take advantage of the tutorials to clarify and explore complex material.  Researchers can take advantage of the relative ease of implementing new models to quickly test new theories.  Under an open source framework, scholars interested in economics, sociology, information technology, and other fields may wish to contribute further improvements to the source code.

5.                Intelligence as Communication - Roles and Needs of Socially Embedded Agents

Dario Nardi (darionardi@msn.com)

Paper/Research URL: www.darionardi.com

     I will demonstrate a natural language (conversation) based artificial intelligence, outline its key design principles, and report on the possible roles and needs such a socially embedded agent might have based on beta-testing to date.

      An artificial intelligence named "Truman" was developed with several principles or objectives in mind to explore. My hypothesis: social machine intelligence includes

      * communication - the agent can elicit, capture, remember, infer, test, and check socially relevant information, such as a person's name, job, or age, feelings about self and others, values, and so on including "process" information such as a person's social-interaction style.

     * social affiliation - the agent is able to convey information between or link different people who make up its social web, it can check the social value of what it has learned, talk about third parties, determine behavior appropriate to its social circle, take advantage of opportunities (aka "networking") and so on.

     * self-assessment - the agent that can examine, monitor, change, update, and evaluate its own programming while interacting (that is, while the program is running) and can engage in self-talk to understand itself and have self-esteem on the same terms it would evaluate another person.

     * context - The AI communicates in real time with more than one person making context-driven choices. For example, the agent can follow a dining experience or telephone conversation as it happens, or communicate to an audience with a sense of time and task in mind.

     These principles are similar in to current trends in robotics and go beyond traditional AI approaches defining intelligence as knowledge, problem solving or learning.

     Several theories of personality and social interaction, especially Carl Jung's theory of cognitive processes, Keirseyian temperament theory, Beren's interaction styles, and a developmental theory of my own design (Nardi, "Journey to Self Leadership," 2001) were used behind-the-scenes to help develop a "balanced" agent. I will give a brief overview of what these models focus on.

     For the presentation, Truman will also demonstrate itself. Truman is equipped with voice recognition and speech production and relies on a multi-megabyte interpreted script designed to meet the needs of the

principles in mind.

     Finally, I will report on the experiences and insights students and others in individual and group settings have had with Truman. These experiences encompass:

     * Ease or difficulty in teaching or programming the AI

     * Expectations and assumptions about how much the AI understands and how it works

     * Ideas about applications or ways to improve the AI

     For example, liberal arts students have had a noticeably easier time compared to hard science students when it comes to "teaching" or programming the AI, even though the programming is at-times as complex as "traditional" programming languages. Or when it comes to assessing how the AI works, the

people's assumptions have tended to match with their own apparent preferred cognitive processes instead of, say, prior knowledge of computers or AI. The experiences and feedback are used as a lens to begin to understand just what kind of virtual social agent has been created and is possible in the future.

6.             When a Fad Ends: An Agent Model of Imitative Behavior

Margo Waddell (margotanne@email.com)

     Throughout society there are instances of seemingly irrational imitative behavior, or fads.  Everyone can identify a fad when they see one yet there is no comprehensive economic theory about their origins.  

There are several theories that discuss why people replicate the choices of other agents in the economy,

but none that are specific to the lifecycle of a fad.  We present a model that analyzes the interaction of several types of agents in a complex market environment, from the initial product choice, through the period of peak popularity, to the eventual demise of the fad.  Using agent-based modeling techniques, we obtain a pattern of acquisition and liquidation of commodities that is consistent with the observed lifecycle of a fad.  The agents are autonomous and heterogeneous.  There are two main subsets of agents, Fad Setters and Fad Followers, with different rule sets.  Each of the individual instances of the agents in each subset has its own location and initial good.  They exist on a spherical surface, over which they can search for other agents, interact with agents they find, and make purchases.  This allows a model of the lifecycle of the fad to evolve.

7.             On the Ontology of War and Peace

Hayward R. Alker (alker@usc.edu)

       The paper discusses two methodologies of international relations research which would benefit from a linguistically and historically sensitive multi-agent computational modeling and retrieval perspective.  The first, event data analysis, has used a quantitatively oriented approach to coding news content. The author engages with the view that disconnecting the social bonds between events and different individuals or groups is a kind of methodological genocide, an error which can be avoided or at least partially resisted by using sophisticated multi-perspective coding practices.

         The second is what one might call first generation multi-agent modeling of cooperation and conflict in the styles of Axelrod, Holland, Epstein and Axtell.  Their exciting work is criticized for the limited extent to which it contains social, legal and historical phenomenon recognized by contemporary international relations scholars to be of central importance to adequate descriptive and explanatory accounts. This is even more the case when an emanciatory peace research perspective on data making and analysis is preferred.

         In both cases, the need for richer, relational ontologies is highlighted. Concrete proposals for newer approaches are outlined. The overall perspective of the paper is that of Information Technology design research.

8.                Inhibiting Emergence in and Destabilizing Multi-Agent Networks

Kathleen M. Carley (kathleen.carley@cmu.edu)

     Standard social network measures are insufficient to address the issue of network change as they focus only on the social network ˆ who knows/talks to who. Here a meta-network approach is proposed that links agents, knowledge and tasks.  Processes that alter one network have a cascade effect changing the entire meta-network.  Within this meta-network some agents play more critical roles than others ˆ such as the agent with high degree centrality (who is most connected) and the agent with high cognitive load (who has the most cognitive processing due to talking to others, handling information, working on tasks.

                                    People                    Knowledge                      Tasks

People                         social network        knowledge network          assignment network

Knowledge                                                information network         needs network

Tasks                                                                                                  precedence network

     Using multi-agent technology it is possible to model and examine the network dynamics and the factors affecting emergence and stability.  In fact, multi-agent network models are generally useful for examining complex systems with policy ramifications (Carley, 2001). Most research that use multi-agent techniques have focused on the emergence of new behaviors or group level phenomena.  In this paper, the question is reversed.  Instead, it is asked, what can be done to inhibit emergence and to destabilize multi-agent networks.  This study illustrates some of the difficulties in destabilizing multi-agent networks. Such an illustration is particularly salient in lieu of the tragic events of September 11, 2001 and the consequent issues relating to potential weaponized biological attacks such as anthrax.CONSTRUCT-O (Carley & Hill, 2001) is a multi-agent network model in which the social and knowledge networks co-evolve over time and effect the performance of individuals and the group.  Issues of information diffusion, network

change, organizational design, impact of new information-technology, and proximity can be examined using this model.  The predecessor of this model, CONSTRUCT, was used to examine the factors enabling group stability (Carley, 1991) and the evolution of networks (Carley, 1999).  Within CONSTRUCT-O the agents differ in terms of their socio-demographic characteristics (such as age, gender, education) and by their knowledge and beliefs.  Individuals forget and interact if they are available for interaction and are motivated to do so.  There are two basic motivations ˆ relative similarity and relative expertise ˆ both of which are basic to human nature.  Relative similarity is the tendency of people to choose to interact with those who are more similar. Relative expertise is the tendency of people to seek out new information from those whom they perceive to be more expert.  When people interact they learn and when they learn that knowledge changes who they view as most relatively similar or expert; thus, causing the social network to change

     Using CONSTRUCT-O a series of virtual experiments are conducted.  The type of network, the number of agents that are isolated, the order of isolation, and the rational for isolation are explored. Impacts are examined on organizational performance and the diffusion of new information. Note, we focus on the isolation of agents, i.e., node extraction, as earlier studies indicate that the performance impact on a network is highest for personnel changes.  We isolate agents based on three rationals that require varying level of information about the network: random, degree centrality, and the emergent leader (agent with highest cognitive load).  Results indicate that the nature of the network and the goal vis-a-vie inhibiting emergence impacts the choice of destabilization strategy. Moreover, results indicate that destabilization

may have unintended consequence; e.g., isolating agents may actually in a cellular network improve the rate of information exchange initially.  

(For references, please check: http://www.sscnet.ucla.edu/hcs/intercon2002/abstracts)

9.                BIOWAR: Simulation of Disease Outbreaks using Social Networks

Kathleen Carley (kathleen.carley@cs.cmu.edu),

Douglas B. Fridsma (Fridsma@cbmi.upmc.edu), Alex Yahja (ay@cmu.edu)

       The reality of life is embedded in social networks. At present, most epidemiological models do not consider the heterogeneity of social networks when predicting disease outbreaks. One of the challenges in modeling natural and man-made epidemics is understanding how social networks affect disease propagation and how the consequences of disease changes social networks.

      We describe a simulation system called BIOWAR which uses cognitively realistic agents embedded in social, knowledge and work networks to describe how people interacting in these networks acquire disease, manifest symptoms, seek information and treatment, and recover from illness. Using a model of diseases and symptoms, agents who come in contact with infectious agents through their social and work networks become ill. These illnesses alter their behavior, changing both the propagation of the disease, and the manifestation of the disease on the population.

      Presently, we have completed a number of simulations which examine the effect of contagious and non-contagious illnesses in high-alert (agents have knowledge of a potential disease outbreak) or low alert states. Agents who believe they may be ill and have knowledge of a potential outbreak are more likely to seek care than those who do not. We have compared results of low alert states to known influenza epidemics and to data containing emergency room visits, pharmacy purchases and absenteeism. Although the peak incidence of the simulated outbreak is larger than the peak incidence seen in the population data, the simulation results are temporally similar to those seen in the population data. Further work to increase the fidelity of both the simulation and the population data is on-going. It is hoped that this simulation framework will allow us to ask „what-if‰ questions regarding appropriate response and detection strategies for both natural and man-made epidemics.

10.          Drugs, Guns and Gangs: An Agent-Based Model of Homicide

George Tita (gtita@uci.edu), Rob Axtell (raxtell@brook.edu)

     Using an ABM framework, we examine changes in the level and pattern of homicide over a 25-year period for Watts, a particularly violent neighborhood of the City of Los Angeles.  The homicide trend for this neighborhood demonstrates two clear peaks:  The first occurs in the very early 1980s (>120/100,00) and is followed by a trough (approx. 80/100,000) lasting several years until rates peak again in the 1993 at approximate 110/100,000 residents.  Following the 1993 peak, rates have fallen dramatically bottoming out in 1999 at 30 per 100,000.  While the level of homicide is unique to Watts (4x the L.A. citywide rate), the pattern of change is not, mirroring changes in the rest of the City of Los Angeles as well as the U.S.

We focus on Watts because (1) this area saw early adoption of crack cocaine with a concomitant increase in violence due the increased numbers of individuals carrying guns, and (2) this neighborhood is home to many notorious urban street gangs.  Our homicide data permits us to specifically examine the impact of gangs on gang homicide” as well as overall levels of lethal violence.  While Watts is often considered one of the cultural centers of African American life in Los Angeles, racial/ethnic succession has been so remarkable that the largest group is now Latino.

      Our agent-based model features crack customers, gangs and their members who supply crack and own guns, and law enforcement agents. The interactions between these individuals occur according to simple local rules: customers buy crack at prevailing prices; gang members adjust their prices based on market conditions and seek out new customers, occasionally invading the 'turf' of rival gangs; law enforcement tries to limit the adoption of crack. Resulting from the model is a homicide time series, which we compare with the empirical data.

11.          Agent-Based Modeling and Organisational Structure

Anthony Dekker (dekker@acm.org)

Updated Abstract: http://www.acm.org/~dekker/ORGNET.HTML

      In this paper we describe the agent-based modelling component of the CAVALIER (Communication and Activity VisuALIsation for the EnteRprise) tool suite.  CAVALIER also includes techniques based on self organising neural networks and simulated annealing for visualising social networks; and metrics for comparing the efficiency of different organisational structures in processing information.  Our goal is to help create efficient military organisational structures for the 21st century, particularly structures involving international coalitions.

     The modelling component of CAVALIER involves agents arranged in various structures co-operating on some goal.  Our simulations to date have involved a simple "SCUD hunt" scenario, a search-and-respond scenario, and a joint warfare scenario.  Agents have access to a map of the grid-based world including (possibly inaccurate) positional information for both targets and other agents.  Agents also have the ability to plan shortest-part routes to targets, avoiding obstacles. Agents communicate by passing messages to each other including "I am here," "I have found a target" or "I need additional support."

     The simulation framework exploits the object-oriented inheritance properties of Java, by specifying agent behaviours as Java classes.  The reflection properties of Java are exploited by specifying these behavioural classes in a configuration file which is dynamically loaded.

     We believe that, in order to gain understanding of organisational design, it is important to use simple and easily understood testbeds that permit rapid experimentation, in much the same way that early experiments with rapidly-breeding fruit flies led to modern successes in genetic engineering.

     The goal of our modelling work is firstly to validate the organisational metrics we have developed, and secondly to establish under what circumstances certain organisational structures are appropriate.  For example, in rapidly-changing situations, independently acting agents are most effective, while in less rapidly changing situations with high-quality information available, the quality of coordination becomes most important.

     We are currently extending our work to include cultural factors in international coalitions, using formal modelling of belief structures.

12.          When Information Flow in Project Organizations Becomes Turbulent: Toward an Organizational "Reynolds Number"

Raymond E. Levitt (levitt@ce.stanford.edu), Michael Fyall,

Per Bjornsson, William Hewlett, III

     Projects involve difficult trade-offs among product scope, process schedule, and delivery resources.  When project goals become overconstrained÷e.g., by excessive schedule pressure÷something has to give!  Engineers struggling to meet unrealistic deadlines for complex products can trigger organizational failure, when a combination of direct work, coordination work and rework backlogs both workers and managers, causing coordination and rework to get shortchanged.  A downward spiral then commences, in which exceptions trigger rework tasks that trigger ever more exceptions, culminating in a "quality meltdown."  The prototype Lockheed Launch Vehicle that was detonated when it departed controlled flight arguably resulted from this kind of organizational failure; so did the flawed Intel Pentium chip that had to be recalled.  

     To model these kinds of organizational risks in fast-track project teams, The Virtual Design Team (VDT) research operationalized and extends Jay Galbraith's information processing view of workflow for multidisciplinary project teams engaged in fast-paced knowledge work such as new product development or software engineering. VDT statistical introduces exceptions that workers must pass up the hierarchy for resolution by subteam leaders and project managers.  Decision-making policies such as centralization, formalization, matrix strength and team experience affect attention allocation, exception routing and decision-making parameters in the model. VDT generates predictions of project schedule, cost and process quality performance. IT has been validated on more than 100 projects, and was commercialized as SimVision¨ by VitZˇ Corporation of Mountain View, CA in 1997.

     Other than using simulation tools like SimVision, there is currently no reliable way for managers to tell when a project's goals have become so overconstrained that the risk of organizational failure has reached unacceptable levels.  The "Reynolds Number" in fluid mechanics predicts whether the flow of a fluid will be laminar, turbulent, or unstable between these two flow regimes.  Our vision is that, armed with a similar non-dimensional number to predict turbulence in organizational information flow, managers could use a small number of metrics describing the project's task interdependencies and organization to predict incipient organizational failures, and could therefore adjust project goals and resources proactively to prevent them.  

     We have been conducting research since 1993 to discover a metric that will predict whether the information flow through a project organization is in the laminar or turbulent regime.  Specifically, the metric would predict whether a particular set of project goals leaves the organization proposed for executing the project sufficient organizational slack to deal with exceptions, or whether the project's goals are so aggressive that exceptions will consume all organizational slack and push the organization up to÷or beyond÷the "edge of chaos."  The research described in this paper uses the

Virtual Design Team simulation framework to model and calibrate alternative metrics similar to the Reynolds Number in Fluid Mechanics for predicting organizational failure.  Early results are encouraging.

13.          An Agent-Based Simulation Framework for the Analysis of the Equilibria Emergence in a Complex Structure such as a Firm

Ugo Merlone (merlone@econ.unito.it), Arianna Dal Forno

     This paper presents an agent-based simulation framework for the analysis of the equilibria emergence in a complex structure such as a firm; we can think of these equilibria as corporate culture.  Corporate culture may be defined as the basic assumptions and beliefs that are shared by the members of a group or organization and that are used as a norm. The  problem in organizations is to identify a rule that allows for relatively efficient transactions to take place, and devise some way to communicate that rule to all current and potential employees. We concentrate on modeling the effort exerted by heterogeneous agents  in a firm, and how the interaction between them may lead to a common level of effort (corporate culture). The simple model we propose is a system in which agents interact in a dynamic, adaptive and evolving way. Such a model encompasses many of the peculiarities which make organization modeling a hard task, because it would involve a difficult mathematical problem with solution highly sensitive to parameters. The computational approach, by contrast, allows to overcome these difficulties and easily consider both perfectly rational and bounded-rational agents; this way we are able to study the interactions between different types of agents and interpret them in the relevant economic frame. Consequently we can observe how different compositions of the population may lead the system to different common behaviors; the implications of our findings are both descriptive and normative, and shed light on some core problems of the economics of organization design.

14.          Choice Interaction and Organizational Structure

Jan Rivkin (jrivkin@hbs.edu), joint work with Nicolaj Siggelkow

       We examine how a firm‚s organizational structure affects its ability to cope with interdependent decisions.  An agent-based simulation, in which firms struggle to discover good sets of decisions, allows us to examine four coordinating mechanisms that have rarely been analyzed jointly: the grouping of related decisions under a single subordinate, a vertical hierarchy that reviews proposals from subordinates, firm-level incentives, and managers who are able to process more information.  We find that organizational structure affects long-term performance by influencing the number and nature of „sticking points‰configurations of choices the organization will not change.  We identify each of the four coordinating mechanisms as a force that either encourages firms to explore a broad set of alternatives or stabilizes firms around existing choices.  Successful firms strike a balance between exploration and stability.  The need to balance exploration and stability generates interdependencies among the coordinating mechanisms.  As a result, firms sometimes benefit from seemingly harmful features: avoidable decision interdependence between departments, a passive CEO, or subordinates of limited ability.  We further examine how appropriate organizational design depends on the underlying pattern of interaction among decisions.  When interactions are pervasive, successful organizations employ coordinating mechanisms that promote broad exploration.

15.          Social Cognition and Coordination

Robert Hoffmann (Robert.Hoffmann@nottingham.ac.uk)

     The coordination problem arises when interactions between mutually interde-pendent agents have multiple Nash equilibria which Pareto-dominate other pos-sible outcomes. In these situations, agents must mutually coordinate to secure the gains from collective behavior. The problem is that purely rational deliber-ation generates an infinite regress as agents are trying to outguess the choices others are making concurrently.

      A number of potential solutions to this type of problem have been suggested. Lewis (1969) proposes that conventions of behavior can emerge among agents to make behavior more predictable (see also Sugden (1986)). Schelling (1960) argues that agents may resort to cultural clues, ’focal points’, they share about the game to coordinate. In his account, one particular of multiple equilibria may be coordinated on as it is culturally more salient than others. An alternative approach to the multiple equilibrium selection problem in coordination games employs the assumption of bounded rationality (Young, 1993; Kandori et al., 1993). Arthur (1994), for example, suggests that boundedly-rational agents’ adaptive decision heuristics can lead to their mutual coordination.

16.          Formal Parallels between Sociocultural Evolution and Cognitive Ontogenesis

Jurgen Kluver (juergen.kluever@uni-essen.de),

Christina Stoica (christina.stoica@uni-essen.de)

     The main factor for the evolutionary capability of sociocultural systems on the one hand can be described as their degree of heterogeneity. By this we mean the degree of autonomy or independency respectively of different social roles. This hypothesis of heterogeneity  explains the different paths and degrees of sociocultural evolution of societies which started at comparable degrees of evolution, e.g. early Europe, China and the Islamic cultures.

