Over the past year, NASA has been uploading software into the Earth Observing-1 satellite, turning it into a testbed for autonomous agents. The agents -- software programs that are able to learn and can function independently -- are used to manage experiments and operate the spacecraft.
The effort is part of a technology initiative that researchers say will reshape IT over the course of many years. Autonomous agents have the potential to become an extraordinarily powerful technology, with the capacity to learn, experiment and act independent of human control. Agents could ultimately improve productivity, increase software reliability and change the operation of markets, particularly supply chains.
NASA uses autonomous agents to handle tasks that appear simple but are actually quite complex. For example, one mission goal handled by autonomous agents is simply to not waste fuel. But accomplishing that means balancing multiple demands, such as staying on course and keeping experiments running, as well as dealing with the unexpected.
"What happens if you run out of power and you're on the dark side of the planet and the communications systems is having a problem? It's all those combinations that make life exciting," says Steve Chien, principal scientist for automated planning and scheduling at the NASA Jet Propulsion Laboratory in Pasadena, Calif.
Like many programs, agent software and related algorithms are often coded in Java. What makes them different is that the designs also incorporate disciplines such as game theory. Agent designers tend to draw from a variety of areas, such as economics and psychology, in an effort to create programs capable of handling complex interactions.
NASA's Earth Observing-1 satellite, which began operation in 2000, was recently turned into an autonomous agent testbed.Image Credit: NASA
Programmers are adept at building systems that respond to a certain set of "if-then" circumstances. But NASA's agents are model-based, designed to achieve the goals and intentions of the designers, not merely to respond to a given event. That means they can react to unimagined events and still ensure that the spacecraft does not waste fuel while keeping to its mission.
At NASA, software agents are performing work previously handled by ground controllers. But the cost-saving potential of agents is something "we don't emphasize, because nobody likes their budget reduced," says Chien. Instead, the focus is on the additional scientific research created by the use of agent-based software.
Making markets, supply chains, telecommunications and other systems more efficient through the use of agents is a subject of intense interest. Some 800 researchers recently gathered at Columbia University for the Third International Joint Conference on Autonomous Agents & Multi-Agent Systems, the leading conference on the technology. In addition to people from universities and government agencies, the event attracted researchers from defense, telecommunications and software development.
The conference stressed both abstract concepts and practical application of the technology. Most presentations detailed the latest research in a wide variety of areas, such as machine learning, which involves creating agents that have the ability to discover interactions and respond to them. Others addressed the behavior of multiple agents: how they work together, exchange information, set priorities and negotiate with one another.
Putting It to the Test
Negotiation was one of the key agent capabilities tested at the conference's Trading Agent Competition. In one contest, computers ran simulations of agents assembling PCs. The agents were operating factories, managing inventories, negotiating with suppliers and buyers, and making decisions based on a range of variables, such as the risk of taking on a big order even if all the parts weren't available. If an agent made an error in judgment, the company could face financial penalties and order cancellations.
Researchers say there is no optimal agent design, and the contest serves as a performance testbed. "If you have a better strategy than your competitor, you're in a more advantageous position in the marketplace," says Nick Jennings, who heads the autonomous agent program at the University of Southampton in England.
Decision-making software is already used in financial markets for functions such as automated trading, but those processes are simpler than the business-to-business transactions that agent researchers are testing. Instead of dealing with fixed prices, researchers are looking at scenarios where agents must constantly negotiate the best prices with a variety of suppliers. In so doing, they must take into account changing market conditions while also considering inventory, production, factory capacity and other issues.
"Our scenario presumes automation beyond the degree that currently exists, but I think things are going in that direction," says Michael Wellman, professor of commercial science and director of the artificial intelligence laboratory at the University of Michigan.
IBM is building agent technology to support its autonomic computing systems, which have the intelligence to reconfigure themselves in response to changing conditions, according to Jeffrey Kephart, manager of agents and emergent phenomena in the computer sciences department at IBM's T.J. Watson Research Center.
Agent researchers "are working toward a world some years hence where these automated decisions are going to be used. Those people in industry who are really thinking about this know that they are going to need technology of this nature, in supply chains," Kephart says. The research work so far bodes well for that future, he adds.
Agents may change the nature of distributed computing environments. Instead of centralized control systems, agents could independently operate parts of a network but also have the ability to take over other functions if problems arise. Interest in building survivable systems, where agents regenerate lost capabilities in communications networks, is of interest to the U.S. defense researchers and scientists at Cambridge, Mass.-based BBN Technologies.
Agent researchers say the technology won't arrive in big-bang fashion. But slowly, over time, agent functions will begin turning up in systems. This technology is "going to be revolutionary," says Richard Lazarus, manager of the enterprise architectures group at BBN, "but it will occur incrementally."