- 20 Jun 2016
Bill Rand is the lead instructor of the eagerly anticipated new Complexity Explorer online course, Introduction to Agent-based Modeling. He is an assistant professor of Business Management at the Poole College of Management at North Carolina State University. He recently co-authored a textbook on agent-based modeling with Uri Wilensky, the author of the NetLogo programming language. While Bill was preparing for his course he came by the Santa Fe Institute and chatted with Gabrielle Beans about the new course and his research. Read on to hear what he had to say.
Bill Rand came to complex systems by way of Artificial Intelligence and Evolutionary Computation, quite accidentally.
As an undergraduate at Michigan State University, he had a research assistantship, and really wanted to study and build virtual reality systems, but at that time no one was really working in that space. Bill’s mentor, Bill Punch, was really interested in genetic programming, and had been working on a system called lil-gp for quite some time. They worked together on that project, and eventually Bill Rand wrote and published his first agent-based model with him, though he didn't call it an agent-based model at that time. During graduate school, he worked with John Holland, and he introduced Bill to the Center for the Study of Complex Systems at the University of Michigan. However, Professor Holland didn't have graduate funding for Bill, so he took another research assistantship to pay the bills, working for Dan Brown, Scott Page, and Rick Riolo on a project that studied suburban sprawl using agent-based modeling. He also took Rick's courses in agent-based modeling at the same time. He went on to do a postdoc with Uri Wilensky at the Northwestern Institute on Complex Systems, worked on the NetLogo project for a few years, thus cementing his relationship with ABM.
According to Bill, agent-based modeling (ABM) and complex systems science are closely connected.
Complex systems science is all about understanding how the various aspects of a system come together to create patterns of behavior that are not defined by any component, but rather are emergent from the totality. Agent-based modeling provides a method for creating computational models in this framework. In an agent-based model you write rules at the level of the individual or the agent. These agents interact to create emergent patterns. Unlike traditional modeling approaches in which the focus is on working backward from a pattern of behavior to create a model that generates that pattern, in ABM you work forward from an individual agent's rules to observe the pattern that is created. This makes ABM a natural method for exploring complex systems.
In Bill’s own area of research, agent-based modeling has been increasingly used to study patterns of behavior with regards to the diffusion of information.
For instance, it has long been speculated that there are certain individuals in society, usually called influentials, who are responsible for spreading information quickly through the population. However, recent agent-based models, including some by Duncan Watts and Peter Dodds, combined with network analysis, have shown that it is potentially possible to configure a network in such a way that influentials are not really important in the spread of information, or at least that they may not have that much more power than any other individual in causing information to go "viral".
One project Bill is currently working on is a paper with Roland Rust at the University of Maryland and Ming-Hui Huang at National Taiwan University, where they are exploring the role of the speed of innovation in companies’ performance. They are examining under what conditions it makes sense for a firm to innovate quickly and rush to market, even if that means the quality of the product will not be that great, or if it makes more sense to wait a bit and allow the product to mature. What they show again and again is that the speed of innovation that is optimal depends on the competitive environment of the firm, but regardless, there is a "goldilocks" speed, i.e., too fast and the firm will fail, but too slow and they will be beaten to market.
What can students expect from the course on Complexity Explorer?
Students can expect to learn why and when agent-based modeling is useful in studying complex systems. They will also be able to build and construct their own agent-based models to understand phenomena of interest to themselves, and analyze the results of those models in a rigorous, scientific fashion so they can make clear generalizations of the results. Finally, they will learn about other domain areas and subject matters where ABM has been successfully applied.
Bill hopes students will come away from the course not only with a knowledge of how to build and construct models, but also with a higher level of model literacy. In other words, they will understand what it means when someone says that they used a computer model to make a decision, and what kinds of questions they should ask to examine results based on a model.
In his spare time, Bill likes to create intricate beers.
He's a homebrewer, and one of his favorite styles is Belgian Lambics. What he finds fun about brewing is that they are really an emergent property of the wild yeast, malt, and the aging process. They change their flavor constantly over time. His most recent batch is a Geuze which is actually a blend of three different years of Lambics, and the taste is different than any of the beers by themselves.
Agent-based modeling is a tool applicable to many diverse fields. Students looking to study complex systems will certainly learn something to their advantage. Take part in the course and see what it can do for you.
