Observations from the AAAI Complex Adaptive Systems Fall Symposium

For the last three days I have been at the AAAI Complex Adaptive Systems Fall Symposium at the Westin Hotel in Alexandria, VA. This has been a normal CAS get-together because the attendees are from a wide variety of fields — business, philosophy, computer scientists, biologists, and all types of social scientists. Our symposium has been held alongside other AI-related symposia on decision making under uncertainty, understanding how to represent common understanding, deploying robots in dangerous environments, and others.

About our gathering I can make several observations:

• The University of Michigan’s Center for the Study of Complex Systems is the indisputable leader in the study of complex adaptive systems. We probably made up over 20% of the attendees here, though attendees also came from Europe, Canada, Alaska, and much of the eastern U.S. (from Maine to Maryland to Pennsylvania). Even beyond this indication, the strongest signal of this leadership was that Carl Simon delivered the Keynote Address to this symposium.
• At least for the attendees this weekend, NetLogo remains the dominant tool for agent-based modeling — probably as a result of the strength of the representation of social scientists and their appreciation of NetLogo’s ease of use. I started using this tool many years ago and it continues to serve me well.
• Another thing I noticed is that it remains the case that articles that use (or are about) this approach are spread widely among many different journals. Further, the top journals in most fields almost uniformly don’t accept submissions that use this approach. Compared with a decade ago, I believe it is true that more, and better, journals accept such research now. At least it’s a start.
• Finally, it seems that most researchers in the field who are working with CAS simulations that address some type of real world problem are approaching model validation in a relatively informal way. They are generally just letting their customers/clients play with the model in order to gain an understanding of how it responds. Also, if some historical data is available, then the researcher demonstrates to the customer how well the simulation conformed with that historical data. And, again, this is usually done in an interactive situation instead of by providing some type of regression statistics or the like.

It was good to see the breadth of CAS research that is being done — as defined by variety of fields and by the range from purely theoretical to highly applied. I look forward to participating as a presenter and author at my next meeting.