|Contributors:||Tom Busey: Author|
Dean Wyatte: Programming
Complex adaptive systems are a reoccurring theme in fields such as biology, computer science, neuroscience, and economics. While such systems may manifest themselves in very different ways, from the organizational behavior of ants to cars in traffic to patterns of neuronal activation, the underlying concept is that simple rule-based parts interact to form complex emergent behavior.
An adaptive system contains variables which are subject to change. These changes occur as a system "adapts" to a problem - attempting to find a good solution. There are many kind of optimization strategies for finding the best values for all the variables in the system. The quantified ability of a system to solve a problem is it's "fitness". Simulated Annealing, the strategy that is demonstrated in our two simulations, recognizes the fact that in a "fitness landscape" (the set of all possible system states and fitness), one can get stuck on a sub-optimal strategy (see "balldropper" for a more detailed explanation). Simulated annealing causes the system to take random jumps to other locations in the landscape, in an attempt to find a better solution.
We use the multi-agent simulator Breve to demonstrate two examples of such systems, and allow manipulation within the simulation so that questions about the composition of such systems can be explored. The user also has direct access to the code through Breve's setup, so that making changes to the simulation is easy.
Download the software Breve from this site. Installation should be straightforward, but there is documentation on the site to help with questions. (Keep in mind that because this is a multi-agent simulator with a physics engine, it's going to take a lot of computing power.)
This module was created by the Indiana University Cognitive Science Program and is also available on their website: IU Cognitive Science Software.