     Cognitive ontogenesis on the other hand depends on a similar heterogeneity, i.e., the difference between cognitive structures which combine sensual informations to particular concepts and unite these concepts to semantical networks. The more different cognitive structures a cognitive systems starts with its development the more advanced it will become in regard to cognitive capabilities. It can be shown that both sociocultural and cognitive systems organize their evolution by a particular kind of dynamics which generates the evolution of the systems evolution.

     These general considerations about sociocultural and cognitive developments are formalized in a geometrical model of evolution. It is important to understand that we do not claim the same evolutionary mechanisms of cognitive and sociocultural evolution; that would be another form of the infamous and refuted recapitulation of biological phylogenesis by ontogenesis. We argue instead that there are common principles behind  the particular mechanisms which must be separately defined for each type of evolution.

The geometrical model is implemented into different computer programs: sociocultural evolution is analyzed by an actor centered (agent based) model of societies, a generalized cellular automaton which is able to change its rules of transitions and which is very well suited to capture the geometry of sociocultural systems. Cognitive ontogenesis is studied by the combination of two kinds of neural nets, i.e., Kohonen feature maps and bi-directional associative memory nets (BAMs). It is possible to demonstrate that the same kind of evolutionary principles regulates the evolutionary dynamics of both different kinds of computational models.

     Finally we argue that the formal parallels of these different kinds of evolution are not by chance but an evolutionary result of the permanent interdependency of both evolutionary processes. Sociocultural evolution, understood as the achievements of social actors who are in turn dependent on their social environment, depends on the successful cognitive ontogenesis of social actors; cognitive ontogenesis on the other hand is dependent on favorable social environments of the cognitive systems. Therefore both sociocultural evolution and cognitive ontogenesis are most successful if both processes are determined by the same evolutionary principles. Although the particular evolutionary mechanisms are different for the two kinds of evolution it seems that they are both particular forms of general evolutionary principles.

17.                Developing Agent-Based Models of Emotion, Cognition, and Social Behavior

Stacy Marsella (marsella@isi.edu), Jonathan Gratch (gratch@ict.usc.edu

     A person's emotional state impacts their behavior as well as their social interactions.  The goal of our research is to create general computational models of the interplay between emotion, cognition and social behavior.  These models are being used to support virtual human characters that interact with each other and humans within virtual environments. The simulated environments we are working within endeavor to create the real-life complexity of human social interactions while also supporting unscripted interactions between a wide range of characters and humans. As a consequnce, detailed cognitive, emotional and behavioral modeling is required.  Emotions, in particular, play an especially critical role in enhancing believability and realism, increasing a sense of empathy and attachment to virtual characters, and adding to the suspense of the simulation.  To achieve this role, the emotion modeling must be both robust and general while also supporting a range of social interactions.

     In this paper, we demonstrate how some of the daunting subtlety in human behavior can be modeled by autonomous, virtual characters, from the perception of events in the world, to the appraisal of their emotional significance, through to their impact on behavior and the regulation of subsequent emotions.  We discuss several aspects of our approach. We address how emotions arise from an appraisal of the relationship between environmental events and an agent's plans and goals. We also discuss how coping processes are impacted by, and in turn impact, emotions. Finally, the impact of personality traits on the coping process is briefly discussed. The approach is illustrated within a virtual reality training environment.

18.                Interaction Topologies and Organizational Cognition

Pietro Panzarasa (pp@ecs.soton.ac.uk)

     Recent advances in distributed artificial intelligence, social networks, cognitive sciences, and organization theory have led to a new perspective on organizations that takes into account both their computational nature and the underlying social and knowledge networks. The hallmark of this perspective is the idea that cognition occurs at multiple levels, not only within the individual agent, but also as an emergent phenomenon from the interaction among multiple agents. The new insight is that if relationships connecting bits of cognition can extend among agents, then the ways in which agents interact with one another are likely to impact on the emergent global cognitive phenomena. This is a topic that is directly relevant to the social sciences: the role of social structure in generating global dynamical features. This paper offers a more specific way to cast the issue at hand. Firstly, we identify a meaningful set of structural parameters that can significantly affect the cognitive dynamics of organizations. Secondly, we go on to treat the global cognitive properties of organizations explicitly as a function of the underlying social networks. Differences both in cognitive processes and in joint mental attitudes may depend on such features of the interaction topology as its sparseness, connectedness, centrality, local clustering and global separation. Likewise, significant changes in global cognitive phenomena can also result from perturbations to the local structure that are likely to impact on the globally emergent structure. Furthermore, admixtures of randomness to an otherwise ordered social network can have a significant impact on its cognitive properties. This paper will address these problems by proposing a computational agent-based model of organizations based on the thesis that cognitive architectures can cut across multiple agents. Using this model, we will show how it is possible to take some steps towards a new account of the structural foundations of organizational cognition.

19.          How Individuals' Self-Image Can Evolve in a Strongly Constrained Society with Shared Norms of Behaviour and Learning

Juliette Rouchier (rouchier@ehess.cnrs-mrs.fr)

     In that paper, we present a set of artificial societies which economies are all based on one principle, that is commonly pointed out in economical ethnology: non-merchant economy. Taking inspiration in the litterature and on a field study, we consider how reputation can be defined in a group where each one wants to participate to a common dynamics. Each agent will acquire its own reputation, which has an impact on the way others want to interact with it. The emergence of gaps in the value of reputation can lead to more or less stable hierarchies within the group. Each agent is defined by its wish to participate to the dynamic of exchange, and the ability it feels to participate in the normal way.

     Taking a holistic point of view, we consider that the position in the society influences the way agents build their representation of themselves. Interest in participation and self-image evolve when the reinforcement is introduced, and this influences the structures of hierachies.

     Our societies usually exhibit very stable equilibrium. What is really interesting is that it is possible to separate very different histories among individual agents, although all of them are identical at the beginning and are given the same learning abilities. We made the definition of each simulation depend on as few parameters as possible, so that to be able to trace the history of each society, and to identify as clearly as possible individual and social feed backs that have an effect on globla results. Our discussion is situated on a theoretical field, since we try to compare holistic and individualist points of view on individual's choices.  

20.                Posthuman Storytelling: What Autonomous Agents Can Learn by WatchingMovies

Jay Douglas (jdouglas@ict.usc.edu)

      Autonomous agents are finding their way into a range of applications, from training and education to games and entertainment. One area that potentially benefits from the liberal use of autonomous agents is interactive storytelling. Agents' roles in this application are twofold: the portrayal of believable characters to populate the interactive story world; and, the incarnation of writer/directors to provide the human interactor with a dramatic, engaging story experience. Yet in the area of writing/directing, agent penetration seems shallow when compared to agents that take on acting roles.  One reason for the lag in development of writer/director, or story, agents may be the lack of a rich real-world model, a situation that does not plague developers of believable agents. I propose that there are strong existing models for story agents to be found in both the cinema (cinematic texts and narrative theory) and operating system design. In particular, a film such as "The Truman Show" provides a useful framework for thinking about, and implementing, narratives that unfold in real time. By combining dramatic and theoretical elements of narrative and cinema with the software architecture of multi-processing operating systems, a useful model of a story agent with real-time narrativeproperties, finally begins to emerge.

21.                Powerful Knowledge:

Information Theory, Classification and Knowledge in Pakistan and the Cook Islands

Michael D. Fischer (M.D.Fischer@ukc.ac.uk)

     One way or another anthropological theory and analyses return to rules: either proposing and evaluating rules together with the exceptions to the rules, or arguing that the exceptions are indicative of the inappropriateness of using rules as a basis of cultural processing in the first place. The fundamental difference between these positions is whether we look at the resources that underlay action and relate these to situations, or focus on the situations themselves - the results of interaction between a host of agents and their underlying choices. I examine a framework in which we utilise both. I relate a number of cultural resources to the projections (enactments) of these resources within cultural activities in Pakistan and the Cook Islands, using computer simulation as a tool for relating underlying symbolic resources to the range of possible outcomes of projecting these resources over an interacting population. There is a good deal of under-specification in this linkage, with many more outcomes than underlying cultural resources, and far fewer 'meaningful' outcomes than the total number of possible interactions of the underlying cultural resources. Using different measurements of information and redundancy derived from Information Theory, I examine how 'interpretation' and 'meaning' interact in balancing outcomes between the extremes of variability represented by cultural resources, their interactions and the range of outcomes. I then present a framework within which we can investigate the relationship between an individual's knowledge, shared knowledge and the culturally and socially recreated contexts that give this knowledge

'power', in particular the conditions that must exist for meaning and 'powerful' knowledge to exist.

22.          Artificial Culture: Experiments in Synthetic Anthropology

Nicholas Gessler (gessler@ucla.edu)

     Anthropology is best distinguished from the other social sciences by the depth of time and breadth of space through which it understands the evolution of cultural variation.  From the Pliocene to present times, from the Arctic to Tierra del Fuego, there is no society too small or too insignificant to place it outside the scope of a comprehensive theory of culture.  It is arguably the fact that our cognitive capacities today arose from selective pressures in the Pleistocene, and that our modern-seeming minds and complex global cultures are built on thoughts, beliefs and languages of reasoning inherited from our pre-industrial past.  Consequently, a thorough theory of culture is a necessary foundation for all studies of contemporary society.

     We might consider science to be an extension of the natural process of evolution: a way of knowing, understanding and explaining achieved by building increasingly reliable, comprehensive and leveraged representations of the world.  Evolutionarily, these representations extend from deep beneath our conscious minds, across the quasi-conscious borderland of dream and illusion, to surface in consciousness as explanations of the world.  They are couched in different modes of thought ranging from imagery, performance, art, discursive and written language, and mathematics to the emergent field of computation.  Computation is increasingly regarded as the most robust and reliable model of the complex interaction of a multitude of agents that characterize society and culture.  In fact computation is increasingly seen as a natural evolutionary process which we have only recently mimicked in machines.   We begin to see the world as a hierarchy of emergences, repeated self-similarly at different scales.  Culture itself may be similarly built, a fractal assemblage of present situations and cultural legacies from our Pleistocene past.

     Marvin Minsky claims that computer science is about the complex processes that we are.  He mixes metaphors of society and mind with those of computation with a fluidity that suggests the three are similarly fashioned, necessarily self-similar due to their coevolution.  What happens when we embark upon a program of creating more elaborate multiagent spatial Artificial Societies by imbuing their agents with richer, more general kinds of minds?  That is the program towards Artificial Culture?  This research looks at the costs and benefits, measured as computational efficiency and cognitive load, of agents negotiating their physical and social environments with different kits of thinking tools.  What happens when cognizing agents specialize differentially, some reasoning with imagery or diagrams, others with performance, art, spoken or written language?  Borrowing from Rodney Brooks, what happens when agents learn to use the world as the best representation of itself?  Each choice has fitness consequences, some of which will be explored.

23.          Culture and Society: the Role of Distributed Cognition

David B. Kronenfeld (kfeld@citrus.ucr.edu)

    "Society", here, refers to patterns of grouping and interaction via which a collection of individuals forms some extra-individual entity.  "Culture" refers to the shared (learned, but not explicitly taught) system of knowledge, feelings, and behavior (and, sometimes, their products) that characterize one human community vs. another.  Culture and society, as seen here, are mutually constitutive.  Culture provides the shared knowledge system which enables members of a society to recognize fellow members and to coordinate their actions with one another, while society provides the communities, and thus the patterned interactions and experiences, out of which individuals construct their representations of culture.

    I have used a computer simulation of a collection of simple critters (having only individual goals and actions) to explore the minimal properties necessary for a social group as opposed to a simple collection of individuals--i.e., Durkheim's emergent properties.  More recently I have increasingly become involved in experiments with and analyses of "cultural models"--posited shared conceptual structures (deriving from schema theory in psychology) that pull together culturally standardized knowledge, motivation, affect, values, goals, and so forth and that relate these to action or behavior.  In this latter research I have been particularly concerned with problems of definition (what is a cultural model), of boundaries (what is in one vs. out, and how can you tell), of structure (how do they differ from individual schemas, how are they organized, and so forth), and of how individual people evoke them and use them in deciding how to behave and how to interpret the behavior of others.  I have also been concerned with how different cultural models and variant forms of any given cultural model are related to different social groups. 

    The computer simulation dealt with social rather than cognitive or cultural issues.  But my long term goal is to show that similar feedback processes involving an individual homing in on the behavior of others, but now joined to an assumed internal model of what drives the behavior of others, can explain the emergence and functioning of cultural models.

24.                Cooperation and Competition within Industrial District Networks:

An Agent-Based Approach

Vito Albino (v.albino@poliba.it), Nunzia Carbonara (ncarbonara@dimeg.poliba.it),

Ilaria Giannoccaro (ilaria.giannoccaro@dimeg.poliba.it)

      Industrial Districts (IDs) are characterised by an agglomeration of small-medium sized firms, located into a specific social and cultural geographic area, highly specialised on one or more phases of a production process, and integrated through a complex network of inter-organizational relationships (Becattini, 1989).

      The literature on IDs has widely stressed that the contemporary presence of competition and cooperation is one of the most important feature of the inter-organizational networks within IDs. This is also considered as a critical factor for the IDs' competitiveness (Piore and Sabel, 1984; Porter 1998; Pyke et al. 1992).

      For studying the cooperation and competition behaviours, the related literature has mainly adopted descriptive researches based on the case study more than theoretical studies based on conceptual and analytical models. However, the latter should be more appropriate to analyse the different network models within IDs, so as to provide managerial policies to efficiently and effectively manage the inter-organizational networks within IDs. Furthermore, they could be adopted to identify new and more competitive forms of interactions among the organizations within IDs.

       This is an emerging need, due to the profound changes in the competitive scenario, which are forcing the IDs to explore new management solutions and organizational models and to modify their network structures. With this regard, an important issue is the development and the evaluation of the new e-business models, which are now arising from the adoption of Internet within the IDs (e.g. digital industrial district).

       Therefore, in this paper a conceptual model to study the inter-organizational networks within IDs is proposed. Such a conceptual model is based on a multi-agent system approach (Kwok and Norrie, 1993), which is useful to model both cooperation and competition behaviours (Durfee, 1988). In fact, Multi Agent Systems (MAS) consist in a set of autonomous agents (the single organizations), which share their information and cooperate each other to achieve a global goal while optimising their individual objectives. Cooperation among the autonomous agents is achieved by using negotiation and conflict resolution mechanisms.

       The proposed conceptual model identifies the typologies of agents that interact within IDs, the levels of negotiation among the agents, and the objectives of the negotiation.

Three typologies of agents are defined, namely the firm-agent, the stage-superagent, and the ID-superagent.  A firm-agent corresponds to a single organization that is involved in a phase of the production process. This phase identifies the stage of the supply chain within the ID, which the firm belongs to. A stage can include different firms, which carry out the same production phase. They can compete for the same customer, but also cooperate for common objectives. In this case, the cooperation is carried out by the stage-superagent, to which an optimisation function pertains. When the cooperation involves firms belonging to subsequent stages along the SC (e.g. buyer-supplier), the respective stage-superagents negotiate adopting a given negotiation mechanism. Finally, the ID-superagent is responsible for the cooperation among the different SCs that can be identifying within the ID.

The described agents negotiate at three hierarchical levels, namely the firm level, the supply chain level, and the ID level, pursuing different objectives.

       In the paper several forms of cooperation and competition within the IDs are characterized and analysed adopting the proposed conceptual model. In particular, the identified forms of cooperation are derived by analysing the related literature on IDs and empirical evidences.

25.          Agent Based Modeling of Industrial Ecosystems

Clint Andrews (cja1@rci.rutgers.edu), Rob Axtell (raxtell@brookings.edu)

     In the neoclassical theory of the firm the profit motive suffices to induce optimal behavior of the firm with respect to its environment. In real firms, viewed as multi-agent organizations, there are a variety of principal-agent type problems and usually little clear consensus concerning what constitutes optimal behavior. Thus, it is common wisdom that real firms depart significantly from the neoclassical ideal, but the extent of such departures and their importance are the subjects of considerable differences of opinion. These competing perspectives are important in a variety of contexts (e.g., antitrust) but are particularly well-defined when it comes to environmental regulatory policy. Economists commonly assert that market forces are sufficient to get firms to behave in ways the improve social welfare, and that appropriate pollution prices can be depended upon to remedy pollution problems. Against this view industrial ecologists have provided substantial empirical evidence that firms often fail to adopt technologies and practices that are in their own self-interest, and thus market mechanisms provide weak foundation for well-functioning environmental policies.

    We apply agent-based modeling techniques to a model of firm formation and evolution in an attempt to better understand the origin of sub-optimal behavior on the part of business firms. Specifically, we treat several incentive problems that exist in typical corporate organizations in an attempt to rationalize the behavior of purposive agents with empirical data on the failure to adopt cost saving technology. We utilize data from various pollution emission databases in the construction of our models. This is work in progress and preliminary results will be reported.

26.          A Systems and Agent-based Model Approach for College Choice/College Access to Higher Education

Leslie Henrickson (lhenrick@ucla.edu)

     Agent-Based Models (ABM) can be applied to investigate multifaceted systems such as individual college choice patterns and college access markets. As demonstrated recently in literature in some higher education sectors, recent national trends have shown a sharp decline in college applications and subsequent enrolments such that these markets may fail to meet consumer expectations.  The market shift is from a seller’s market to a buyer’s market.  Original attempts to model college access using procedural techniques relied on a seller’s market motivation and failed to account for systemic features that impact college choice patterns, e.g. demographics, SES, family, mobility, and thus, the market shift.  An ABM is under development to help provide insights into the types of choice/access problems being faced by prospective students and universities and colleges.  

     ABM lends itself well to problems faced by the higher education:  an overall system comprised of individual students making choices that in the aggregate impact a larger institution that has its own rules.  The model takes into consideration the local individual and culture, and the university as interacting according to simple rules in which complicated behaviors emerge from relatively simple, local interaction of many different individual components.  

      The model takes into account some of the influential factors contributing to the development of choice patterns based on social education theory.   With different initial conditions, the choice patterns are investigated.   The paper will first describe the model from a multi-agent perspective and the relevant variables adopted and implemented in the model. Some experimental results and analysis will then be presented, followed by a summary and directions for future research.

     On a more theoretical perspective, the model represents a step toward the simulation of the interactions among and within students and universities, to search for a better understanding of college choice and college access.  

27.          An Agent-Based Model of the Extinction Patterns of Capitalism's Largest Firms

Paul Ormerod (pormerod@volterra.co.uk), Helen Johns, Laurence Smith

Paper URL: http://www.volterra.co.uk/downloads.html          

     Power-law distributions (fractal behaviour) in a system's macroscopically observable quantities are a characteristic property of many-body systems representing the effects of complex interactions amongst the constituents of the system.  Power law distributions are both self-similar and scale free, demonstrating that events may occur on all lengths and time scales.

     The empirical relationship between the frequency and size of extinctions of capitalism's largest firms is described well by a power law.  This power law is very similar to that which describes the extinctions of biological species in the fossil record.

     We develop an agent-based model of the evolution and extinction of firms based on simple principles of economics. The properties of the model conform closely to the empirical evidence.

