Artificial Life Simulation of the Textile/Apparel Marketplace: An Innovative Approach to Strategizing About Evolving Markets Investigators: E.L. Brannon, P.V. Ulrich, L.J. Anderson, T. Marshall (Auburn), A. Donaldson (NCSU) Research Associates: S. Thommesen and N. Terase (Auburn) Contact Person: E.L. Brannon, 308 Spidle Hall, Auburn University, AL 36849 voice: 334 844-6457, fax: 344 844-1340, ebrannon@humsci.auburn.edu Project Goal Investigate the feasibility of using artificial life (A-Life) simulation techniques to model the textile/apparel marketplace by: n developing A-Life simulations sufficient to demonstrate emerging behaviors that are interpretable within theoretical frameworks, n assessing A-Life simulation as method of inquiry by using the simulation as a laboratory for computer-based experiments. n assessing A-Life simulation as an aid in strategizing about evolving markets. n fostering a three-way network between complex systems scholars at the Santa Fe Institute, researchers studying marketplace phenomena, and industry executives managing change in a non-equilibrium environment. Abstract A-Life techniques allow researchers to build rigorous, replicable simulations of non-equilibrium systems. The simulations are laboratories where the behavior of adaptive agents can be observed across time, rule sets can be recalibrated to investigate different scenarios, and emerging patterns of behavior can be identified and interpreted. The aim of A-Life simulation, like all computer modeling, is to link the micro behavior of the simulated system with the macro behavior of society. In this project, the textile/apparel marketplace is viewed as a complex, adaptive, self-organizing system—that is, the actions of agents (consumers, firms, etc.) create a system in rolling equilibrium. Through the agents’ mutual accommodation and mutual rivalry, new structures and behavior patterns emerge continuously. Two simulation prototypes— InfoSUMERS, a simulation of diffusion of innovation, and SPHERE OF INFLUENCE, a simulation of consumer/supplier interactions—have been designed, programmed, tested, and revised. Two other simulations are in the design and programming stages. The process of simulation development includes stages for design and programming, verification, validation, and sensitivity analysis. A network of simulation developers, textile/apparel scholars, and industry executives assist the core team in developing A-Life simulations for theory building, generating hypotheses for empirical research, strategizing about evolving markets, and conducting computer experiments. A-Life Simulation as a Method of Inquiry Theorists and researchers (Levy, 1992; Waldrop, 1992; Holland, 1995) working in the field of selforganizing systems are forging an alternative to the traditional linear, reductionist approach to inquiry. The new paradigm is based on a basic insight: a system of agents following simple rules can exhibit complex behaviors. Only recently have tools like graphic workstations made it possible to study adaptive agents through simulation. As a new method of inquiry, A-Life (Langton, 1989) techniques allow researchers to build rigorous, replicable computer-based experiments of non-equilibrium systems. In this new science, computer simulation acts as both theory and experiment. Developing simulations requires deep understanding of the inputs, outputs, and behaviors within a system—a theory of the system. For complex, self-organizing systems, the simulation is a laboratory where agents’ actions can be observed across time, rule sets can be recalibrated to investigate different scenarios, and emerging patterns of behavior can be identified and interpreted. Unlike other modeling approaches, no agent operates with perfect global knowledge. Instead agents in A-Life simulations operate in their world with incomplete knowledge using heuristics rather than pure rationality, much like consumers and firms in the real world. At each time step, each agent assesses its own condition in comparison to its preference set, evaluates the environment within its sight range, and determines its behavior in the next time step. There is no coordinating authority to dictate the system’s behavior. Rather the collective behavior of all agents determines the pattern and environment at each successive time step. The method is particularly valuable for phenomena where the actions occur over a long time span, the system is inaccessible for observation, or the dynamics of a system are subtle. This method of inquiry can even be used to recreate an historical system and play what-if games involving alternative scenarios. The aim of A-Life simulation, like all computer modeling, is to link the micro behavior of the simulated system with the macro behavior of society. An investigator using A-Life simulation begins with a theory of the system gleaned from observation and previous research. The theory is translated into precise specification of system parameters, agent variables, rule sets, and timetables and programmed in the simulation software. Programming is not a trivial process because decisions about the how to display data, the stages of initializing the simulation, and selection