Ant Colony IT

Computer scientists studying ants are using them as models for new distributed systems.

"Go to the ant . . . consider her ways, and be wise."

- Proverbs 6:6

Computer scientists are doing just that, and they say the humble insect heralds a revolution in software design.

Rather than relying on complex, centralized logic, systems that mimic ant behavior use many small, autonomous software agents. Each acting on the simplest of rules, these agents together can solve problems that are enormously complex when viewed as a whole. Higher-level, or "swarm," intelligence emerges from those rudimentary rules in ways that would be difficult to program into conventional software.

Such autonomous systems have no single point of failure. Like ants that repair their nest if it's disturbed and columns of ants that route around obstructions, these systems adapt to unanticipated problems in a way that ordinary centralized systems can't.

Today, such systems are typically used for optimization applications such as factory scheduling and vehicle routing. For example, General Motors Corp.'s Saturn subsidiary uses distributed systems to control the flow of part assemblies. Instead of having each factory machine's actions preprogrammed and unvarying, Saturn's machines can sense the type and state of a part and adapt their actions on the fly.

Other applications draw on different ant behaviors. For example, researchers are working on a way to analyze financial transactions based on the way ants sort and cluster their brood and their dead.

The technology is a decade old, but it's just beginning to blossom, says Doug Currie, president of Flavors Technology Inc., a Londonderry, N.H.-based company that sells agent-based systems. "It's a great way to leverage technology, because of the surprise factor - it does things you don't expect, but in a way that's natural," Currie says.

He adds that antlike systems will increasingly be used to design production lines, not just to control them. He predicts they could also have a use in medicine, where antlike agents could mimic the interactions of simple chemicals in the body and in drugs.

Individually, ants have primitive brains, yet collectively they run surprisingly sophisticated and efficient operations. With no central direction, they divide responsibilities among themselves, find food, build and maintain their nests, tend to their young and respond to attacks.

Using a few simple rules of logic, ants can find the shortest path to a food source. They lay down chemical scents, called pheromones, and over time the most successful paths - originally found by trial and error - are reinforced by the repeated deposit of pheromones.

Similarly, multiagent systems can solve routing problems by reinforcing good routes with "digital pheromones" and discouraging the choice of poor routes by allowing the pheromones to "evaporate." London-based British Telecommunications PLC has developed a method for routing data packets around a network that way.

Tucker Balch, a robotics researcher at Carnegie Mellon University in Pittsburgh, says the thousands of ants in his laboratory amount to a "library" of logic that can be applied to solving human problems.

"A multiagent system could be the stock market, or anything where there are thousands of individuals interacting with one another," he says.

Today, most agents have program logic built in from the beginning. But Balch says agents will become much more powerful as computer scientists enable them to learn from their environments. "These agents will empower us," he says.

Swarm intelligence pioneer Eric Bonabeau, chief scientist at Paris-based EuroBios, says much research and development remains to be done. "There is no clear understanding of why it works so well in a number of cases, and no proof it is going to work in every situation," Bonabeau says. "But there will be more and more applications, as there is no alternative when you want to control armies of small, dumb entities."

There are also workforce barriers to overcome, says Van Parunak, chief scientist at the ERIM Center for Electronic Commerce in Ann Arbor, Mich. "Saturn is a good example, because it is able to diagnose itself and configure itself. The action it takes at any moment can't be predicted, and people find that disturbing," Paranuak says.

And sometimes it isn't just factory workers who can't predict the behavior. "Systems of interacting parts have the potential to self-organize, but also the potential to chase their own tails, to go into random-like operation and otherwise fall apart," says Parunak.

Copyright © 2001 IDG Communications, Inc.

  
Shop Tech Products at Amazon