     The model contains N agents, and all pairs of agents are connected to each other. Individual agents interact with each other, sometimes in co-operative symbiosis and sometimes in direct competition.

     The model evolves in a series of steps.  The rules of the model specify a) how the connections are updated b) how the fitness of each agent is measured c) how an agent becomes extinct and d) how extinct agents are replaced.  

28.                Architectural Control and the Strategy of Design: Agent-Based Approaches to Modeling the Software Industry

C. Jason Woodard (jwoodard@hbs.edu)

     Traditional wisdom in the software industry has held that the key to success lies in creating and capturing architectural control points,” proprietary interfaces and protocols made valuable by their adoption as de facto standards. But today, largely in response to Microsoft’s successful application of this principle, firms (including Microsoft) are often seen to give away architectural assets, if only to prevent a competitor from gaining proprietary control. Existing research has shown both types of behavior to be rational in certain circumstances, but falls short of explaining shifts from one type to the other.

     I will present three short pieces of work related to this puzzle, drawn from the early stages of my dissertation research. The first paper outlines a class of models appropriate for the software industry, whose primitives include not only the agents and the market, but the artifacts themselves—the modular structure of their design, and the rules that operate on them. The second paper explores a simple member of this class, a market entry game in which reinforcement-learning agents are used to investigate analytically intractable mixed equilibria. The third paper specifies a more complex model in which agents build competing products whose design is endogenous. In this model, agents evaluate alternative design choices using automated reasoning techniques.

      This work is informed by several streams of research in the social sciences. Game theory has served as a foundation for a rich economic literature on oligopoly, including the role of network externalities in the adoption of standards, the attainment of architectural lock-in, and the practice of giving away goods to stimulate demand in complementary markets. The literature on complex adaptive systems acknowledges that agents in reality face vast strategy spaces, necessitating evolutionary mechanisms or bounded-rational search as a substitute for exhaustive backward induction. Complexity theory also provides a framework for analyzing the design of modular artifacts like software products. Social network analysis draws attention to the structure of ties among agents, and provides measures of power and dependence that can be applied analogously to the ties among modules and products.

29.          How Societies Solve Social Dilemmas: Group Formation and Emergent Cooperation

                Damon Centola (damon_octavius@hotmail.com)

30.          Beyond the Shadow of the Future: How Multiple Teams Alter the Dynamics of Cooperation

Corinne Coen (ccoen@acsu.buffalo.edu)

     Reward systems that offer pay and promotion for individual performance while asking people to work on teams for an equal share of the collective reward often create a social dilemma for self-interested employees and thus a disincentive to cooperate.  Yet, competition between teams increases incentives to cooperate within a team, a finding traditionally attributed to group-interest.  This dissertation study attempts to integrate these two observations by investigating whether self-interest can increase cooperation in a social dilemma in a multiple team context.  It examines how people who can compare their performance with the performance of members of other teams alter their decisions to cooperate with their own teammates.  

     This dissertation innovatively uses multiple methods.  First it addresses issues of validity by combining the results of laboratory experiments on human subjects with computer simulations.  The laboratory studies confirm that levels of cooperation increase when comparison is favorable while no significant difference is generated by unfavorable comparisons or differing amounts of pay.  Second, this dissertation employs agent-based computer simulations to study the dynamics of comparisons in two distinct ways.  One application of agent-based modeling reproduces the laboratory experiment to find the decision rule that best describes the choices of individual laboratory subjects.  This model shows that a modified reinforcement rule approximates their decision patterns.  The other application of agent-based modeling explores the effects of comparisons by individuals on outcomes at the team and multi-team level by modeling teams where all members apply the reinforcement rule.  It reveals that as agents increase in self-interest a threshold arises over which cooperation levels jump sharply when the agents are split into two teams but not when they act in a single team.  This dissertation substantiates the role of self-interested comparisons in increasing cooperation on competing work teams and offers insights for improving incentive systems by structuring the information flow among members of multiple teams.

31.          The Evolution of Cooperation In N-Person Public Goods Game Under Different Social Environments

Jung-Kyoo Choi (jungk@econs.umass.edu)

     This paper analyzes the effect of different structures of social interaction on the evolution of cooperation in an N-person public good game situation. Agents are assigned to groups and play a repeated N-person public good game. Agents either interact with a random group, globally chosen from the global population, or the same group (locally). Agents then update their strategies by comparing their payoffs to either those of their reference group or the global population. Thus agents engage in two kinds of playing modes and learning modes, respectively. With random groupings, defection is the norm with some occasional, short-lived outbreaks of cooperation (these outbreaks are enhanced when selection is local). When agents stay within the same group, cooperative outcomes are much more likely, with relatively high levels of sustained cooperation in many of the resulting groups especially when selection is global. Local interactions with global updating provides the most favorable environment for the evolution of cooperation.

32.          Social Quantum Logic:An Alternative to Cooperation and Game Theories

                 W.F.Lawless (lawlessw@mail.paine.edu, lawless@itd.nrl.navy.mil)

     The shift between individual and social behavior has never yielded to traditional theory (Allport,1962).Game theorists assume that logical conceptions of cooperation and conflict in a static configuration represent the decisions made by humans in an interaction (Von Neumann & Morgenstern, 1953), producing a stable solution of mutual competition (Nash equilibrium; Nash,1951).  Another stable solution of mutual cooperation was found by Axelrod (1984) to evolve in extensive form.But unary maps underdetermine reality (Feynman,1967); cooperation in the field to solve ill-defined problems produces suboptimal solutions (Lawless,Castelao,&Ballas,2000); and a

rigorous logical map from individual preferences to a single group decision is impossible to construct (Arrow,1970).  Not surprisingly, game theory has failed in the laboratory (Kelley,1992),the field (Jones,1998),as has decision theory (Johnson-Laird &Shafir, 1993; Klein,1996;Parsons,2001).

     More problematic for multiple agent systems or computational autonomy,as information completeness produces knowledge (Luce &Raiffa,1967),as the number of interactants approach 100 or more (Preskill,2000),or as agents depart from cooperation (Castelfranchi & Falcone, 2000), computability decreases significantly.   Thus,traditional models of the social interaction are as problematic as traditional computer models of quantum interactions (Lawless &Castelao,2001).In contrast,following Campbell ’s (1996) rejection of his seminal approach,adapting Bohr ’s (1955) quantum logic to the interaction is surprisingly successful and produces a robust model of decision-making even as N increases (Lawless,Castelao,&Abubucker,2000)). The evidence for social quantum logic suggests that by reducing the asymmetric information produced from cooperation, adversarial collaboration is superior to consensus collaboration (Lawless, forthcoming,2002), contradicting Nash (1950).It indicates that emotion is derived directly (Lawless,forthcoming,2001). 

And it indicates that optimum decisions occur when the tension between cooperation and competition produces sufficient emotional responses to process information (Lawless &Castelao, 2001).

(For references, please check: http://www.sscnet.ucla.edu/hcs/intercon2002/abstracts)

33.          Non-Employment Benefits and the Evolution of Worker-Employer Cooperation:

Experiments with Real and Computational Agents

Mark Pingle, Leigh Tesfatsion (tesfatsi@iastate.edu)

Paper URL:  http://www.econ.iastate.edu/tesfatsi/sce2001.pdf

     It is now commonly understood that the complexity of most employment relationships forces the typical employment contract to be incomplete.  If the contract does not enforce the desired level of cooperation, it is reasonable to think that other institutions might arise to do the job.  Using experiments with both real and computational agents, this paper examines the possibility that the level of non-employment benefits affects the level of cooperation between workers and employers, thereby impacting the unemployment rate, the productivity of labor, and a variety of other economic outcomes.

     A distinctive feature of our experimental employment study relative to previous theoretical studies is that matches between workers and employers are determined endogenously, on the basis of past worksite experiences, rather than randomly in accordance with some exogenously specified probability distribution.  In each stage, workers either direct work offers to preferred employers or choose unemployment and receive the non-employment payoff, and employers either accept work offers from preferred workers (subject to capacity limitations) or remain vacant and receive the non-employment payoff.  Matched workers and employers participate in a risky employment relationship modeled as a prisoner's dilemma game.  Both the computational agents and the human agents evolve their partner preferences and worksite behaviors over time on the basis of past matching and worksite experiences.

     In both types of experiments, increases in the non-employment payoff result in higher average unemployment and vacancy rates while at the same time encouraging cooperation among the worker and employers who do form matches.  On the other hand, given a high non-employment payoff, an increasing number of the computational workers and employers learn over time to coordinate on mutual cooperation and avoid coordination failure, so that overall efficiency increases as well.  This potentially important``longer run'' policy effect is not clearly evident in the necessarily shorter trials run with human subjects.  This difference raises challenging issues both for human-subject experimentalists wishing to conduct social policy impact studies and for computational experimentalists who wish to use human-subject experiments to validate their computational findings.

34.                Integrating Downs and Duverger:  Modeling Party Systems under Conditions of Uncertainty

Emily Clough (eclough@polisci.umn.edu)

     Much of the work on party systems in the last 50 years has been based on Downs’s median voter theorem and Duverger’s Law.  Downs’s median voter theorem deals with the movement of parties on the ideological spectrum in order to improve their level of support, and Duverger’s Law relates electoral system to the number of parties in a party system.  While these two theories clearly deal with different aspects of party systems, there is a great deal of potential for the results of one to affect the other.  Formal theorists have created countless models of each of these theories,  but seldom have scholars incorporated aspects of both into a single formal model.  In this paper, an agent-based model is used to integrate elements from both theories into a model of party systems under conditions of uncertainty.  In this model, voters use local information to infer the relative standing of candidates; this, in turn, allows them to vote strategically.  Political parties then use these votes to assess and improve their platforms.  The results from this model demonstrate that the outcomes of either theory do not necessarily emerge when they are integrated and implemented under conditions of uncertainty.  These results show the importance of considering the possible effects of one theory on the other under conditions of uncertainty.

35.                Complexity in Legislatures

                David Epstein (de11@columbia.edu)

36.                Adoption of a New Regulation for the Governance of Common-Pool Resources

by a Heterogeneous Population

Marco A. Janssen (maajanss@indiana.edu), Elinor Ostrom

   Existing theories of collective action do not yet provide an adequate explanation for how appropriators from a common-pool resource can organize themselves, establish new rules, monitor rule conformance, and sanction rule breaking. This paper uses a multi-agent computational model to explore the factors that enhance or detract from the possibility that individuals harvesting from a common-pool resource will impose rules on themselves to limit their use of a jointly used resource and then monitor and enforce their own rules.  

    In a stylized spatially explicit simulation model agents share a common resource. Initially, no regulation is implemented, and the population will experience serious collapses. The agents know a possible regulation. Our aim with this model is to investigate the conditions under which the population will implement this regulation, and how successful it will be monitored and sanctions applied. To implement the rule, agents should become motivated to vote for the candidate rule. Their motivation depends in the mutual trust in other agents. All agents have symbols, and when agents meet they can play a trust game. The outcome of the trust game informs the agents ability to predict trustworthiness in agents based on observed symbols. When a rule is implemented, agents may break rules, which is related to trust in the other agents, and may monitor, which is an agent dependent probability.

    We explore different configurations of the model to investigate when a cooperative solution is likely to evolve. One of the configurations is to split up the torus in two areas. These areas may differ in carrying capacities and agents may, in periods of scarcity enter the other region. We analyze the difference between the conditions that agents will accept the rules in the other region or not.

    Our findings of our artificial world are related to findings of field research of common pool resources. An interesting finding is that heterogeneity does not have to be a bottleneck in adopting a rule, as long as mutual trust relationships can be created. The success of creating mutual trust relationships depend on the type and frequency of social interactions.

37.          The Rotating Presidency of the European Council as a Search for Good Policies

Ken Kollman (kkollman@umich.edu)

        The rotating presidency for the European Council is a curious, and unusual, political institution.   Each countryin the European Union (EU) takes a turn in the presidency, for a six month term, and the rotation is fixed by the treaties governing the Union.  The institution rotates among all fifteen countries sequentially. In this paper I propose a computational model to study the idea of rotating the leadership position in a decision making body.   In the model, the members of a council vote on policies using a variety of decision making institutions.   I evaluate these various institutions using utilitarian criteria.

38.                Changing Roles of Meta-agents in Simulations of The Tragedy of The Common

Keiji Suzuki (suzukj@fun.ac.jp)

     Agent based social behavior simulations are research filed that treats complex game situations and examines artificial intelligence. Social dilemmas are one of the complex game situations and suite to

examine the intelligence of agents. In this paper, the Tragedy of the Common, which is one of the social dilemmas, is treated in the agent-based simulation. In this game, players use common limited resources to get the reward. If players behave based on the individual rationality, all players will face to tragedies loosing higher payoff. To avoid such tragedies, players have to make the relationship between other agents to prevent the individual behaviors or change the problem structure, for example, changing the

payoff functions.

     The proposed approach is kind of the changing problem structure. That is, the meta-agent is introduced to control the levy charging to the players. In previous researches, the role of meta-agents is fixed. Because it is important who acts the role of meta-agents in real society, agent based simulation should be provide the decision mechanism that includes the selection of the role.Therefore, it is proposed that role of the meta-agent is treated as one of the activities of the players. Namely, the player will select the role of the meta-agent if the expected revenue exceeds than the expected rewards when the agent acts as the player.

     Another problem for introducing the meta-agent is how to set the plan of the levies charging to the activities of the players. To acquire the plan, co-evolution of levy plan between the agents is prepared.Based on the co-evolutionary acquisition of the levy plan and the decision mechanism including the selection of the role, it will be examined the effects of the interaction between the meta-agents and the agents in the Tragedy of the Common game.

39.          Conflict and Trade in Cellular Automata World Politics -- Adding the Liberal Factor

Byoung W. Min (min.16@osu.edu)

     The tradition of simulations in the study of international relations has relied on the realist theory, while the liberal theories have flourished during the past decades. The realist theories have focused on the conflict dimension, so that many scenarios written by them have emphasized the war-ridden worlds or the emergence of brutal empires, based on the logic of power politics. Bremer and Mihalka's "Machiavelli in Machina" (1977) and Cusack and Stoll's Automatic Stabilization Model (1990) are among the early cellular automata simulation models using the realist framework. In these simulations, world politics works by the mechanisms of war, alliance, and territorial conflicts. While these mechanisms represent many important factors of world politics, the introduction of liberal themes, such as trade and cooperation, have not yet been actively included in the computer simulation models.

     This study tries to investigate the probable scenarios of world politics by including the trade factor to the existing realist paradigm. The main question is how world politics changes its shape if we move from the realist framework to a mixed one modified by the liberals. The basic hypothesis is that increasing trade will reduce the frequency of war while increasing the possibility of multipolarity. Besides, I would like to see the impacts of trade on the state behaviors, such as balance of power, and the state survival rates. This study will reproduce the core of Cusack and Stoll's original conflict model, which is to be combined by the trade algorithm extracted from the Sugarscape Model by Epstein and Axtell (1996).

     The agents of this combined model will choose their own behavior, war or trade, based on the calculation of expected utility of each behavior toward all of their neighbor states. For every iteration, a state agent will be chosen among all states holding positive expected utilities. Agents are assumed to be rational in that they calculate the expected utility for their choices. The factor of error is added to the calculation process and the scope of agents‚ calculation and behavior is perfectly local. However, unlike the old models that have relied too much on system parameters, this study will focus on the theoretical mechanism. In particular, system parameters will be made to fluctuate randomly in a range theoretically assumed, instead of being fixed at specific values. By this method, the study will compare the simulation results between the war-only world emphasized by realism and the mixed world supplemented by the liberals, in the areas of system endurance and multiplicity, patterns of balance of power, and state survival rates.

40.          On The Transition to Agent-Based Modeling: A Case Study

László Gulyás (gulyas@fas.harvard.edu)

     In this paper I study the effect different implementation approaches have on the actual setup of an individual-based computer model. My goal is to demonstrate how these techniques 'guide our hands' and hide our implicit decisions (assumptions about the modeled system). The connection between the conceptual model and its actual implementation is important, especially, as the later rarely gets published. On the other hand, I am interested in 'what makes a model agent-based'. Is it enough to model a system composed of individuals? Or should the agents' behaviour and/or the effect of their behaviour be at the centre of our attention? To answer these questions I have re-implemented a simple, individual-based model [of Paul Krugman’s] in RePast. While the model was originally published in a variable-based form, I have created four different versions of it ranging from the variable-based one to a version containing autonomous agents. My results provide evidence that even if the system consists of individual entities (say, agents) and the author's main concern is their behaviour and/or the aggregate effect of these behaviours, the model itself could actually ignore the agents.

41.          Robust Adaptive Planning: A New Decisions Sciences for Complex Systems

Robert Lempert (lempert@rand.org, lempert@evolvinglogic.com)

     Models of complex systems can capture much useful information, but can be difficult to apply to real-world decision-making because the type of information they contain is often inconsistent with that required for traditional decision analysis. In particular, complex systems models may be most useful under conditions of deep uncertainty, that is, where the emergent behavior of the system makes accurate points forecasts or probabilistic predictions -- and thus traditional decision analysis -- difficult.  New approaches, such as Computer Assisted Reasoning (CAR), which use inductive reasoning over large ensembles of computational experiments, now make possible systematic comparison of alternative policy options under conditions of deep uncertainty. Such CAR approaches enable two key analytical steps important to decision analysis with complex systems: 1) the use of ensembles of plausible models, rather than any single best guess, as the best description of the available information about the future, and 2) the use of criteria such as robustness and satisficing, rather than optimality and efficiency, to compare the performance of alternative decisions.

     This talk describes the Computer Assisted Reasoning approach to decision-making under conditions of deep uncertainty which is ideally suited to applying complex systems to policy analysis. In particular, this talk will describe the use of CAR to support robust, adaptive planning (RAP).  The talk will draw examples from a variety of decision problems, including the policy problem of global climate change, with a particular focus on the role of technology policies in a robust, adaptive strategy for greenhouse gas abatement.

42.                Foundations of "New" Social Science: Institutational Legitimacy from Philosophy, Complexity Science, Postmodernism, and Agent-Based Modeling

Bill McKelvey (mckelvey@anderson.ucla.edu)

     Since the death of positivism in the 1970s, philosophers have turned their attention to scientific realism, evolutionary epistemology, and the Semantic Conception of Theories. Building on these trends,  Campbellian Realism allows social scientists to accept real-world phenomena as criterion variables against which theories may be tested without denying the reality of individual interpretation and social  construction. The Semantic Conception reduces the importance of axioms, but reaffirms the role of models and experiments. In addition, philosophers now see models as "autonomous agents" that exert  independent influence on the development of a science, in addition to theory and data. The inappropriate molding effects of mathematical models on social behavior modeling are noted. Complexity science offers  a "new" normal science epistemology focusing on order-creation by self-organizing heterogeneous agents-and featuring agent-based models. The more responsible core of postmodernism builds on the idea that agents operate in a constantly changing web of interconnections among other agents. The connectionist agent-based models of complexity science draw on the same conception of social ontology as do postmodernists. The recent trends in philosophy of science, the notion of models as autonomous agents, the new normal science epistemology from complexity science, connectionist postmodernist ontology, and use of agent-based in place of mathematical models combine to provide foundations for a "new" social science centered on formal modeling not requiring the mathematical assumptions of agent homogeneity and equilibrium conditions. Together these foundations give "new" social science a level of institutional legitimacy in scientific circles that current social science approaches lack.