of a suitable pseudo-random number generator can effect simulation validity. Once a prototype of the simulation is complete, the next stage is verification and validation. In verifying the simulation, investigators run trials of the model exploring combinations of parameters and variables to determine if the theory of the system has been captured in the simulation design and that the internal logic of the model is operating in a consistent and reasonable way. Verification may also include evaluation of the model by experts in the field. In the validation stage, results from the model are matched against existing data. Problems at this stage lead to revision of the system specifications (Figure 1). I95-A20/3 Figure 1 ARTIFICIAL LIFE SIMULATION Steps in Building Rigorous, Replicable Simulations EMPIRICAL FINDINGS AGENT THEORY RULES 1 -------2 -------3 -------- SIMULATION DESIGN & PROGRAMMING EMERGENT BEHAVIOR EXPERIMENTATION & STRATEGIZING VERIFICATION VALIDATION SENSITIVITY ANALYSIS If the purpose of the project is exploratory—for theory building, generating hypotheses for empirical testing, or strategizing—this level of development may be sufficient to produce a useful simulation, one that simplifies a phenomenon so that it can be understood and studied while retaining real-world veracity. Theory building may be the most immediate use of A-Life simulation as investigators oscillate between specifying theoretical parameters and variables and exploring the behavior of the model under various scenarios in an inductive approach to science (Seror, 1994). If the purpose of a project is hypothesis testing or prediction, then a model that is plausible against existing data and persuasive with experts is not sufficient. For methodological completeness, investigators must establish experimental validity and perform a sensitivity analysis (Doran & Gilbert, 1994). Experimental validation means designing experiments that fully explore all possible combinations of parameters and variables to adequately sample the model’s behavior. Unlike laboratory experiments where standard procedures have been established, procedures for computer experiments are still being developed. Questions remain about how many time steps and how many runs of the simulation represent adequate sampling, what data should be captured and how, and what documentation is sufficient for replication by other scientists. Interpretation of results from computer experiments can be problematical, especially if findings are counterintuitive, but sensitivity analysis can bolster confidence in the model. In a sensitivity analysis, the investigator examines assumptions underlying the model to identify their relative significance to the functioning of the simulation. A-Life simulation is a relatively new technique. From an applications standpoint, A-Life simulations scenarios can be used to stimulate strategizing among industry experts. As a method of inquiry A-Life simulations can be useful for theory building, generating hypotheses for empirical testing, and for hypothesis testing. When the technique is more completely developed, it may be possible to use A-Life techniques to build a “flight simulator” for the exploration of rare or novel phenomena (Figure 2). A-Life Simulation for the Textile/Apparel Marketplace There are a number of tools available to researchers interested in simulation where a population of interacting agents are linked in a network of connections. The problem is that such models tend to settle into a self-consistent pattern that fails to provide information about the constant adaptation present in real world systems. With co-evolutionary models agents must adapt to each other and to changes in their world using incomplete or inconsistent information and heuristics rather than perfect rationality—agents are not able to optimize their fitness or utility. The resulting models are a class of behaviors that exhibit perpetual novelty where agents adapt to niches of opportunity as they occur. Such models have been demonstrated in A-Life experiments and are now being developed as models of real world phenomena. Fashion systems and marketplace dynamics are a new arena for such approaches (Anderson, Arrow, & Pine, 1988). The marketplace is constantly forming and reforming with shifts in tastes, lifestyles, immigration, technological developments, prices of raw materials, and myriad other factors. In this project, the textile/apparel marketplace is viewed as a complex, adaptive, self-organizing system—that is, the actions of agents (consumers, firms, industries, economies) create a system in rolling equilibrium. Through the agents’ mutual accommodation and mutual rivalry, new structures and behavior patterns emerge continuously (Holland & Miller, 1991). I95-A20/5 Figure 2 ARTIFICIAL LIFE SIMULATION Why build an A-Life simulation? “Flight Simulator” for rare or novel phenomena Computer Laboratory for Hypothesis Testing Strategizing Theory Building and Generating Hypotheses The idea that complex behaviors result from relatively simple sets of rules offers a construct for examining and modeling consumer behavior and the competitive marketplace. Simulations of marketplace systems offer