Industrious Ants Teach Powerful Lesson In Simplicity

When Eric Bonabeau ponders how to build adaptive systems, he thinks small.

Bonabeau spent four years at the Santa Fe Institute studying insect behavior before founding a consulting business, Icosystem Corp., that's based on the idea that behavior copied from ants and other biological systems offers a model for self-aware software and business processes.

Bonabeau was working as a network engineer at France Telecom in 1991 when his fascination with ants took hold. He was looking for a system for network routing in which distributed software components could organize themselves and find the optimal solutions to problems in the network. That experience led to his stint at the Santa Fe Institute in New Mexico, a place best known for its pioneering work in complexity theory and its application to business-optimization problems.

While in Santa Fe, Bonabeau and some of his colleagues hit upon a new field of study, based on ant behavior, called Swarm Intelligence, which attempts to solve common computing and business-optimization problems by modeling software tools on behaviors that scientists have observed in nature. Ants are relatively simple organisms, but when working in a group, they can achieve astonishing things. An ant, for example, communicates to other ants when it finds a food source by laying down a chemical trail of pheromones. Each ant following the chemical trail lays down more pheromones, which reinforces the message to the group that this is the path to a good food source. When the source dries up, the ants stop dropping pheromones.

This is the kind of naturally occurring optimization behavior that gets Bonabeau excited. Swarm Intelligence solves problems from the bottom up, rather than the top-down approach found in business decision-making. It's best applied to environments in which lots of complex scheduling changes happen rapidly.

For Swarm Intelligence to work in computing environments, software needs to be built from many pieces of component code that are fairly simple, with a finite set of instructions--kind of like an ant's brain. "Software is too complex, and developers need to run too many tests to ensure its reliability. This is becoming an untenable situation," Bonabeau says.

Instead, the software Bonabeau and his colleagues have developed contains what's referred to as "emergent behavior," in which objects and patterns arise from simple interactions among software components. This lets software become self-organizing to solve problems and adapt to changes in the computing environment by discovering ways to work around system outages.

General Motors Corp. implemented software using adaptive technology on the factory floor to schedule car paint jobs and avoid the scheduling conflicts and inefficiencies from which the manual system suffered. The new system resulted in a 30% productivity improvement and 35% fewer business-process changes.

Swarm Intelligence also is being used in the design of computer algorithms and in complex financial systems. The U.S. military uses it as it tries to develop robots that can work in cooperative teams to figure out the best way to move goods. Like an ant colony, software made up of small, simple elements could add up to something far greater than its individual parts.