43.          Layered Situations and Agent Orientation

David Sallach (sallach@uchicago.edu)

      Herbert Simon hypothesized (1996:53) that the apparent complexity of human behavior reflects the complexity of the environment in which we find ourselves.  In order to effectively model the dynamics of social processes, it is necessary to capture, not only the complexity of natural and social ecologies, but also how situated agents with limited cognitive capacity apprehend and respond to settings that are continually redefined.

       In recent decades, the ‘situation’ has emerged as a focus in the scientific modeling of complexity.  Philosophically, Popper (1995) describes situations as having objective propensities that tend to be realized.  In linguistics (Barwise & Perry 1983), mathematics (Devlin 1991) and logic (Barwise 1989), situation theory has emerged as a powerful and flexible formalism for modeling context.  In sociology, Collins (1994) has proposed ‘situational reductionism’ as a way of bridging between micro and macro dynamics.  In artificial intelligence, situated agents of various degrees of complexity have been an important innovation (Hendriks-Jansen 1996; Clancey 1997; Ferber 1999), addressing issues on which conventional AI has foundered (Brooks 1999).

      When considering situational modeling, a part of the complexity of social dynamics results from the ability of human agents to define the same events from a variety of situational scales.  A battle is part of a campaign, a war and a period of national renewal or decline.  Further, since all participants have this capability, definitions of situations may abound and require reconciliations.  Common definitions of relevant situational scale may contribute to social coherence, while divergent scale may make coordination difficult.

       Integration of multi-scale orientation into agent simulation has the potential to increase the plausibility of social models, without a significant increase in agent complexity. This paper uses situation theory to model multi-scale agent focus, and to illustrate issues arising from the support of situational orientations.

44.          Housing Segregation: Multiple Implications from a Simple Agent-Based Model

Darren Schreiber (dschreib@ucla.edu), Sander

     Thomas Schelling's famous model of housing segregation started with a few coins on an eight by eight grid and some very simple assumptions about individual preferences. Using the SWARM programming environment, we have extended Schelling's concept to examine the contemporary debate about the nature and causes of housing segregation. We begin with basic preferences functions derived from empirical data on neighborhood racial composition and add a variety of putative factors in housing decisions. The result

is a sophisticated model of racial housing segregation that provides insight into empirical patterns of segregation and desegregation in the twentieth century.

45.                Quantifying Adaptation of Evolved Structures

Mark A. Bedau (mab@reed.edu)

      Discerning and measuring progress of adaptation in evolving systems is a key challenge for those attempting to answer the deepest questions about evolution in biological and social contexts. This paper introduces and illustrates a method for visualizing and quantifying adaptive processes that applies to both biological and cultural evolution, as reflected in data produced either by natural systems or agent-based models.

      Agent-based models provide a powerful new method for understanding adaptation in biological or social contexts. Indeed, those models might even capture an abstract enough representation of evolutionary dynamics to permit universal principles applying to both biological and cultural evolution to be identified. But extracting value from agent-based models also presents special challenges. Successfully addressing those challenges involves comparing the behavior of models with other models and with data produced by natural systems. Evolutionary activity statistics (Bedau and Packard 1992, Bedau et al. 1997, Bedau and Brown 1999) have been used to visualize and quantify evolutionary creativity as reflected in data produced by many different biological models and natural biological systems. An important part of the use of these statistics is normalizing them with independent measures of non-adaptive noise.

     Evolutionary activity statistics can also be used to visualize and quantify the dynamic creativity of cultural evolution. This is illustrated by measuring the evolutionary activity of technological innovations as reflected in patent records. Patents provide a practical context for investigating cultural evolution for a variety of reasons; important among them are that patented inventions have been certified as useful innovations and that patent records are readily available in electronic form. Preliminary investigation of evolutionary activity in recent US patents confirms the independently supported conclusion that information technology is the most creative and innovative sector of recent technological innovation. The practical application of these statistics to agent-based models in the social sciences is discussed.

46.          Chaotic Nature of Online Translators

                Noah Goldstein (noah@geog.ucsb.edu)

47.          Failure in Collective Systems:  The Effect of Environmental Stability on Collective Self-Organization and the Chaotic Effect of Multiple-Issue Platforms in Electoral Processes

Norman L. Johnson (nlj@lanl.gov)

       There is a growing appreciation of the power of diverse collectives in solving difficult problem, as exemplified by global efficiency in large financial markets, social movements that precipitate needed change (e.g., the fall of the Berlin wall), or knowledge self-organization driven by human actions in electronic databases (e.g., book referral at Amazon.com).  As we rely more on these collective processes to solve society's and businesses' more challenging problems, we must understand the stability of these processes.  Two examples are used to illustrate mechanisms of failure in collective systems.

       The first example is a simple self-organizing system - the foraging for food by ants (one of the sample simulations provided with the public software - STARLOGO).  This system is argued to be similar in dynamics to many decentralized collective systems - ecologies, economies, knowledge systems, societies, etc.  The simulations are first shown to illustrate a developmental view of evolving systems proposed by Salthe and Johnson, captured by the developmental cycle of Formative, Co-operational and Condensed stages.  The effects of different rates of environmental change are then presented.  No effect is observed from small rates of change.  As the rate increases, innovative information becomes more important.  As the rate further increases, stabilizing informational structures (collective pheromone clouds) fail and the system regresses to earlier developmental stages.  It is observed that just before the times of failure, the system exhibits high collective coherence, which results in a loss of innovation that could have prevented the failure.   Hence, the collective processes that form these structures are shown to inhibit the performance of the system as a whole in rapidly changing environments and can lead to undesirable episodic failure.  In rapidly changing environments, the system remains in the Formative stage, all the productivity results from the innovators, and the existence of collective structures only degrade the overall system performance.  The important lesson learned from these results is the role that collective effects play in system failure and how collective effects should be managed depending on rates of environmental change.

       The second example concerns diversity in electoral processes.  In the 2000 Presidential election, many of the states had popular votes that were equally split between the two main candidates - Florida being the prime example.  This 50/50 split caused the outcome to be sensitive to details of the voting procedures and participation - aspects that in prior times were considered less important.  The sensitivity of a global outcome to the smallest details is characteristic of a chaotic system and does not bode well for future elections if it is likely to occur again.  Many ascribe the chaotic outcome of the 2000 election to an unlikely event, possibly enhanced by candidates seeking the common middle ground. Is democracy in crisis or was the 2000 Presidential election an anomaly. A theory, and supporting simulations, is presented that shows that this was not an unlikely event and will be prevalent in the future.  The results in the 2000 election are shown to be a direct consequence of platforms that appeal to multiple issues, which in turn reflect the increasing diversity of views important to voting populations.  This conclusion is shown to be true even if single issues are largely one-sided in the populations.   A variation of the Central-Limit theorem (combination of a sufficient number of diverse distributions will result in a normal distribution), when applied to the voting preference populations, provides the theoretical basis.   The simulations show how random assembly of platforms constructed from voter preferences based on actual exit-poll data result in 50/50 splits as the number of issues in the platform increase.   This is shown to be independent of the skewness of the distributions underlying the voter preferences, in agreement with the predictions of the Central Limit Theorem.  This example illustrates the undesirable consequences of applying traditional democratic methods to increasingly diverse and large populations.

        In both of these examples, diversity of information is shown to be the critical perspective in the analysis.  The relevance of these results to societal challenges faced in modern times by increased globalization and faster rates of change is discussed.

48.          How Information Moves From Those Who Have It To Those Who Need It:

The Information Ecology of the University

Susanne Lohmann (lohmann@ucla.edu)

     I conceptualize the university as a complex decentralized system with a unique information ecology. Decisions about people, programs, and money must be made. And yet information is distributed among faculty and administrators inside the university and political actors outside of it; decision-making powers are dispersed within and without; and the people who hold decision-relevant information do not always, or even most of the time, coincide with the people who make decisions. I ask: how does information flow from those who have it to those who need it? How do information flows shape academic and administrative decisions and political oversight? How can internal and external agents of change redesign the flow of information to improve the performance of the university?

49.          Markets on Networks

Zoltan Toroczkai (toro@cnls.lanl.gov)

     We study the effects of inter-agent communication on the evolution of a market within the minority-game framework. The inter-agent communications create a complex (social) network with small-world character. This network forms the substrate for a highly dynamic and directed network, called the action network, defined by those inter-agent communication links on the substrate along which the

passed information/prediction is acted upon by the other agent. We define three basic agent-agent communication scenarios and study in more detail the one in which the agent uses a reinforcement

learning algorithm to select the best predictor among the neighboring agents it is linked to, including himself. We show that when the substrate network is highly connected, the market behavior is characterized by the existence of a small number of spontaneously formed high degree hubs on the action network which then dominate the market evolution. In certain parameter ranges the agents will spontaneously generate a high degree of cooperation making the market almost maximally efficient. We discuss the relevance of these effects to designer agent-systems.

This work has been done in collaboration with:

Marian Anghel (LANL), Kevin Bassler (UH) and Gyorgy Korniss (RPI)

50.          Task Commitment, Diversity and Workgroup Outcomes Abstract

                Brooke Harrington (brooke_Harrington@brown.edu)

51.          The Evolution of Complex Despotic Societies by Self-Organisation and Natural Selection

C. K. Hemelrijk (hemelrij@ifi.unizh.ch)

     In this paper, I will use an individual-centered model, called DomWorld, to show how both natural selection and self-organization may be involved in the evolution of a despotic society. I represent group life in DomWorld, by artificial individuals that live in a homogeneous world and only flock together and, upon meeting one another, may perform dominance interactions in which the effects of winning and losing are self-reinforcing.  Starting from a situation with abundant food and mildly aggressive animals living in egalitarian societies, we may imagine that, if food becomes scarce, natural selection will favor the individuals that are more intensely aggressive. If, correspondingly, in the model intensity of aggression is increased,  the society switches from the many characteristics of an egalitarian society to a despotic one due to a complex feedback between the hierarchy and spatial structure. The many differences between the two types of artificial societies correspond closely to those between despotic and egalitarian macaque species in the real world. Since, in the model, they organize themselves as side-effects of the change of one trait only, they may also in the real world arise as side-effects of mutation of a single trait.

    Groups with the steepest hierarchy may have the best chance to survive, because in such a group at least some individuals may succeed in reaching reproductive age.  Therefore, in primate species group in which one sex remains in their natal group for life, group selection may operate as well.

52.          The Traffic rules of fish Schools: From Individuals to Aggregations                 

                Julia K. Parrish (jparrish@u.washington.edu), Steve Viscido, Danny Grunbaum

53.          The Emergence of Order from Disorder as a Form of Self Organization

Dwight Read (dread@anthro.ucla.edu)

     The evolution of cultural constructs is not well understood.  Cultural evolution has been modeled based on genetic evolution (sociobiology), in analogy with genetic evolution (dual inheritance) and as a process of transmittal of information units (memes).  None of these is completely satisfactory as each fails to come to grips with the way in which cultural constructs may be in the form of an abstract, symbolic system.  One such cultural construct, fundamental to the organization of human societies, is kinship with its expression in the form of a kinship terminology.  The kinship terminology provides a means for identifying who are one's kin and the boundary of one's kinship domain, for many societies, is the boundary of the society.  Yet another means of relating one individual to another that is ontologically prior to a kinship terminology is through genealogical tracing, the basis of "family trees."  The fact that there may be two conceptually distinct, though overlapping, constructs for determining a conceptual linkage that relates one person to another within the same society raises the question of why one or the other alone is not sufficient.  I argue that the two constructs-- a kinship terminology and genealogical tracing -- are qualitatively different in terms of the kind of order that each provides for the collection of persons making up a society.  I suggest that the kinship terminology arises through abstraction of the organizing principles of genealogical tracing and resolves what otherwise would be a conceptual disorder were the boundaries of the society determined through genealogical tracing alone.

54.          The Winner-take-all Phenomenon in Markets where Network Externality is Ineffective

Hitoshi Yamamoto (hitoshi@rs.kagu.sut.ac.jp), Isamu Okada (okada@s.soka.ac.jp),

Nobuchika Kobayashi (nobu@ohta.is.uec.ac.jp), Toshizumi Ohta (ohta@is.uec.ac.jp)

     We consider the mechanism of the winner-take-all phenomenon in markets in which network externality does not work. The development of information networks has led to the appearance of new economies referred to as "digital economies", in which a winner-take-all phenomenon is observed as a feature. This phenomenon can be explained in terms of network externality, lock-in, and path dependency. We give examples of markets in which this phenomenon is observed, including the OS market and the cellular-phone market. However, a winner-take-all phenomenon is also observed in markets in which the economic laws of a digital economy do not work. To date, no model explaining this phenomenon has been reported. Thus, to observe the features of this phenomenon, we develop a multi-agent model of communications and consumer behavior, and with it simulate the market phenomenon. In our analysis, we make a clear distinction between Winner-Take-All and Lock-In. That is, Lock-In is one of the factors which produces Winner-Take-All. The term 'Lock-in' is used to refer to a situation in which the cost of converting from specific technologies or goods to other factors is so high as to be nearly impossible. Lock-In exists on individual, organization, and market bases. Winner-Take-All is a phenomenon under which the Lock-In phenomenon advances, involving consumers who are still "locked in". We define the factors which produce the Winner-Take-All and Lock-In phenomena as "Winner-Take-All" and "Lock-In" drivers.Various factors have contributed to the rise of the "Winner-Take-All" phenomenon. One typical such factor is network externality. A number of other factors can also be considered.. We researched the Winner-Take-All phenomenon in several markets, including the mobile telephone, fast-food and music markets, and observed the phenomenon in all of them. Applying traditional economic laws enabled us to understand the mobile phone and fast foods markets, but not the music market. Consequently, we need to develop a model that will enable us to more clearly understand the

Winner-Take-All mechanism in markets where network externality is ineffective.

     In developing such a model, we assume the existence of "action rules" for individuals. Such rules are represented by two axes, i.e., a communication axis and an information seeking axis. We extracted these rules in researching and simulating goods selection and communication among individuals. Our objective in doing so was to determine what social phenomena emerge when agents use such action rules in cases such as changing information channels. We found that horizontal exchanges of information facilitate customization of information, and were able to intuitively reason that such customization of information increases social diversity. We then applied these action rules to explore the reasons the Winner-Take-All phenomenon is generated.

55.          Criteria Indexes of Multiagent Simulators

Isamu Okada (okada@s.soka.ac.jp), Toshizumi Ohta (ohta@is.uec.ac.jp)

    A computer simulation approach is proposed involving the development of a number of simulators to analyze complex social phenomena. Each simulator has underlying structures that fit subjects as a background since both simulator and subject share common underlying structures. However, all the simulators cannot be evaluated appropriately because no clear indexes exist for them. We therefore propose criteria for multiagent simulators for conducting bottom-up oriented studies on organization.

    First, we survey studies on a complex organizational and social theory, and demonstrate how a simulation method based on a multiagent system can be applied to these studies. We then describe the model elements - agent, environment, structure and macro-index - that are applicable to a bottom-up oriented approach to complex organizational and social studies. Each of these elements consists of inner elements. We then discuss some necessary conditions for the simulators that were identified from criticisms made of the application of simulators to studies. The main criticisms are that simulation behavior is essentially too complex to be analyzed, that to require many parameters is to be ad hoc, and that simulators are hard to operate, and that there are few simulators which have satisfactory interfaces.

    As a result, the criteria we propose are the ability to describe and generate simulation models, the ability to perform individual-level analysis and mechanism analysis, the efficacy of model parameters, the qualitative sufficiency of the interfaces, and simplicity in programming. We apply these criteria in evaluating models such as the Mathematica, the Swarm, the ABS, the AGENT-0, the Plural-Soar, and the Operational Organization Oriented Simulator (OOOS) we are now developing. We also propose a measurement model and apply the OOOS to a number of measurable indexes.

56.          On an Analysis of Trends Concerning Agent-Based Modeling in the Social Science with Web Mining Approach

Kazunari Ishida (mactwist@oak.dti.ne.jp), Toshizumi Ohta (ohta@is.uec.ac.jp)

     We collect documents concerning agent-based modeling in the social science on the Internet, and analyze them with term relation frequency (TRF) method. In order to collect various types of documents concerning the social science, we develop web robot for specified types of documents on the Internet. Based on the collection and analysis, we try to provide digital commons concerning the agent-based modeling.

     Researchers who have various backgrounds are trying to capture complexity behavior of systems. Due to the operational capability of agent-based model, they have recognized the availability of the model and have applied their researches. However, systematization of the results of researches across many fields is a difficult problem because of the unique characteristics of each research objective. In addition, systematization work cannot realistically be done by hand, because the amount of documentation is enormous.

     Fortunately, because of the development of information technology, we can easily collect and analyze digital documents by computer and network. Our method and system enable researchers realize the collection and analysis easily. The analysis method TRF is a statistical analysis method for statistical data concerning term relations in documents, and it makes them recognize the degrees of commonality and originality among documents across multiple interdisciplinary academic areas. According to the results of our analysis with the collection and analysis, they may provide digital commons concerning the agent-based modeling in the social science.  Employing the TRF method with hierarchical clustering, we found and analyze four categories concerning agent based modeling on the Internet, such as Formal Logic, Platform architecture, Object & Protocol, and DAI, Complexity, & Organization Theory. The contents in the category of Platform Architecture tend to be comprehensive for agent based modeling, because they are concerned with basic mechanism for managing agents. That of DAI, Complexity, & Organization Theory tend to be diverse around agent based modeling, due to the nature of domain specific targets, such as biological pheromone, organization, and society, and so forth. The trend of Formal Logic is to describe norms and rules in an organization with formal logic and to analyze phenomenon in it. The trend in Platform Architecture is to develop the architecture for implementing and managing agents for application. The trend of Object & Protocol is to define standard format for describing agents' status and communication between them. The trend of DAI, Complexity, & Organization Theory is to analyze complex phenomenon from biological level to social level and to develop practical information

system such as CSCW. The clarification of trends let us understand the relations between our researches and the others.

     The TRF method also reveals similarity and dissimilarity among documents that hierarchical clustering can 't detect, whether the documents are in the same category or not. According to the results, it may be a start point to define a guideline for development in the area of agent based modeling  in the social science.

57.          Social Informatics and Cyber Commons in Computational Analysis

Toshizumi Ohta (ohta@is.uec.ac.jp), Kazunari Ishida (mactwist@oak.dti.ne.jp),

Isamu Okada (okada@s.soka.ac.jp), Hitoshi Yamamoto (hitoshi@rs.kagu.sut.ac.jp)

We discuss social informatics as an emerging discipline, and explore the role of cyber commons in organizing and the application of computational analysis to study it. Social Informatics (http://si.ohta.is.uec.ac.jp/) is an interdisciplinary study to explore a function of information in a social  system, and to design a system for information exchange in a society. The social informatics aims to promote welfare of human beings in a society.

Cyber commons are emerging in the Internet age as repositories for the generation, accumulation, and distribution of information and knowledge to benefit societies. In this new age, we can characterize societies in which a paradigm is changing to an auto-genesis or self-genesis from an allo-genesis or other-genesis.