industry executives a window through which to observe these systems in action under various sets of parameter and variable settings. The effect is similar to flight simulators—a way to get the feel for how situations develop and important variables interact. Such simulations will not predict the outcome of a new acquisition or what new fashion will emerge next season, but they will provide statistical and structural measures of the process and guidance in answering business questions like: When do you introduce a new product? How big an impact will it cause? How many other goods and services does it bring to market with it, and how many old ones go out? How do you recognize when a good has become central to a system as opposed to a fad? A-life simulation can be used to: n gain insights about marketplace dynamics based on the emergent behavior in systems of adaptive agents. n revitalize fundamental theoretical frameworks that guide industry decision making and joint industry/academic research programs. n leverage expertise and accelerate understanding of marketplace phenomena by involving industry executives and the academic community in building and testing simulation. A-Life simulations allow executives and researchers to view patterns in data and introduce a richer environment for theory building and strategizing about evolving markets. InfoSumers: A Diffusion of Innovation Simulation Rogers (1983) proposed a diffusion of innovation model in the form of a bell-shaped curve. The left side of the curve represented the earliest adopters of innovation, the center section majority adoption, and the right side late adoption by laggards. The model launched innumerable studies of the demographics and psychographics of the consumer and of the way innovation moves through a population. Rogers’ concepts are important because today’s marketing managers use these ideas as a guide for the dissemination of new technologies, products, and services. Yet research on diffusion within consumer social systems has reached a state of malaise. Critical elements like competitive activity, segmentation strategy, and marketing mix decisions are rarely part of the conceptualization and empirical research on diffusion. Identification and targeting of consumer innovators and opinion leaders for product promotion is difficult, partly because an individual may be an early adopter in one product category and a laggard in another. The emphasis in past research was on direct relationships (main effects) (Gatignon & Robertson, 1985). Simulation offers the opportunity to consider interactions and the effect of marketing initiatives. Diffusion of innovation is the theoretical framework of the A-Life simulation InfoSUMERS, designed by Brannon, Thommesen and Duffield and programmed in Swarm software (Santa Fe Institute). InfoSUMERS models the interaction between change agent influence (advertising, promotion, product placement in media) and personal influence, between innovative early adopters and fashion followers. In the world inhabited by the adaptive agents, influence both diffuses and evaporates. At the beginning of a simulation, parameters can be set for the rate of diffusion and evaporation, the number of each kind of agents, and the number and kind of fashion looks available for adoption. Each agent evaluates the situation and determines the action at each time step according to its individual rule set. The rule sets include probability functions so that each decision is unique and adaptive. During each run of the simulation, data are gathered at each time step on the agent segments and their adoption profile. Additionally, screen capture is used to collect and record patterns of interaction at selected time intervals (usually every 100 time steps). The intent of the design team is to evaluate InfoSUMERS through all the stages for methodological completeness. The original simulation prototype was evaluated against the theory of diffusion of innovation and revised. The revision provides more complex rule sets for agents and a redesigned data collection scheme. Validation procedures for the revised simulation includes comparison of results from the simulation with published results from studies of diffusion of innovation. Sensitivity analysis determines the relative importance of the assumptions underlying the simulation. When fully implemented InfoSUMERS can be used to perform computer experiments to investigate differences in patterns of diffusion among various consumer segments. Sphere of Influence: Influence: A Simulation of Supplier-Consumer Relationships SPHERE OF INFLUENCE is an A-Life simulation designed by Marshall and Terase to represent the interaction between consumer agents with various preference sets and firms organized around product mix. When a consumer agent reaches a supplier’s sphere of influence, it uses a utility distribution curve and a preference distribution curve to determine whether to move toward or away from the supplier. The resulting assessment can produce a clear decision to move toward or away from the supplier, but if the decision falls into a fuzzy area of indecision, then determination of the behavior for the next time step is dependent on