By David Schwager and Brian Kleiner
Among the three biologically influenced modeling techniques neural networks, genetic algorithms and ant colony optimization (ACO), ACO is the newest and the most intriguing because of its close connection with easily observable nature. Looking at actual ants shows several possible improvements.
Many of the ACO models include a memory or map. In the traveling salesman problem, for example, the imaginary ants remember which cities they visited and know which cities they have not yet visited. Real ants, however, have no memory. Mathematicians could improve on nature in other ways. For example, one virtual ant could pass the exact turns and length of the best known path to a second ant, refining the random searches.
Although we can consider how the model can improve upon nature, it is more interesting to consider the ways in which nature can improve upon the model. ACO is a simplification of nature, and several aspects of social insects could have commercial applications.
Examining ant colonies naturally leads one to consider other social insects. Termites, bees or wasps combine to form a “meta-organism” somewhat like the way cells in animals and plants cooperate to make one organism. This is the link between ACO and neural networks. One simulates cells in an animal while another models animals that cooperate as if they were cells. There has been some effort to find business applications of the behavior of honeybees. Older bees look for food while younger bees work at the hive. If food runs low, however, young bees also forage outside the nest. Researchers at the University of Warwick in the United Kingdom applied this concept to flexible manufacturing. A multifunction machine tool, for example, may be dedicated to a particular job, but change jobs when it senses the need.
Read more about ant colony optimization and other modeling techniques from nature in David Schwager and Brian Kleiner's article "Modeling Systems by Nature" in the January/February 2010 issue of Industrial Management.