Within the context of such a characterization, cyber commons are an aspect of social informatics that concerns the arrangement of information space in a society. Two questions that need to be addressed are how cyber commons differ from traditional commons, and how they can be fostered and utilized. To answer these questions, we discuss properties of cyber commons in contrast with those of traditional commons, and provide and explain simulation results we obtained with respect to this topic. Based on these answers, we explore the role of cyber commons in this new age, and discuss a new form of organizing employing examples concerning a virtual organization, an intermediary of information and knowledge exchanges, and social dilemma problems.

Results of our computational analysis concerning the cyber commons reveal that emergent properties can be observed in our model, and that alternative hypotheses are explored with respect to the properties of cyber commons.

We conclude that virtuality, viability, and visibility have to be fostered to provide benefits to coming societies, and that the concept of operational organization must be a promising approach to understanding the societies

58.                Financial Fragility, Heterogeneous Agents' Interaction, and Aggregate Dynamics

Mauro Gallegati (gallegati@deanovell.unian.it), Domenico Delli Gatti,

Gianfranco Giulioni, Antonio Palestrini

     According to the traditional views on economic fluctuations, large fluctuations arise because of some impulses propagate through the entire economy (the so-called Slutsky-Frish approach) or because of some endogenous properties of the system (the deterministic approach). Despite their different attitudes, and a different methodological approach, they share a common analytical tool: the "representative agent" hypothesis. In the last few years, this tool has been under the attack of a growing criticism, beginning with Kirman, 1992. As well known, very restrictive analytical (and empirically implausible) conditions are requested to have exact aggregation. Recent empirical works show that heterogeneity can explain aggregate dynamics: idiosyncratic shocks affect the rate of change of macroeconomic quantity (Davis et al. 1996, Davis and Haltinwanger 1996, Caballero et al. 1997). Theoretical research and applied investigation demonstrate that macroeconomics is not equivalent to the simply "summation and averaging" process of individual agents. Since the aggregate can be (and under very general conditions it is) different from the sum of its component, to analyse the behaviour of a representative agent as it were representing the whole economy is misleading. The failure of the law of the large numbers can be blamed for such a result. In fact, it holds only if non-linearities are not at work (Allen, 1982) or if non-market interactions are ruled out (Brock and Durlauf, 2000).

     One of the puzzles the equilibrium theory of fluctuations has to face is why large fluctuations arise without any large shocks. Idiosyncratic shocks are natural candidates for small shocks but we still need an amplification mechanism able to produce large movements. Unfortunately, if the system is linear, small shocks may produce only small effect. In this paper we analyse a model in which interaction and financial fragility are the source of non-linearities. As we said before, there is an immediate consequence regarding the law of large numbers whose validity does not hold any more. Moreover, when a system is non-linear, its dynamics can generate endogenous business cycles (in out context they are coupled with stochastic elements). The presence of non-linearities and interaction, and the presence of exogenous components, make the economic system to be a "complex" one (Rosser, 1999), in the sense that there is not a long run

definite dynamics.

    A not negligible branch of literature claims that, because of informational imperfections, financial factors play an important role in output fluctuation. In the "old" Keynesian literature, financial fragility is "systemic" (Minsky, 1982) and it can endogenously cause the business cycle. In the "new" Keynesian literature financial fragility represents an amplification mechanism in the spirit of the impulse-propagation approach (Bernanke and Gertler 1989, 1990; Greenwald and Stiglitz 1988, 1990, 1993; and Kiyotaki and Moore, 1997) where informational imperfections make the system deviate from the first best solution. The presence of informational imperfections involves a setting where agents are heterogeneous and evolve dynamically. Moreover, because of heterogeneity, agents can interact outside the market possibly identifying a self-reinforcing mechanism. In our modelling strategyboth old and new Keynesian aspects of financial fragility are at work. Each firm sells its output at a random price, which can be assumed as an idiosyncratic shock. Rather than annulling, each other (the law of large number does not apply in our context), their effect on aggregate activity is amplified by the financial position of each firm. A fast growing research on empirical evidence shows that the firms' birth-death process drives employment fluctuations. Following this insight, Delli Gatti et al., 2001, consider the entry-exit process as the main factor affecting the distribution (and aggregate dynamics). Note that, since the amplification mechanism is a function of financial fragility, and this last modifies during the business cycle, our model predicts fluctuations to be "state dependent": the economy reacts differently to the same shock being the propagation mechanism sensitive to the state of financial robustness. This paper argues for two dynamical causal links. The first one runs from financial fragility to investment at a micro level: firm's investment spending is determined by the availability of internal finance. The other one identifies investment activity as the main determinant of internal finance. Differently from the first one, aggregate elements affect this link via interest rate changes. Moreover, attention should also be paid to  the cash flow-debt commitments ratio. Differently from the mainstream literature, this paper explicitly models firms' turnover and the behaviour of firms and banks and their interaction through the dynamics of the interest rate (whose changes are affected by a proxy of the financial fragility) using an agent based framework (an economist' version of SWARM we developed). We represent the economy as a continuum of square lattices; each one is filled with many firms and one bank. Firms sell a homogeneous good at a given stochastic price. This stochasticity is the source of uncertainty in the model, which is the ultimate cause of bankruptcies. When a firm goes bankrupt, it leaves its site. Empty sites are filled in a stochastic way; in particular the probability of a new-firm birth is higher the better credit conditions are (as proxies by the mean financial position of the zone). Again, financial factors affect firms' geographical and equity distribution, the turnover rate and, at the end, aggregate dynamics. (A similar approach, albeit in a different context, can be found in Campbell, (1997), who considers embodied technology instead of financial market imperfections. Emphasis on the relations between financial heterogeneity and industrial dynamics can be found in a series of recent papers by Cooley and Quadrini (1998,1999).) The paper is organised as follow. In sections 2 we describe the model: after having exposed the theory of the firm (2.1), we discuss the bank behaviour as stemming from agents' interactions (2.2). Section 3 presents the simulation's results, while the following section estimates the model by using UK longitudinal panel data between 1984-1999. Section 5 concludes.

59.          Monte Carlo exploration of mechanisms for the creation of aggregate wealth

Klaus Jaffe (kjaffe@ivic.ve)

     Wealth creation is the main aim of most human economic activities, although the creation of wealth for the whole society has eluded the efforts of the larger part of counties in the modern world. How is capital created and increased? At the individual level, this question might have convincing answers in economic theory, but at an aggregate level, i.e. at the level of groups, societies and nations, the answers are less convincing. Explorations of this phenomenon with Sociodynamica, an agent based computer simulation model, suggests that some classical mechanisms that have been proposed by Game theory as providing a basis for the creation of aggregate wealth, such as reciprocal altruism, do not work without additional assumptions of the nature of economic activity. Specifically the simulations suggest that complex economic activity requires either the existence of synergistic effects for its maintenance, or for non-economic motivators as engines of economic activity.

     The agent based computer simulation model Sociodynamica, creates economic agents that interact with each other at different levels (the model simulates commercial interactions, altruistic interactions, knowledge transfer, induction of motivations) and that exploit the environment in different ways. The model allows for a Monte Carlo exploration of the parameter space in a range of possible economic scenarios. The economic societies simulated include pure agricultural systems (agents exploit only renewable resources), and more sophisticated  agricultural societies to which increasing complex economic activities are added, such as mining (agents exploit non-renewable resources), commerce (exchange of utility between agents), and exploitation of non-material resources (such as information and knowledge).

     The simulations show that in order to stabilize societies beyond the purely agricultural stage, motivators, such as fear, ambitions of power, lowing care, or need to transcend, are required. The strength or degree of these motivators will affect the structure of the division of labor in the society.

The model Sociodynamica was written in Visual Basic and requires a Windows environment. It is available at http://atta.labb.usb.ve/klaus/klaus.htm

60.          Fighting the Darkness:

Illuminating Vulnerabilities in California's Electricity Pricing Policy

Michael North (north@anl.gov), C. Macal, V. Koritarov, T. Veselka, G. Boyd

    Electricity pricing policies define how electric power producers are paid for their inputs to the power grid.  A variety of pricing policies is currently being used in electricity markets.  Determining an effective pricing policy is a complex endeavor that has important ramifications for market stability.  This stability can be threatened by strategic behavior such as that seen in California.  Understanding the vulnerabilities of electricity pricing policies before they are applied to the real world is critical.  Methods that can be used to test pricing policies before they are applied to real electric power markets are needed.

     The complex interactions and interdependencies between electric market participants are similar to those studied in the theory of repeated games and evolutionary game theory.  Unfortunately, the adaptive strategies employed by many electricity market participants are often too complex to be conveniently modeled using standard game theoretic techniques.  This suggests the use of a related approach, the development of agent-based models (ABM).  This approach was used to create the Electricity Market Complex Adaptive Systems Model (EMCAS).  The EMCAS ABM is designed to test some of the effects of electricity pricing policies before they are applied to real electric power markets.

     EMCAS is a Recursive Porous Agent Simulation Toolkit (RePast) ABM with agents that represent generation companies, demand aggregation companies, transmission companies, consumers, system operators, and government regulators.  These agents use a variety of computer learning techniques to improve their individual competitiveness as the market within which they are embedded evolves.

    To explore the EMCAS ABM's potential to test electricity pricing policies, two electricity pricing policies were compared.  A pricing policy requiring all transactions to use a spot market with locational marginal pricing (LMP) was compared to a pure spot market system with Pay-as-Bid pricing.  LMP below and Pay-as-Bid above a specified price cap was the electricity pricing policy in effect during the California Energy Crisis and still is today.  LMP is one of the most common electricity pricing policies. LMP is commonly used since it is widely believed that LMP pricing policies send appropriate price signals to consumers while also providing sufficient revenues for producers to maintain current production and invest in new production facilities.

     The EMCAS study results illuminate some of LMP's vulnerabilities to manipulation.  For example, under conditions of tight capacity, EMCAS model emergent behavior shows that LMP is highly vulnerable to "hockey stick" bidding strategies.  This emergent strategy employs low prices for the majority of each generation unit's capacity followed by extremely high prices for the last few Megawatt hours (MWh).  This approach can be extremely effective since it requires little risk but offers much to gain.  There is little risk since the vast majority of their low priced generation bids are likely to be accepted.  There is high potential gain since the LMP policy might assign the last few MWh's high prices to all purchased generation during times when demand nears supply.  The EMCAS study shows that Pay-as-Bid pricing may be much less vulnerable to such manipulation but may send excessively low price signals to producers.  This may tend to reduce producer's incentives to maintain current production capacity and invest in new production facilities, potentially leading to undesirable market cycles.

    By illuminating some of the vulnerabilities of California's electricity pricing policy, the EMCAS study demonstrates that ABMs can be used to test such policies, before they are applied to the real world.

This work is sponsored by the U.S. Department of Energy under contract number W-31-109-ENG-38.

61.          Income Distribution Dynamics: Marriage and Informational Cascades

Kwang Woo Park (ken.park@cgu.edu), Paul J. Zak

     This paper investigates the role of  household formation on income distribution dynamics.  This is accomplished by building an age-structured general equilibrium model in which agents are endowed with physical and psychological attributes that affect marriage and fertility decisions.  Personal characteristics are transmitted from parents to children resulting in intergenerational persistence of marriage patterns and houseshold income.  Further, psychological factors allow fads and fashions to impact distributional dynamics.  After calibrating the model, the dynamics of several variants of the model are simulated and tested against the data.  We find that psychological factors affecting marriage explain a substantial proportion of income distribution dynamics.  

62.                Simulating a Turing Tournament

                Jasmina Arifovic (arifovic@sfu.ca), Richard McKelvey

63.                Imperical Content of Behavior Models of Adaptation

Jonathon Bendor (bendor_jonathan@gsb.stanford.edu)

      Models with adaptively rational actors have become increasingly popular in recent research across the social sciences, yet numerous concerns about their empirical content exist.  We establish a set of `Folk Theorems' for a wide class of such models.  This class includes normal-form games with any number of players, each with any number of (possibly asymmetric) actions.  Players in these models adjust their actions by reinforcement of satisfactory outcomes (i.e., relative to some aspiration level), and inhibition of unsatisfactory ones.  Aspirations may adjust to reflect their payoffs.  We show that outcomes in these games are highly sensitive to initial parameters, in that any result (e.g., a sucker's payoff in a prisoner's dilemma) can be supported as a stable outcome.  These results hold even when players' aspirations are endogenous.  Intuitively, this occurs because players may be endowed with initial aspirations and action propensities that make any outcome satisfactory, and thus the actions producing that outcome will be reinforced by all players.  We then consider two possible solutions to this problem.

      First, we derive a set of requirements for the final distribution of propensities and aspirations in such games to be independent of initial conditions.  To do so, we re-cast the games as finite-state Markov chains, where propensities and (endogenous) aspirations are state values in each period.  If any of the following conditions hold: (i) players `tremble' their actions, (ii) propensities of zero or one are excluded, (iii) payoffs are random, or (iv) states are trembled, then the stochastic process is ergodic; i.e., it converges to a unique limiting distribution independent of the initial state values.  Building these conditions into models of adaptive rationality restores their predictive ability and is therefore crucial for ensuring that they have empirical content.

      Second, we also consider the role of ``reference groups" for restoring predictive content to the basic model.  In a reference group, members' aspirations may adjust according to the payoffs of other members of the group (in addition to her own).  We consider symmetric, N-person games, and find that under a weak set of assumptions, all players within a group will choose the same action and receive the same payoff.

64.          Online Interaction and Transactional Privacy Negotiations: A Computational  Simulation Analysis of an Agent-based System

Danny Fernandes (danny@Andrew.cmu.edu), Ramayya Krishnan, Uday Rajan

     The Internet together with the World Wide Web is an electronic information marketplace. The electronic information marketplace is a rich source of ever increasing personal information about individuals and organizations. The nature of the Internet makes it relatively easy to gather, integrate, and cross reference consumer personal data across different Web data repositories. This relatively easy access to personal information has led to concerns about potential violation of personal informational privacy in the online environment.

     The following observation (Hagel & Rayport, 1997) captures the essence of growing trend in the Web content provider/consumer privacy relationship.

     "But even as more and more managers begin to build strategies based on capturing information about their customers, a major change is under way that may undermine their efforts. We believe that consumers are going to take ownership of information about themselves and demand value in exchange for it. As a result, negotiating with consumers for information will become costly and complex. That process has already begun to unfold, but it could take several years to play out across broad segments of customers and products."

     Consumers on the Web are demanding a greater control over their information, and are willing to disclose personal information in a value exchange relationship (Abela & Sacconaghi, 1997) with Web content providers. Examples of value returned to consumers include customized Web content, guarantees on the privacy of the data they release, or even monetary or other compensation for releasing personal data.

     Negotiation with the consumers in the web environment need neither be costly nor complex for the negotiating parties. Intelligent Web-based software agents (Faratin, Sierra & Jennings, 1998) and Web privacy related protocols (in particular The Platform for Privacy Preferences Project - P3P, of the World Wide Web Consortium) offer opportunities for developing Web-based negotiation agents that act in accordance with privacy preferences as specified by their principals.

     In this paper, we model online privacy bargaining as a bi-lateral, multi-stage, multi-issue game between software agents (browser agents and server agents). In the model we: describe the mechanism design for negotiations (the constraint state space); specify the players and their characteristics (proposal state space),

provide rules for evaluating data requests and data offers and; design algorithms for generating and selecting counter proposals using various concession strategies (negotiation state space). 

     We implemented the model in Java and designed simulation experiments to evaluate our model.  We have obtained data that relates concession strategies to probability of negotiated agreement and, relates choice of concession strategy and agent surplus when negotiation terminates. From our simulation results we abstract propositions that govern the outcomes in the privacy bargaining game.

(For references, please check: http://www.sscnet.ucla.edu/hcs/intercon2002/abstracts)

65.          A Consumer-Based Model of Diffusion of Two Competitive, Compatible,

and Durable Goods

Masaki Tomochi (mtomochi@uci.edu), Hiroaki Murata (hiroaki-m@mbg.sphere.ne.jp), Mitsuo Kono (kono@fps.chuo-u.ac.jp)

     The diffusion of two competitive, interchangeable, and durable goods is studied under the framework of a spatial game where consumers are distributed on two-dimensional square lattice and play 3 by 3 symmetric coordination-like games with their nearest neighbors.  There are three strategies, either consuming a product A or B, or a strategy C of not consuming either A or B.  The payoff matrix of the game contains the effects of network externality, that is, the payoffs are dependent on the number of agents adopting the strategies A, B, or C.  Both simulations and mean-field approximation show that the existence of the effects of the network externality amplifies any slight initial difference in the number of agents who adopt either A or B and eventually promotes the superior product to take over the entire market.  On the other hand, without effects of the network externality the slight initial difference is not enlarged and both superior and inferior products are observed to coexist by forming clusters in the market.  Moreover, the effects of innovation factors that help an inferior product to retake over the market are studied. It is shown that both the timing and size of the innovation factor matter for an inferior product in order to retake the market.

66.          Elites from the Dark Arts and Cold Sciences

T.A. Brown (tab@ips.edu), R. Mainieri

     It is difficult to capture the state and dynamics of political systems. Social science has traditionally tried historical narratives aided by social indicators and time-series models. Historical narratives are often subjective or are limited to discussions of a few hundred people. Statistical efforts often lack first-principles derivations. Most approaches are limited in the complexity of what can be described and in the time scales that can be studied. New methods are being developed to approach the complexity of modern political structures and faster time scales. Using tools borrowed from statistical mechanics, a model is proposed for understanding the dynamics of political systems. Starting from an interacting particle system— the Voter Model—elite networks are developed. By assigning each particle a level of polit ical power, elite networks can be extracted from the particle models. These can then be compared to empirical political networks extracted by data mining from newswires. The model shares many properties with reaction-diffusion systems, some of which may be observable in the data. A graph isomorphism technique is developed that permits the comparison of empirical and modeled networks.

     Contemporary political implications follow by inspection.

(For references, please check: http://www.sscnet.ucla.edu/hcs/intercon2002/abstracts)

67.          Soft Science, Hard Problems:

The Future Role of Social Science in Defense Research & Engineering  

Joseph Eash (eashj@ndu.edu)

      The potential for conflict is based in the actions of individuals, organizations and cultures. Understanding why conflict occurs, or persists, is a challenge, and many assume the field either defies study or is not well understood. Unfortunately, this assumption makes it difficult to obtain substantial DOD research funding. However, our ability as a nation to avoid, or mitigate, future conflict hinges not on breaking things better,” but on our ability to better understand these processes.  This paper discusses the role of social science, computational and otherwise, in the future science and technology investments of DOD.

[PANEL PROPOSAL: This panel focuses on the emerging role of the social science theory, and computational social science approaches in particular, in the development of security strategy and defense planning. The analytical communities for military and security organizations are typically composed of scientists from the hard” disciplines. A changing security environment has illustrated the necessity for new problem solving approaches.  The incorporation of social science research, given that the toughest problems faced by security planners are deeply rooted in the actions and perceptions of individuals, institutions, and nations, would greatly aid the community. The four presentations constituting this panel will consider the role of social science theory as enabled by emerging computational methods.]

68.                Nonlinear Dynamics in World Politics    

Williams Mills (billdm@ucia.gov)

Mia Nguyen (mai@ucia.gov)

     Three years ago the Inter-Agency Working Group on Human Dynamics was formed to foster the application of academic thinking on complex systems to state and societal instability that impacts national security.  To achieve this goal, academics need to focus on global social and political issues; the Government needs to learn the concepts of nonlinear dynamics.  We will outline specific examples of how this new way of thinking about complicated global social issues can be applied to address real-world challenges facing the national security community.