probability. Parameters on the number of suppliers and consumers of each type, the diffusion rate for influence in the world, and the probability that a consumer will initiate a move are set at the beginning of a simulation run. The simulation allows for alliances among various suppliers and for an increased motivation to move among consumers whose needs have not been satisfied in previous time steps. During the simulation run, data are gathered on the level of satisfaction among consumers and the number of consumers by type under the influence of individual suppliers and supplier alliances. The intent of the design team is to evaluate the simulation as a tool for strategizing about the marketplace using graduate students in business. In addition, the team intends to complete validation and sensitivity analysis so that the simulation can be used to perform computer experiments related to consumer-supplier relationships. Other A-Life Simulation Prototypes Design teams are in the development stage on the following prototypes: • An extensive expansion and revision of Sphere of Influence to allow more detailed descriptions of both firms and consumers (Marshall and Terase). • A simulation of product lifecycles to use in exploring the dynamics of new product introductions given a range of existing products with varying success at satisfying consumer needs (Ulrich and Thommesen). • A simulation of market share capture by competing manufacturing approaches—mass production and mass customization (Anderson and Thommesen). Network of A-Life and Textile/Apparel Experts Developing rigorous, replicable simulations requires a deep expertise in the system being modeled, the ability to translate that model into simulation parameters and variables, mastery of the technology to program the simulation, and competence in research methods for verification and validation of the simulation. Individuals are unlikely to possess all these skills and abilities. Teams are the ideal unit for simulation development. To accomplish the goal of this project, the core team networked with A-Life simulation developers through the Santa Fe Institute, with researchers studying marketplace phenomena through the International Textile and Apparel Association, and industry executives through the National Textile Center. Until recently, developing an A-Life simulation meant programming all the elements, schedules, and displays. Langton (1995) and the software development team at the Santa Fe Institute have made Swarm, a “virtual computer” for managing concurrent interactions of large numbers of agent-type objects, available to a wider range of users. As Swarm beta testers, investigators on this project participate in an international online network of simulation developers and attended the 1997 Swarm-users meeting in Santa Fe. Thommesen, research associate on this project, has been named to the board of directors for Swarm development and will participate in planning the future extensions of this simulation platform. Additionally, investigators for this project networked with scholars by conducting workshops on Alife techniques at two national meetings of International Textile and Apparel Association. The 27 scholars representing 22 universities corroborated the potential of the complex systems approach as a way to understand, investigate, and model marketplace phenomena including diffusion of innovation, theory of the firm, product lifecycle, and fashion change. These scholars along with industry executives participate in simulation verification and validation. To facilitate networking, the project Web site provides: • an overview A-Life simulation, • an explanation of the application of A-Life simulation to textile, apparel, marketing, and retail sectors, • an introduction to each of the prototypes developed during this project, • a survey requesting input on an agenda for A-Life simulation development. The Web site is linked to the Santa Fe Institute/Swarm, ITAA, and NTC Web sites. References Anderson, P.W., Arrow, K.J., & Pines, D. (Eds.) (1988). The economy as an evolving complex system. Reading, MA: Addison-Wesley. Doran, J. & Gilbert, N. (1994). Simulating societies: An introduction. In J. Doran & N. Gilbert (Eds.), Simulating societies (pp. 1-18). London: UCL Press. Gatignon, H. and Robertson, T.S. (1985). A propositional inventory for new diffusion research. Journal of Consumer Research, 11, 849-867. Holland J. (1995). Hidden order: How adaptation builds complexity. Reading, MA: AddisonWesley. Holland, J. & Miller, J. (1991). Artificial adaptive agents in economic theory. American Economic Review Papers and Proceedings, 81, 365-370, Langton, C. (1989). Introduction. In C. Langton (Ed.), Artificial Life. Redwood, CA: Addison Wesley. Langton, C. (e-mail, December 14,1995). Levy, S. (1992). Artificial Life. New York: Pantheon. rd Rogers, E.M. (1983). Diffusion of innovations (3 edition). New York: The Free Press. Seror, A.C. (1994). Simulation of complex organizational processes: A review of methods and their epistemological foundations. In J. Doran & N. Gilbert (Eds.), Simulating Societies (pp. 1940). London: UCL Press. Waldrop, M.M. (1992). Complexity. New York: Simon & Schuster.