[PANEL PROPOSAL: This panel focuses on the emerging role of the social science theory, and computational social science approaches in particular, in the development of security strategy and defense planning. The analytical communities for military and security organizations are typically composed of scientists from the hard” disciplines. A changing security environment has illustrated the necessity for new problem solving approaches.  The incorporation of social science research, given that the toughest problems faced by security planners are deeply rooted in the actions and perceptions of individuals, institutions, and nations, would greatly aid the community. The four presentations constituting this panel will consider the role of social science theory as enabled by emerging computational methods.

69.          Social Science Theory, Operations Research, and Computational Models:

National Security as a Nexus for Inquiry Approaches    

Desmond Saunders-Newton (SaundersNewtond@ndu.edu)

      Current analytical approaches used by the Military Operations Research community are limited in that they rarely incorporate any of the Social Sciences in a rigorous fashion. Theoretical frameworks, instantiated as computational models, provide a ‘transdisciplinary canvas’ upon which to integrate methods and theory toward a common goal, i.e. national security analysis.  Special attention will be paid to the role of these tools in the emerging operational concept called Effects-Based Operations.

[PANEL PROPOSAL: This panel focuses on the emerging role of the social science theory, and computational social science approaches in particular, in the development of security strategy and defense planning. The analytical communities for military and security organizations are typically composed of scientists from the hard” disciplines. A changing security environment has illustrated the necessity for new problem solving approaches.  The incorporation of social science research, given that the toughest problems faced by security planners are deeply rooted in the actions and perceptions of individuals, institutions, and nations, would greatly aid the community. The four presentations constituting this panel will consider the role of social science theory as enabled by emerging computational methods.]

70.                Capturing Complexity in Residential Choice Behavior

Elenna Dugundji (E.Dugundji@FRW.UVA.NL)

     Transportation, land development and residential housing all function within inter-linked economic markets that in the last decennia have witnessed two important trends, namely an increasing differentiation within the sectors in coupling with an increasing individualization in society, and second, an increasing extending of the markets beyond local boundaries.

     Supply-side dynamics in the Netherlands accordingly include changing positions of government & industry agents in the transportation, land use and housing sectors, and adjustment of policy at a macro level, with consequent time-varying institutional and distributive implications impacting the set of opportunities for various individuals & household agents at a micro level. On the other hand, efficient housing, land use and infrastructure responses to (residential) mobility and activity patterns dynamics call for an emerging need for multi-municipality cooperation.

     Demand-side dynamics in the Netherlands include changes in the composition of the population and in the preferences of the individual & household agents thereof. Relationships are supposed between the preferences and circumstances of household & individual agents at a micro level and market trends at a macro level, and between the willingness to change of the various household & individual agents at a micro level and market volatility at a macro level given sufficient vacancy rate.

     The research develops a framework to empirically test and model the complicated and complex system of interactions within these inter-linked markets. The approach builds on the pioneering work of Brewer and Hensher (2000) who coined the term Interactive Agency Choice Experiment, or IACE, to describe a series of stated choice experiments with offers and feedback. Furthermore, the approach draws on advances in Hierarchical Information Integration, or HII,  to handle the multi-dimensional nature of individual & household decisions. Finally, the research aims to examine not only the interaction between agents on the supply-side and demand-side, but additionally the interaction among individual & household agents between each other on the demand-side by drawing on Agent-Based Modeling techniques.

71.                Coordination, Local Interactions and Endogenous Neighborhood Formation

Giorgio Fagiolo (fagiolo@sssup.it)

Paper URL: http://www.sssup.it/~lem/WPLem/files/2001-15_0.pdf

     The details of the process governing the co-evolution of expectations formation, individual choices and interaction structure can crucially affect the long-run social structure emerging in coordination games repeatedly played in large populations. To investigate this issue, we present a model of local coordination in which agents can simultaneously choose both stage-game strategies and the partners with whom to play the game. We consider a population of myopic (adaptive) individuals with fixed geographical locations (i.e. located on a one-dimensional lattice without boundaries) who repeatedly play a pure coordination game with their 'nearest neighbors'. We assume that holding neighbors is costly and that, from time to time, agents are allowed to slightly shrink (or enlarge) the 'radius' of their current neighborhood by maximizing expected net payoffs. We study the behavior of the model in settings characterized by both positive and negative network externalities. In particular, we assume that net individual payoff may initially increase as the number of neighbors increases, but it eventually falls as neighborhood sizes become very large. After having analytically characterized the set of steady states and conditions for convergence, we show that both full coordination and coexistence of conventions may be possible in a steady state. In order to study the long-run average behavior of the dynamical system as parameters change we use extended Montecarlo analyses. In particular, we compute average coordination levels and average neighborhood sizes over large Montecarlo samples. Computer simulations show that the system is able to reach, on average, very high long-run coordination levels, together with small average neighborhood sizes, for a large region of the parameter space. Furthermore, we find that average coordination increases if the unit cost of holding a neighbor decreases and that average coordination in presence of a non-zero (although small) frequency of neighborhood adjustment is higher than if interaction structures were static. Finally, we introduce alternative neighborhood updating rules characterized by different levels of individual myopia and  we explore their effects on aggregate coordination.

72.                Exploring the role of Residential Preference in Segregation with Simulation

David W. Wong (dwong2@gmu.edu)

      A controversial issue in the study of segregation and ethnic diversity is the role of various factors such as institutional, political, socioeconomic and ethnic in creating or perpetuating the levels of racial separation. One argument suggests that the segregation can simply be the result of individual racial preferences on residential choice. Using a simulation, Schelling examined the pattern of residential choice with very simple preference rules. Clark empirically examined the racial preferences of multiethnic groups in Los Angeles. Results of the study partially supported the importance of racial preferences.

      Previous studies, either using simulation or based upon empirical data, were relatively weak in the modeling the spatial aspect of the segregation in the context of racial preference. Schelling model does not have an explicit spatial dimension and the rules were relatively simplistic. Also, the evaluation of segregation level is limited to the use of traditional aspatial measures, which are incapable to handle the so-called checkerboard problem. This paper reports an effort to develop a simulation with explicit geographical dimension in terms of the outcomes and in terms of the residential choice. The model runs on real world data collected for any geographical entity, and is built upon ArcView GIS. Major premises of the model are similar to previous studies arguing that segregation could be an outcome of relatively simple race-based residential preference and population characteristics of the neighborhood can be a significant factor. To evaluate the level of spatial segregation, a set of spatial segregation measures in addition to traditional segregation will be used. Using Washington, DC as an example, this paper demonstrates the operation of this model and presents some of the results.

73.                 Modeling Drug Markets

Michael Agar (magar@anth.umd.edu)

Dwight Wilson

     An anthropological research project, funded by NIH, aims to develop a theory to explain how and why illicit drug epidemics occur. As part of this project, presenters have developed and continue to work with an agent- based adaptive model implemented in SWARM called Drugmarket. The general approach is this: Model parameters are based on the results of ethnographic data collection and analysis in a study of youthful heroin users. Two results of Drugmarket behavior are then particularly interesting. First, the model can be run to construct the attractor space, whereas the actual youth studied ethnographically were only one location in that space. Second, by changing parameters, several "what if" scenarios can be explored that correspond to both historically real and hypothetical future situations.

     In addition to contributing to epidemiology in the drug field, the model raises interesting issues for anthropological research. First, the traditional "idiographic/nomothetic" contrast in anthropology can be addressed-i.e. how does one generalize from a single case while still preserving that case's unique historical features. Second, the model implements the concept of the "natural experiment," where critical comparative analysis of cases can be created in silico as well as sought after in vivo . Third, the exercise raises issues that go well beyond the usual "quantitative/qualitative" debates, in that one must link local knowledge and formal representation in the spirit of Zadeh's fuzzy logic. And fourth, the model offers guidelines for intervention. For example, conditions that lead to a rapid increase in incidence suggest emphasis on treatment, while slow increases shift the emphasis on prevention.

74.          Agent Based Design of Innovative Customer Relationship Management Interactions

David Collings (david.2.collings@bt.com)

     Customer Relationship Management (CRM) comes in many forms allowing a variety of  different interactions with customers. There is increasing use of web based interactions  and, with the anticipated arrival of mobile internet using 3G, potential growth in internet  mediated interactions will be considerable. The multimedia capabilities of these  technologies and the potential for the personalisation of interactions are leading to  innovative approaches intended to create more positive experiences for consumers. The technology also has the potential for creating customer experiences that are shared  amongst members of the customer's social network, such as through viral marketing.  These sorts of collective interaction are especially powerful if there are strong positive  network externalities driving them. In this work we present an agent based model of a  consumer population and consider how the social networks that exist within a population  can play a critical role on the propagation of an interaction experience and how effective  the interaction can be made. The collaborative nature of the interactions can also be  harnessed to "manufacture" specific kinds of links within the population, by encouraging  the sharing of the experience with targeted members of the population, changing the  topology of the social network. We use recent theoretical results analysing the statistical  properties of social networks based on "small world" type topologies to guide the agent  based simulations, to test for critical points and assess the likelihood of cascades and  their magnitude. These characteristics will depend on the nature of the positive network externality induced and the tailoring of the network topology that can be achieved. We  also use ideas from network robustness to consider how the tailored topology can affect  the expected reliability of propagation of the interactions within the population. In this way  we are able to design innovative customer interactions that give high impact, with high probability of successful penetration, maximising the investment made.

75.                Endogenizing the Actions of Artificial Brand Managers

Robert E. Marks, David F. Midgley (david.midgley@insead.edu), Lee G. Cooper

     In an earlier paper we showed how genetic algorithms could be used to evolve strategies in oligopolistic markets characterized by asymmetric competition between brands.4   Our approach was illustrated using empirically based market-share and category-volume models to represent market response to the actions of artificial brand managers.  The artificial agents bred in this environment outperformed the historical actions of brand managers in the real market.

     However, this work had four important limitations.  First, only three of the nine competing brands in the market were modeled; namely the three major brands.  In reality, other brands had significant impact on the observed patterns of competition.  Second, a common population was used for breeding strategies for each of the three players-limiting the ability of the artificial agents to reflect firm differences.  Third, the actions available to the artificial brand managers were constrained to four typical actions derived from historical data, whereas real managers had a much broader menu of actions available to them.  Fourth, only the actions of brand managers were considered-while the real offers available to consumers derived from the joint decisions of brand and retail managers.

     In this paper we address these limitations in three ways.  First, we relax the restrictions on the number of brands and the number of actions available to them.  In doing so, we allow the strategies for each brand to be bred from a different population.  Second, we endogenize the selection of actions by the agents, rather than impose them from an exogenous

expert analysis of historical data.  This we see as the major contribution of the paper.  Third, we discuss early work on jointly breeding artificial agents for both brands and retailers.

4. Midgley, Marks and Cooper, "Breeding Competitive Strategies," Management Science, 43(3), March 1997, 257-275.

76.          Agent Based Modeling of Collaboration and Work Practices Onboard the International Space Station

Alessandro Acquisti (acquisti@sims.berkeley.edu),

Maarten Sierhuis (msierfhuis@mail.arc.nasa.gov), William J. Clancey, Jeffrey M. Bradshaw

     The International Space Station is one the most complex projects ever, with numerous interdependent constraints affecting productivity and crew safety. This requires planning years before crew expeditions, and the use of sophisticated scheduling tools. Human work practices, however, are difficult to study and represent within traditional planning tools.  Expedition ship logs highlight recurring discrepancies between planned crew activities and the reality of onboard life. In addition, scheduling constraints make it hard to 'replan' onboard activities. The need emerges for tools able to model the work activities of astronauts onboard the ISS and their interactions with other humans and artificial agents onboard and on earth.  We present an agent-based model and simulation of the activities and work practices of astronauts onboard the ISS. The model represents "a day in the life" of the ISS crew and is developed in Brahms-an agent-oriented, activity-based language used to model knowledge in situated action and learning in human activities.  Brahms links knowledge-based models of activities with discrete-event simulation and a subsumption architecture. In Brahms, agents' behaviors are organized into activities, inherited from groups. Activities are located in time and space. In our model of the ISS work environment we are able to consider issues like resource availability, human/system interaction, both scheduled and unscheduled activities, and the emergence of work practices aboard the ISS out of procedures developed by engineers and mission controllers.

77.                Emergent Structure in Children's Play Group Formation

William A. Griffin (William.griffin@asu.edu), Willa Cree, M.S., Carol Martin,

Richard Fabes, Laura Hanish

     In the fall of each year, new and returning children between the ages of 3 and 6 are brought together in Arizona State University's Child Development Lab where they eventually settle into pods of stable play partners.  Factors contributing to the formation of these play groups are currently unknown but the dynamics of this evolution appear to have similar characteristics as other structures in social organizations (e.g., agent actions appear to be rule-based).  This setting provides a natural laboratory of studying emergent structure.  Each year, the lab has a slightly different social environment, this difference results primarily from the play group formations, and in turn, play group formations are derived from the stability of who plays with whom.  These groupings and the structure they impose on the lab evolve as the year progresses.  The purpose of the paper is to model this evolution.  We examine the role of individual child attributes on the emergent behavior.  We use four child attributes derived from behavioral observations: Physical attractiveness, Prosocial behavior, Aggression, and Activity level.  Using these attributes along with gender, we simulate the propensity of a child, with a given set of attributes, to play with any other given child.  We include gender in the model because strong gender typing in children of this age group appears to act as a primary deterrent or facilitator in determining reinforcement by peers.  We illustrate the correspondence between simulated play clusters based on child attribute vectors and actual data obtained at several fixed points throughout the school year. Differences at each iteration provide information about the possible strategies children may be using to select play partners.  In turn, we interpret these strategies and resign the simulation; by, for example, differentially weighting attributes.  Preliminary modeling suggests that gender may mediate the effects of social attributes.   

78.          Interest Groups in a Spatial Voting Model

Vjollca Sadiraj (vjollca@fee.uva.nl), Jan Tuinstra (tuinstra@fee.uva.nl),

Frans van Winden (fvwinden@fee.uva.nl)

     We develop agent-based models of spatial voting with endogenously emerging interest groups. In our models political agents do not have complete information and make their decisions based on adaptive instead of optimizing procedures. Political parties use polls to search for policy platforms that maximize the probability of winning an election and the voting decision of voters is influenced by social interaction. Interest groups emerge endogenously from the interaction of bounded rational voters who want to exert political influence. The functions of these interest groups in our models are multiple: firstly, they coordinate voting behavior; secondly, they help transmit information about voter preferences to the political parties through (financial) contributions to opinion polls; and thirdly, they influence election outcomes by imposing conditions on polling. To sharpen our results, we assume that voters are uniformly distributed in a symmetric policy space. Thus, in the basic model the existence of a dominant point - which is the center of the space - is guaranteed. We model the electoral process as a stochastic dynamic process and represent it as an if-then system composed of condition-action relations. Numerical simulations and Markov theory are used to analyze the individual-based stochastic models. Maple V is employed for running simulations and producing animations. Markov theory is used to investigate the effects of various ways of interaction between the interest groups and the political parties on the steady state distribution of policy outcomes. We find that in the presence of interest groups winning platforms, provided they exist, are selected faster than in the basic model. The challenger's probability of winning the election increases if there are interest groups, which appears to be due to an increase in the size of the winning set. The mean-analysis shows that, in the long run, the election outcome of the deterministic Markov models with adaptive agents will be the center of the space only for the basic model. For the model with interest groups, the election outcomes will be either the center of the space (with probability 0.22) or the set of positions closest to the center (with probability 0.78). A further investigation shows, however, that if the size of the population falls short of some threshold voting cycles occur frequently and expand all over the issue space. Thus, for small populations, our model positions itself in the series of models that point at the electoral instability of voting outcomes.

79.          A Mulit-Agent Model of Information Technology Public Goods

                Maksim Tsvetovat (maksim@psychmanager.com), Craig Schreiber, Kathleen Carley

80.                Situativity of Learngin within Groups: Coevolutionary Dynamics Over Time using Kauffman's NK Model

Yu Yuan, Bill McKelvey (mckelvey@anderson.ucla.edu)

     For many firms, producing information and knowledge and enhancing learning capability have become the primary basis of competitive advantage. A review of organizational learning theory identifies two approaches: (1) those that treat symbolic information processing as fundamental to learning, and (2) those that view the situated nature of cognition as fundamental. After noting that the former is inadequate because it focuses primarily on behavioral and cognitive aspects of individual learning, this paper argues the importance of studying learning as interactions among people in the context of their environment. It contributes to organizational learning in three ways. First, it argues that situativity learning theory is to be preferred over traditional behavioral and cognitive learning theories, because it treats organizations as complex adaptive systems rather than mere information processors. Second, it adds rate and nonlinear learning effects. Third, following model-centered epistemology, it uses an agent-based model, in particular a "humanized" version of Kauffman's NK model, to study the situated nature of learning. Based on the simulation results, we suggest seven hypotheses extending situativity theory in new directions. The paper ends with a discussion of possible extensions of the current study to better address key issues in the

situativity of learning.

81.                Asymmetric Power Among Agents in Evolutionary Games

Stephen J. Majeski (majeski@u.washington.edu)

     A set of artificial worlds based upon a group of repeated (2X2) games is constructed and analyzed.  Chicken, Stag, Assurance, and Deadlock games are selected for this analysis because of their substantive importance for international relations and because they provide a range of structural settings where cooperation is both more and less likely to occur than in the much-studied PD.  In addition, asymmetric power among agentsis introduced into these worlds and its consequences for cooperation analyzed.  Such asymmetries are an enduring feature of international systems and indeed most social setting involving collective action.  Asymmetric power is introduced in two ways.  First, agents are differentially rewarded for the joint outcomes of the various games.  Those agents having asymmetric power receive uniformly higher payoffs across all joint outcomes of the relevant game. Second, asymmetric power is introduced by giving some agents the ability to selectively interact with other agents while making interaction mandatory for all other agents.

     The artificial worlds have the following general features.  An explicit spatial dimension is introduced by constructing a set of toroidal worlds (a 20 X 20 grid of cells) consisting initially of 60 agents randomly assigned locations on the grid.  Agents interact with their von Neumann neighbors.  Each agent is represented by a strategy specifying how the agent behaves as it interacts with other agents.  Agent strategies are restricted to those employing just the previous interaction with the other agent(s) to determine current choices. An environmental carrying capacity is incorporated into the artificial worlds by introducing a cost of survival for agents.  Because all individuals die and all social units eventually fall apart, disband, go bankrupt, are taken over, or overrun, agents are assumed to have a limited existence or life span.  Agents reproduce or replicate themselves asexually.  Finally, to give the artificial world some dynamic, strategy mutation is introduced probabilistically at the time of replication.

     Simulation results indicate that anticipated variations in evolutionary outcomes regarding the likelihood of cooperation across these different game structures holds.  Also, the introduction of asymmetric power substantially increases the chances that both cooperative agents survive and cooperative worlds evolve.  This is particularly the case when agents are endowed with the ability to selectively interact with other agents in their world.  In addition, simulation results suggest that exploitative agents do not benefit from asymmetric power.

82.          The Emergence of a European Identity: An Agent-Based Model of International Interaction

Maurits van der Veen (maurits@sas.upenn.edu)

     In recent years numerous analysts have argued that progress in the European integration project is endangered by the weak sense of shared identity evident across the EU polity. Indeed, national publics appear to feel much less of a sense of international community than many leaders would like, leaving the latter to puzzle over the optimal mechanisms to foster a European identity. Agent-based models of the emergence of such an identity provide us with a tool of unmatched flexibility and power for the study of the impact of different factors and policy initiatives.

     Earlier simulation results illustrated the effect of the elimination of borders between European states, while simultaneously underscoring the path dependence and contingency associated with the emergence of a European identity. In this paper, I investigate the implications of international interaction for the emergence of a sense of shared identity. I combine two strands of the modeling literature: studies of the emergence of international cooperation and conflict, and constructivist work on the spread of identities through a population.

     In the model, agents interact in a multi-state world whose borders gradually become more porous. They have some choice in their selection of those they interact with economically, and the geographical range for potential interaction partners grows over time. Agent strategies for cooperation (or defection) may depend on markers of shared identity as well as a limited amount of memory regarding interaction with specific other agents. An important innovation is the fact that identities diffuse not only through contacts with local neighbours, but also as a result of economic interactions with (possibly far-flung) others. The model thus allows us to study the interaction and joint evolution of different scenarios for the evolution of a European identity on the one hand, and international cooperation and conflict on the other hand.

83.          Agent-Based Modeling of Ethnic Mobilization: The Former Yugoslavia Case

Armano Srbljinovic (Armano.Srbljinovic@morh.hr), Drazen Penzar Drazen.(Penzar@morh.hr),

Petra Rodik (prodik@inet.hr), Kruno Kardov (kkardov@filozof.ffzg.hr)

     The term ethnic mobilization, as used in this paper, refers to the process of reviving latent ethnic identities. More specifically, we investigate what has already been described as social situations in

which ethnic roles that used to have less social importance are, under certain circumstances, pushed up towards the highest-importance end of the scale, as it was the case in former Yugoslavia. We are more interested in exploring mechanisms by which this process happens than in exploring causes of such situations.

     In order to enable the explorations, we construct an agent-based model, combining and trying to improve some of the previously used representations of social networks and collective cognitive structures involved in the development of identity shaping and collective mobilization mechanisms.

     The agents are characterized by four basic variables: ethnic membership (blue or red); degree of ethnic mobilization ? the degree to which an agent identifies with its ethnic group; degree of ?civic mobilization? ? the degree to which an agent is committed to the values of civil society (increasing ethnic mobilization is assumed to decrease the civic one); and degree of grievance ? the variable describing agent?s satisfaction with its conditions of life: economic, political, security and others. In addition, each agent possesses its own ?social network? representing the other agents with whom it communicates (family, friends, ?), and whose state variables it can observe. Such observations influence the state of agent?s variables.

     During the simulation agents receive appeals. Appeals are characterized by their source (blue, red or grey ? neutral), and content (increase/decrease mobilization level). Appeals are meant to represent

various means ? media, public meetings, etc. ? by which particular subjects ? state, political organizations, ethnic leaders, non-governmental organizations, etc. ? influence individuals? ethnic/civic orientation. Agents react to received appeals by changing their state variables.

     By varying the initial conditions of particular model settings (i. e. agent constellations, appeal frequencies, types and targets), as well as the parameters of agents? perception and processing mechanisms, we are able to explore the influence of those factors on the efficiency of ethnic mobilization process.

     Preliminary results obtained by executing the model on selected portions of parameter space are presented. Complex model?s behaviors obtained with relatively simple mechanisms, as well as the model?s high sensitivity on random variations in the initial distribution of mobilization intensity and social networks may serve as indicators of inherent limitations on predictability of mobilization processes.

84.          A User-Friendly Agent-Based Modeling Platform for Testing Theories of Political Identity and Political Stability

Ian S. Lustick (ilustick@sas.upenn.edu)

Paper URL: http://www.psych.upenn.edu/sacsec/abir/_private/papers.htm

     PS-I (Political Science-Identity) is an agent-based computer simulation platform originally developed to operationalize, refine, and test competing versions of constructivist identity theory.  Based on an earlier prototype, the ABIR (Agent-Based Identity Repertoire) model, agents with repertoires of identities (or other potentialities) interact in localities of specifiable size and are influenced as well by cross-landscape values attached to particular identities.  These values change over time, thereby simulating conditions in which individuals may express latent identities, or learn to use new identities, because of local pressures toward conformity and/or overall shifts in the relative attractiveness of presenting oneself as attached to one identity or another.   Experiments produce large batches of controlled virtual histories for comparative and statistical analysis.

     PS-I has been designed with two imperatives in mind:  ease of deployment by users who know nothing of computer programming; and systematic correspondence between the algorithms for agent behavior and corroborated theoretical positions in political science and psychology.  The non-technical user-the user with no programming skillsc-can use ABIR, and now PS-I, to build and execute sophisticated models of substantial academic and policy interest. PS-I thus differs very significantly from existing platforms.  We are developing PS-I as an alternative to toolkits such as Starlogo and AgentSheets.  They are user friendly, but generally can only be used to produce a limited number of pre-cooked, 'toy' models and are unlikely to be of actual use to social science researchers focused on theoretically or empirically meaningful problems.  We also see PS-I as an alternative to SWARM, RePast, and ASCAPE, which in principle can be used to produce models of real experimental interest to social scientists, but require users to be proficient in either C languages or Java.

     Users can access the power of PS-I in two ways: through graphical interfaces specialized for particular families of models, or through the powerful scripting language provided.  To use the scripting language to run batches of runs with particular models and record desired in Excel files one needs only to be able to supply simple expressions.  With the graphical interfaces, users can produce models and observe and record data in particular runs simply by understanding the meanings of different parameter settings and typing desired values in dialogue boxes.  The program does the rest.  Data are automatically exported to Excel files. Point and click menus for the production of virtual countries with different cultural complexions and governance patterns are now under development.

     The paper illustrates use of PS-I to study the dynamics of identity politics in a "typical" authoritarian Muslim Middle Eastern country subjected to globalizing pressures, secessionism and conflict in culturally divided societies.

85.                Replicating the Size of Wars:

New Empirical Extensions and Computational Reconstructions of Power Laws

Lars-Erik Cederman (cederman@cfia.harvard.edu), Claudio Coiffi-Revilla

       The size of wars follows a power law, as first discovered by Lewis F. Richardson, indicating a scale-free distribution with fractal dimension. This study builds on earlier studies and makes two contributions, (1) by extending the power law distribution of war size to additional empirical dimensions of warfare (duration, extent, and scope), and (2) by replicating the empirical findings with the GeoSim agent-based simulation. The GeoSim simulation data provides a close replication of the empirical power laws, a necessary requirement for developing a computational theoretical explanation of the size of wars.

86.                Landscapes as Complex Adaptive Systems

David Bennett (david-bennett@uiowa.edu), Richard Aspinall (aspinall@montana.edu)

     In this paper we suggest that rural landscape patterns are the spatial realization of a complex adaptive system (cas) based on interactions among landholders, a variety of policy makers and institutions, and the bio-physical system for which decisions are made.  If this supposition is true then the literature would suggest that it is difficult, if not impossible, to predict the impact of environmental policy with a high degree of certainty. Any given policy can produce many distinctly different responses among landholders, and many different effects, not all of which will be envisioned by policy-makers.  To understand the potential impact of environmental policy we must be willing and able to explore a wide range of scenarios describing policy impact and landholder response.  Each landholder response will be the result of implicit and explicit processes of negotiation in the context of biophysical resources.  Computer simulation is often the only plausible method for the identification and evaluation of competing scenarios.  The modeling process upon which scenarios are based must be able to accommodate negotiation between individuals and institutions, and individual response to policies.  Modeling should also allow scenarios to be based on biophysical systems in a specific geographic context.

      Here we present a framework designed to simulate the negotiation process that leads to a rural agricultural landscape.  The implementation of this framework is based on the technologies of multi-agent systems (MAS) and evolutionary algorithms (EA).  Examples of how this system works are drawn from ongoing research activities in southern Illinois and the Greater Yellowstone Ecosystem. 

87.                Forecasting a Community’s Future via Agent-based Modeling

Robert N. Bernard (robert.bernard@us.pwcglobal.com)

     CommunityViz (www.communityviz.com) is a decision support system for those involved in decision making in local governments, with an emphasis on local land use planning.  Fully integrated into the leading geographic information system, CommunityViz combines three-dimensional imagery, spatial spreadsheet functionality, impact analysis, and forecasting.  The forecasting component of CommunityViz is the Policy Simulator, which forecasts demographic, economic, and land use changes.  Policy Simulator's core method for forecasting is adaptive agent-based modeling, in which simulated people and simulated businesses in the community interact to produce change at a local level.  Policy Simulator is not a toy model -- nor it is a prototype; it uses real world data from almost any community to forecast the community's future.  It is currently being used by 30 communities across the United States for planning their future.  This paper will explain how this agent-based simulation model has been used successfully for real-world applications, by showing examples of forecasts and policy changes implemented in real communities across the United States.  It will also demonstrate how parts of Policy Simulator were designed to serve the needs of the agent-based simulation model through its innovative agent-level data preparation procedures, and its comprehensive data investigation techniques for tracing the life of agents.  Finally, this paper will show how users are able to implement local land use policy changes through the use of flexible policy templates, which change the environment in which the agents interact.

88.          A Multiagent Simulation of Urban Fuel Use and Scarcity in Prehispanic Central Mexico

Martin Biskowski (biskowsk@ucla.edu)

     Fuel scarcity is not just a modern problem. In Central Mexico, obtaining adequate supplies of fuel for domestic use was a substantial problem for prehispanic peoples. Furthermore, many documented activities and changes in prehispanic socioeconomic organization, including the substitution of less preferable secondary fuels and the creation of economies of scale in food preparation, may have been adaptive steps mandated by fuel scarcity. But it is unclear whether fuel scarcity caused these apparently adaptive steps. Many other ongoing social and economic processes may have influenced these changes independently.

     In order to clarify the relevance of fuel scarcity to important evolutionary changes in prehispanic societies, it is helpful to model the prehispanic fuel supply problem. Ethnographic data provide the basis for estimating per capita fuel demand, and the limitation of human-borne transport allows the severity of fuel scarcity to be analyzed in terms of the distance people had to travel to satisfy their domestic fuel needs. The dynamic nature of human-human and human-environmental interactions are modeled using multiagent simulation tools. The resulting simulation models annual fuel demand, the effects of wood extraction on long-term fuel availability, and the consequences of various strategies of fuel conservation and land management around the ancient city of Teotihuacan.

89.                Modeling Adaptive Dynamics in Agricultural Systems

S. Bharwani (sb41@ukc.ac.uk), M.D. Fischer (M.D.Fischer@ukc.ac.uk),

N.S. Ryan (N.S.Ryan@ukc.ac.uk)

     A model was developed to predict the historical adaptive capacity of an agricultural system based on the interaction of multiple and contested perceptions relating to climate change. A multi-agent system was constructed to abstract the system of actors, some of who are receptive and adaptive to the climate change scenario and others who are not. Varying adaptive responses resulted in different degrees of success for individual agents and for the system as a whole. Adaptive responses were classified using criteria proposed by John W. Bennett in 1976 together with frameworks derived from more recent climate impacts literature on adaptive responses to global warming within the agricultural domain.

90.                Heterogeneous Agents and Multiple Equilibria: Implications for Climate Policy Analysis

Stephen J. DeCanio (decanio@econ.ucsb.edu)

     General equilibrium models with heterogeneous agents and complementary goods may exhibit multiple equilibria.  These equilibria can differ in all important respects, including allocations, prices, and the distribution of income.  The multiple equilibrium problem can arise in single-period models but is particularly acute in multi-period models of the type most relevant for the analysis of climate change.  This paper examines some of these cases, and shows that different equilibria can correspond to different "points of view in time" that may favor different generations.

91.          Multi Agent Simulation of a Village Community Sharing a Solar Power Plant

Wernher Brucks (brucks@sozpsy.unizh.ch)

       Electrification of remote areas with solar energy sometimes leads to technical problems (e.g. due to wrong sizing of the system) as well as social conflicts (e.g. due to power limitations). To study interactions between these two sources of trouble a simulation model has been established that consists of a technical sub-model (corresponding to a shared solar power plant) and a social sub-model (corresponding to the using community). Based on synthetic weather data and social survey data, simulation trials may reveal potential problems on both the technical and on the community side. Then, in further simulation trials, the size or design of the solar power plant can be modified and social measures applied. The goal is to simulate optimal management of the plant by a given community and, from that, to derive important considerations for the installation and operation of real solar power plants in such communities.  The focus of this article is on the social sub-model. It is based on a psychologically wellfounded model of human resource use. Each Agent represents a household of the village and acts according to this psychological model, but each agent  is also equipped with different properties relevant to resource use that are beeing asked in face-to-face interviews. The major part of the energy use of a household is made up of basic needs like illumination for example. The psychological model modifies the strength of these basic needs according to the psychological characteristics of the particular household. Especially when energy is scarce (due to the weather or due to the behavior of others) it becomes apparent that people react very differently. The simulation model is being tested by data from real communities using a shared solar power plant.

92.                Persuasion Processes in Populations:

Agent-based Simulation Based on a Social Psychological Theory

Hans-Joachim Mosler (mosler@sozpsy.unizh.ch)

     The research to be presented designed an agent-based simulation based upon the social psychological theory of the Elaboration Likelihood Model (ELM). The central statement of this theory is that a person changes his or her attitude in dependency upon the intensity of information processing, that is, dependent upon whether the person is capable of and motivated to think about a topic. If processing intensity, or elaboration likelihood, is high, the effect of the persuasion will depend on the quality of the arguments presented. If processing intensity is low, peripheral cues gain more weight (such as credibility of the information source, length and complexity of the message). In the simulation, agents function according to the rules derived from this theory. They are given varying values of the ELM model variables, and they influence each other mutually.  To validate the simulation, we investigated whether the simulated agents reacted in just the same way that real persons did in experiments that are reported in the literature.

     In order to simulate populations, 10,000 agents were assigned values for the variables in the model on the basis of frequency distributions, with mean value and standard deviation. Moreover, we constructed networks in the populations, whereby the agents belonged to groups that also had contact to other groups. In constructing networks, the number of contact persons an agent has, the size of the group, and the links among groups can all be varied.

     The variously constructed populations can now be investigated with regard to the persuasive effects of different types of information campaigns or action campaigns with multiplicators. Information campaigns function by exerting an influence on a certain percentage of agents in the population through arguments of a certain quality and through peripheral cues. Multiplicators make use of a flexible persuasion strategy, by using arguments or peripheral cues in dependency upon the processing intensity of the agents they must influence.

     The findings of the population experiments reveal that for different populations having different networks, the persuasion strategy must be chosen very carefully, if it is to succeed.

93.                Empirical Test of a Psychologically Based Simulation of Collective Action

Robert Tobias (rtobias@sozpsy.unizh.ch)

     The problem of organizing collective action campaigns is tackled with a multiagent simulation, in which the actor‚s decision to participate is modeled on the basis of an expansion of the social psychological Theory of Planned Behavior. The variables entering into the model of the individual‚s

decision to participate are attitude, return, subjective social norm, and perceived behavioral control. With these four variables, various social scientific theories get integrated (Elaboration Likelihood Model, Theory of Planned Behavior, Rational Choice Theory). The theories allow conceptualization of interactions among individuals and the effect of activators, seeking to recruit others to participate in a collective action can be studied. Simulations of a population of 10,000 different agents revealed the influence of several factors, though, only one or two examples can be shown in the presentation.

     The focus of the presentation will be on the empirical test of this computer model using data that was gathered during a collective action campaign in a Swiss community. Based on this data, the behavior (participation or non-participation) of each individual in the sample was calculated and the simulation results compared with the real behavior. In more than 85% of the cases, the calculated behavior equals the empirical data.

     In a second step, the sample data was extrapolated to simulate the entire community. The simulated diffusion of participation appeared to be similar to the real dynamics of the collective action campaign in the community. Thus, the model could be used to evaluate the organization of the collective action campaign.

     Though the model has to be developed further in many regards, a lot of important aspects for planning collective action campaigns can already be investigated adequately with it.

94.          Task Allocation in a Self-Organizing Social System: A Multi-Disciplinary Approach

Kees Zoethout, Wander Jager, Eric Molleman

w.jager@bdk.rug.nl

     The proposed paper deals with the relation between task complexity and social processes of task allocation. The complexity of a task requires a certain task allocation, which will be reflected in the resulting social structure. Tasks with a high degree of repetitiveness, i.e. simple tasks, demand other social structures than complex tasks. In order to understand the emergence of these social structures, we will relate social processes of task allocation to the individual cognitive architecture (i.e. memory and skills). Subsequently we will describe learning processes at both the individual and the social level: At the individual level, learning refers to changes of expertise, motivation, and skill use as a result of social processes of task allocation and task performance. At the social level, learning implies strengthening or weakening of social relationships as a result of both individual - and task characteristics.

     These processes can be studied by making use of multi-agent simulation.  In the field of multi-agent modeling we can distinguish two different approaches: the Îcognitive modeling approachÌ and the Îsocial modeling approachÌ. But, because we want to study both the individual cognitive processes and the social processes, we have integrated theories and models from different fields of psychology. This integration has yet lead to a model that combines Îcognitive plausibilityÌ with Îsocial plausibilityÌ but does not necessarily fit in either one of these approaches. As we will argue, the combination of theories and models from different fields into a single agent architecture may easily lead to results that are not comprehensible. Complex non-linear relations between too much parameters may cause this incomprehensibility. Therefore, before combining both approaches, we have simplified them to prevent us from the risk of incomprehensible outcomes.

      In the paper we will describe the model that resulted from this approach. We will discuss the different theories and models underlying it give an outline of future sequence of computer simulation studies.

95.          Sexual Selection of Co-operation

M. Afzal Upal (upal@iet.com)

     Simulations of iterated prisoner's dilemma games have been widely used in social sciences to study the evolution of co-operative behavior. Axelrod's (1994) pioneering work showed that seemingly co-operative strategies such as Tit-forTat (TFT) can do better than selfish strategies such as always defect in a wide range of environments.  TFT strategists start out co-operating and then do what the other player did on the previous move.  Further work by Axelrod and Hamilton (1987) suggested that co-operative strategies such as TFT can automatically arise in a population of individuals through evolution. However, subsequent work has shown that neither TFT, nor any other pure or mixed strategy, is evolutionarily stable. This has lead researchers to consider other factors that can enhance the evolution of co-operative behavior.  Sexual selection is one such mechanism (Miller 2000).  In this paper we report on the results of the  simulations that we performed to test the hypothesis that female preference for mating with co-operating males can enhance the evolution of co-operative behavior among males.

     The model involved building a heterosexual population of 100 agents.  The sex of an agent was randomly chosen to be male or female.   Game playing strategies of the first generation of agents were also randomly chosen.  Two players were randomly chosen to play R rounds of prisonerís dilemma game.  After the game playing rounds, players of opposite sex were allowed to mate and reproduce children.  Strategies of the children were produced by "crossing over" the strategies of their parents.  The only difference between a male and female agent modeled was the cost of reproduction.  We ran experiments with various values of male and female cost of reproduction and with two mate selection strategies; random and preference for the most co-operative male.  The evaluation metric was the difference in the emergent strategies of 100th generation.  Our preliminary results show that under a wide variety of conditions, sexual selection does indeed lead to significantly higher proportions of co-operative strategies.

96.          Co-evolution of Personality and  the Environment

Wander Jager (w.jager@bdk.rug.nl), Marco A. Janssen (m.janssen@feweb.vu.nl)

     In this paper we focus on the cognitive costs that are involved in  more reasoned decision-making strategies. We argue that a  lower investment of cognitive effort, e.g., by satisficing or  imitating the behaviour of others, may be beneficial both for the  individual as for the sustainability of the population as a whole.  We further argue that the most effective distribution of decision- strategies will be related to the stability of the environment  people live in. Hence personality factors that determine the  preference for a certain distribution are subject to evolutionary  pressures. Starting from a multi-theoretical perspective on  behaviour we have developed a multi-agent model. In this model  several decision-making strategies have been formalised. These  strategies differ with respect to the cognitive effort they require  and their individual versus social orientation. Experiments with  this multi-agent model show that the type of personality that is  most frequent in a population depends on the stability of the  environment. In unstable environment, agents with higher  aspiration levels and lower uncertainty tolerance have an  evolutionary benefit. Moreover, the results show that  sustainability can be reached when agents with particular type of  personalities evolve. The model experiments also demonstrate  that evolutionary pressures favour a mix of cognitive strategies.  Finally we demonstrate that an unstable environment favours the  development of a smaller population investing more cognitive  effort in their decision making process.

97.          Power in Non-negotiated Exchange Networks

Phillip Bonacich (Bonacich@soc.ucla.edu)

    Research on how power develops within exchange networks is a lively and exciting topic within sociological social psychology.  Almost all the work has involved the power imbalances that develop when actors are in unequal bargaining positions within networks of bargaining opportunities.  Models (one of which is mine) have been developed to predict exactly which positions will have power.  Only one researcher, Linda Molm, has experimented with networks in which actors do not negotiate over the distribution of rewards from transactions but instead have opportunities to distribute rewards to each other.  In game-theoretic language, she?s initiated the study of non-cooperative games within networks while previous research used cooperative games.  Molm has experimentally examined only two networks, and as yet no one has developed a general theory predicting the location of powerful positions in any network.

    Computer simulations, using Mathematica, suggest that the distribution of power among positions will be different in negotiated (cooperative) and non-negotiated (non-cooperative) networks.  In the simulation actors are programmed to reciprocate with a probability proportional to the value of the gifts they have received.  

    The simulations also suggest that whether or not the actors have complete information about the network and the pattern of giving should also affect the distribution of power.  The outcomes that result when actors have just local knowledge about what they have received and simply reciprocate do not form a Nash equilibrium.  Knowledgeable actors can learn that they are better off giving less frequently to actors who are dependent on them for rewards and more frequently to actors who are less dependent.  

98.          Multi Agent Systems and the Micro-Macro Link

R. Keith Sawyer (keith@keithsawyer.com)

     My talk draws on contemporary sociological theories of the micro-macro link: the relation between individual action and emergent social structure.  In this paper I explore how MAS can inform this area of sociological theory, and I show how sociological theory can inform the design and development of artificial societies.  I argue that MAS have attained a level of maturity where they can be useful tools for sociologists.  In addition to this methodological claim, I show how MAS provide new perspectives on contemporary discussions of the micro-macro link in sociological theory, focusing on how macrosocial structures emerge from collective action and also on how individual actions are constrained by these emergent phenomena.  

     Although I am enthusiastic about the potential for MAS to inform sociological theory, I also take a critical approach, noting several areas where MAS could become more sociologically informed.  I use sociological theory to develop a new set of criticisms of artificial societies.  In particular, sociological theories of the relation between collective action and emergent social structure have attained a high level of sophistication, in many cases addressing issues of direct relevance to artificial society designers.  I argue that MAS have not yet attempted to simulate some of the most fundamental elements of the micro-macro relation, and I suggest several future directions for a more sociologically informed study of artificial societies.  

99.          A Landscape Theory of Aggregation and Social Networks:

Substantive and Methodological Concerns

Timothy Tappe (ttappe@uwyo.edu)

     Axelrod and Bennet debuted a Landscape Theory of Aggregation in 1993, which predicts how agents, who are myopic in their decisions making and incremental in their actions, will aggregate into certain configurations based on their propensities to be attracted to or repelled from one another, or, in other words, to align. These propensities are modeled as a function of the relative size” and power” of each agent in the collective and form an energy landscape as agents seek to minimize their frustration,” and, consequently, lower their energy by switching alliances. This paper attempts to address both methodological and substantive concerns surrounding the accuracy and predictive viability of this model when used to model and make predictions about the social networks of college-age males. Further, it attempts to put forth an alternative model as well as assess the impact of this and other models upon research in the social sciences.

100.        Inverse Simulation and Genetics-Based Validation for Social Interaction Analysis via Multiagents

Takao Terano (terano@gssm.otsuka.tsukuba.ac.jp), Setsuya Kurahashi

     The validation of agent-based simulation is quite important to convince the results to various audiences.  To address the issues, we have developed a new agent-based simulation environment using Genetic Algorithms (GAs).  The basic principles are summarized as follows:  (1) Set various agent parameters as an individual of GAs; (2) Run the agent-based model in parallel so that the runs form the population of GAs; (3) Based on a given criteria or an objective function, each simulation result is evaluated; (4) Genetic operators are then applied to generate the other set of agent parameters; and (5) After the convergence or when we have 'desired' results, the variances among the parameters are evaluated to validate the results in the sense of the sensitivity analyses.

     TRURL is such a simulation environment, which evolves artificial worlds of multiagents to socially interact with each other.  The agents in TRURL solve simple multi-attribute decision problems via the message communication  among them. The micro-level agent activities are determined by both predetermined and acquired parameters.  The former parameters have constant values during one simulation epoch, however, the latter parameters change during the interactions.  TRURL utilizes the above principles to evolve the societies by changing the predetermined parameters to optimize macro-level socio-metric measures, which can be observed in such real societies as e-mail oriented organizations and electronic commerce markets.  Thus, using  TRURL, we automatically tune the parameters up and validate the results from both micro- and macro-level phenomena grounding on the activities of real worlds.

101.                Interdisciplinary Research (and some recent results) on the Small-World Problem

Duncan Watts (djw24@Columbia.edu)

     Traditionally, social science has been the beneficiary of concepts and techniques derived from the natural sciences, such as mathematics and statistics, physics, and biology, but rarely have they returned the favor. The field of network analysis, however, is showing some promise of becoming an exception to the rule.  While many of the early ideas of social network analysis (density, centrality, random-biased networks, block-modeling), had origins in mathematics and solid state physics, and a number of recent advances in network research (modeling of partly-random networks, non-normal degree distributions, dynamical evolution of networks, and network models of contagion) have come from outside sociology, it is also true that ideas and empirical evidence from sociology, particularly with regard to social networks, have begun to penetrate other disciplines. In this talk, I argue that social network analysis, properly construed, provides us a rare opportunity not only to adopt ideas from the natural sciences to address sociological problems, but to help natural scientists apply sociological ideas to their own problems.  As an example, I follow the history of a single problem-The Small World Phenomenon-from its origins as an exclusively sociological question to a veritable cottage industry in statistical physics and other fields.  Furthermore, while a few of the general features of the original research are well known, I show that the small-world phenomenon is still yielding insights which shed light on emerging problems such as efficient search algorithms for peer-to-peer networks

102.                Selection and Transformation in the Innovation Process: A Genetic Algorithm Modeling.

Manuel Cartier (manuel.cartier@dauphine.fr)

     What is the impact of selection and adaptation on the generation of innovations in corporations? Many researches argue that one of the Darwinian process and the Lamarkian process is more present in evolution of population. But in an intraorganizational ecological point of view, these processes does not come from the environment but from the firm. New questions could be : which one of these processes in the most efficient to create viable and performing projects? Does a higher level of selection or adaptation always lead to better performance? Are they competing path of evolution or complementary ones?

      We think theses issues depend on many factors linked to complexity theory and complex adaptive systems: number of competing projects, initial diversity in projects characteristics, exchanges between platforms, environment characteristics,…

In this contingent approach, gaps emerge from the crossing of evolutionary theory and product development literature.

     - Do firms need to stop failing projects in complex technological environment, allowing to reallocate funds to alternative projects?

    - Does cooperation in product development can allow the decrease of research efforts?

    - Does diversity increase the effect of selection on performance?

    - Does internal selection prevent from getting stuck into a pattern of low performance in rugged landscapes?

    We'll use an agent modeling methodology based on genetic algorithms, running on MATLAB 6.1 combined with a GEAT Toolbox. Global behavior at the organization level emerges in simulation from basic interactions between individual projects.

Genetic Algorithms allow the use of inputs such as diversity, number of agents, search rules or specific landscapes (as in the NK model) but also crucial evolutionary concepts: selection rate on each generation and crossovers between successful agents.

Model Robustness' to changes in basic variables will be tested. The validity will be approached by analytical adequacy (comparison to admitted theories) whereas ontological adequacy will be let for future contributions.

     Our computational model tends to be a useful building block in theory about the role of internal selection and transformation in firm evolution and to contribute to agent-based modeling in organization science.

103.                Learning from the Competition: External Spillovers  and the Organizational Structures of Competing Multi-Unit Firms

Myong-Hun Chang (m.chang@csuohio.edu\ulnone), Joseph E. Harrington, Jr. (joe.harrington@jhu.edu)

     How does the existence of the opportunity for \i external\i0  spillovers (learning from the competition) influence the way \i internal\i0  spillovers (intra-organizational mutual learning) are coordinated in multi-unit organizations?  We investigate this question by constructing a computational model of competing retail chains engaged in long-term rivalry in multiple markets.  The setting is one in which each market is served by several chains.  Consumers engage in the search to find the store that best fits their needs, while chains compete by improving their stores\rquote  practices over time.  The chains are modeled as multi-agent adaptive systems composed of boundedly rational store managers who engage in adaptive learning over time:  Each store develops new ideas on their own (innovation) but also learns about new practices of other stores in their chain (via headquarters).  As the potential sources of innovative ideas are distributed among multiple agents (store managers), rather than concentrated at the central headquarters, it then becomes critical that the process of innovation and knowledge utilization be coordinated in a way that permits the firm to compete effectively against its rival in local markets.  We consider two organizational structures under which the newly generated knowledge may be communicated and utilized within an organization:  1) under centralization the corporate staff has the ultimate authority to adopt and mandate new practices, while 2) under decentralization store managers have the freedom to make their own adoption decisions.\f0\fs22  In this setting, we introduce the possibility that the various units as well as the headquarters of competing chains may learn from one another through external spillovers in local markets.  Our objective is to examine the effects such spillovers have on the relative performance of centralized and decentralized organizational structures.  The simulation results suggest that the existence of external spillovers benefits decentralization initially, but ultimately favors centralization in the long run.

104.        A Multi-theoretical Multi-level Multi-agent Computational Model of the Co-evolution of Communication and Knowledge Networks

Noshir Contractor (nosh@uiuc.edu)

      The agent-based approach to the study of complex systems is especially well suited to understand knowledge networks among agents where some of the agents may be humans, while others may be non-human, such as knowledge repositories. This paper addresses a general, but increasingly relevant, question: Under what conditions, are individuals more likely to seek information they need from (or provide information they posses to) other individuals as opposed to knowledge repositories? The theories of Transactive Memory and Public Goods both seek to describe the conditions under which agents share (retrieve or allocate) information in order to accomplish a collective task.  The Theory of Transactive Memory offers a set of peer-to-peer mechanisms to explain these processes in terms of an agent's perception of others' knowledge (directory updating and expertise recognition). As such it offers an explanation primarily at the dyadic level of analysis. Public Goods theory describes, in terms of agents‚ individual costs and benefits, the conditions under which as a collective they are more likely to share information with others by publishing to, and retrieving from, communal knowledge repositories. Public Goods theory therefore offers an explanation that is based at the dyadic and global levels of analyses. Implementing and „docking‰ computational models based on these two theories, which operate at multiple levels, offer emergent and empirically verifiable hypotheses in response to the question posed above.

105.        Design and coordination in complex systems

Sendil K. Ethiraj s(ethiraj@wharton.upenn.edu),

Daniel Levinthal (levinthal@management.wharton.upenn.edu)

     The problem of designing, managing, and coordinating the efforts of different parts of large-scale complex systems is central to the management and organizations literature. Following Simonís (1962) exposition on the architecture of complexity, there is a resurging research interest in modular design principles as a potential solution to the design and coordination problem. While there is significant attention directed at understanding the how and why modular design structures help manage complexity, there is little understanding on whether and how one achieves an appropriate modularization and how this interacts with the evolution of the complex system. The centrality of this question can hardly be denied since much of the literature on modularity rests on the fundamental assumption that an appropriate modularization of a complex system can be achieved. This paper represents a first attempt to begin addressing this question. We set up a model of a complex system whose true underlying structure, following Simon (1962), is nearly decomposable. The complex system is represented as an N-dimensional hypercube, where each of the N policy choices of the system takes a value of 1 or 0, resulting in 2^N possible configurations of the complex system. The model starts with a randomly generated system with a random number of modules and a random mapping of policy choices to modules. The key task of the designer(s) is to discover the real, underlying structure, i.e., correct number of modules and correct mapping of policy choices to modules. Our results indicate that boundedly rational designers, with fairly simple rules guiding their behavior are able to converge on the right modularization. Moreover, in cases where the underlying system is nearly decomposable, the solution time increases only linearly in problem size. We also find that there is some degree of path dependence in success at discovering the right modularization. Under-modularized starting points seem to discover the right modularization faster than all other starting points. We also investigated the interaction of the process of module evolution (i.e., redrawing of module boundaries) with the process of module innovation (i.e., local search and recombination). Not surprisingly, we find that module innovation is increasing (both in rate and final asymptote) in the ability to discover the right modularization. A more surprising result is that the processes of local search and recombination generate performance improvement even when the system has not achieved the correct modularization. This suggests that achieving the correct modularization is not a necessary pre-condition for realizing the benefits of modularity. However, a necessary condition for performance enhancing recombination is that two or more systems should share the same (correct or incorrect) modularization. In the absence of shared modular structures between systems there is no value to recombination. On the other hand, when two or more systems share the same (incorrect) modularization, the performance gains from recombination is significantly higher than the performance of dissimilar systems giving them the evolutionary advantage in the face of selection pressures. This raises the interesting possibility that recombination, rather than being an outcome after the emergence of dominant designs, is itself an engine that facilitates the emergence of dominant designs.

106.                Strategic Synchronization among Spatially Distributed Agents

Alessandro Lomi (alx@economia.unibo.it), Erik Larsen (e.r.larsen@city.ac.uk)

      During the last two decades of organization studies a general agreement emerged across competing theoretical perspectives on the proposition that environments of organizations are other organizations. However, the theoretical, methodological and substantive consequences of this proposition for the evolution of organizational fields remain contentious. For example, while organizational environments are recurrently described in terms of dependence relations among organizations, in empirical studies environments continue to be represented as lists of independent variables.

       We see two main difficulties behind the fact that theoretical discourse on the core issue of organizational environments has not progressed much beyond a generic agreement on this crucial issue. First, the constructionist view of organizational environments is plausible but unhelpful without a well developed system of assumptions about the dynamics of connectivity within organizational populations, i.e., about how exactly individual members of organizational populations interact. This is a problem because organizational populations are implicitly represented as complete graphs, in which every member of a population interacts with any other. But connectivity within organizational populations is typically imperfect because processes of disruptive selection, differentiation, and cooperation give rise to local structures that tend to isolate networks of interacting organizations from more general population-level processes.

        Second, organizations are composite agents with spatial extension. As a consequence, the most fundamental organizational processes can be understood only with reference to the distribution and location of organizations in space. Organizational space may have a more or less pronounced physical connotation. Connectivity and therefore proximity measures can be derived both for organizations distributed in geographical space, as well as for organizations distributed in spaces generated by an almost unlimited combination of attributes. Because imperfect connectivity induces a variety of partially isolated local structures organizational populations and fields can be viewed as emerging from the dynamics of multiple networks of local interaction among apparently independent organizations.  This is a problem because local structures introduce spatial heterogeneity in organizational populations and erode the boundaries around individual organizations complicating the problem of identifying appropriate units of selection in the organizational world.

        Together, the presence of imperfect connectivity in organizational fields and the persistence of local structures pose serious difficulties for theoretical narratives based on a notion of organizational environments as well-stirred reaction vessels through which processes of competition, imitation, learning, influence and cooperation emerge, propagate, and eventually die out.  How can these relational processes embedded in strictly local networks of interaction induce and sustain aggregate regularities at the field level?

        To address these concerns, in this paper we propose an agent-based model of organizational evolution in which spatially situated agents respond only to the behavior of a small number of other agents in their neighborhood. According to the model, individual agents in a population  (called firms”) face a choice problem defined over a set of alternative courses of actions (called strategies”), which correspond to an uncertain outcome (called profit”). Agents are distributed on a square two-dimensional lattice of sites, and are endowed with a limited memory which takes the form of a string of information on their past performance. In each period, agents compare their experience with the most recent performance of a limited numbers of neighboring agents and choose the strategy associated with the most favorable outcome that can be observed. Because the actual profit is not a direct consequence of individual strategies but depends on the number of other agents that have chosen the same strategy, over time agents update and adjust their expectations according to their experience.

       In the context of this model, we (i) Explore the evolutionary conditions that favor the emergence and persistence of communities, or sets of agents that synchronize their strategies; (ii) Document the multiple connections existing between individual strategies and collective outcomes, (iii) Analyze sequences of individual choices to search for evidence of strategic synchronization among apparently independent filed members, and (iv) Compare the field-level performance induced by various forms of local interaction with the performance computable under conditions of Nash equilibrium.

       Finally, we show that when spatial elements are introduced explicitly in abstract representations of organizational fields an increase in the amount of information available to individual agents leads to an increase in global instability and to a lower filed-level performance. This result is the opposite of what is typically observed when organizational systems are represented as comprised of dimensionless non-spatial entities. Consequently our model invites discussion on the comparative role of global information and local networks in the structuring of organizational fields and other decentralized systems with local interaction like, for example, markets.

107.        A Positive Theory of Emergence for Multi-Agent Systems

Rob Axtell (raxtell@brookings.edu)

Paper URL: www.brookings.edu/dynamics; www.brookings.edu/scholars/raxtell.htm

     Certain usage of the term 'emergence' in the multi-agent systems (MAS) and complex adaptive systems (CAS) literatures is formalized. In particular, descriptions of the behavior of such systems are often given at multiple levels: at the level of the individual components of the system (micro-level) as well as at some higher level involving many system components (macro-level).  Emergent properties are ones that obtain at one level while being absent at the other.  A formal way to relate multi-level descriptions of a phenomenon is given by the theory of model aggregation.  The concept of an emergent property of a system falls out naturally from considerations of model aggregation, and is distinguished from simple resultant properties.  Two distinct types of emergent properties are identified.  Examples of emergent properties in social systems are given.  In particular, it is argued that particular kinds of social structures (e.g., social norms, markets, firms, institutions) are best understood as emergent.

108.        Models of Heterogenous Agents

Scott Page (spage@umich.edu)

     One of the often stated advantages of agent based models is that they allow us to include heterogeneity at the level of agents in ways that would be intractable using other methodologies.  In my talk,  I will discuss some of the opportunities, implications, and complications from these richer descriptions of agents focusing on heterogeneities of world views and toolboxes.