ANN ARBOR—Imagine a day when you merely tell your computer that you want to take a trip to Boston, and it automatically seeks out and buys the best airfare, reserves the most affordable hotel room that meets your wishes, and gets you a ticket to a Red Sox game. All you’d have to do for yourself is flag down the hot dog guy when you get to your seats in Fenway Park.

Such a day may come if artificial intelligence researchers continue improving computer programs known as “trading agents.” This software goes beyond merely searching for options and presenting the user with choices; it bargains on your behalf with other agents in an automated auction environment and tries to cut the best deals, explains Michael Wellman, associate professor of electrical engineering and computer science at the University of Michigan.

For e-commerce, especially the huge and growing business-to-business sector where companies buy and sell commodities like steel bars and toner cartridges online, autonomous agents would be very handy, says Peter Wurman, an assistant professor at North Carolina State University and former Ph.D. student of Wellman’s who teaches e-commerce. “Ideally, you would tell the agent just enough about what you want and it can go out and make the choices for you,” Wurman says. Of course, people who get recreational value out of shopping would probably prefer to keep making the choices themselves, he adds.

To encourage the development of trading agents and to allow software authors to pit their agents against others in a “real-world” contest, Wellman and Wurman are co-organizing a trading agent competition in Boston on July 8. The event will be a part of the International Conference on Multiagent Systems.

Trading agents that compete will be assigned the task of planning trips for multiple clients at once, buying airfare at the best prices, booking the best hotel rooms, and scheduling and buying entertainment as well. Not only will multiple agents be working for numerous clients, but the price and availability of each item will change continuously in an auction environment.

“What makes it so hard is that the dynamics of the environment are changing, based on the actions of other agents,” Wurman says. “This game was designed to be difficult to predict.”

“Our plane ticket thing is very realistic,” says Wellman. “The prices change pretty much at random.” So the agent has to try to predict the best time to get a ticket. Room prices for the hotels—one very nice and one a flea bag—will rise continually during the 10 minutes each contest runs and competing agents
The winning agent will have created the most customer satisfaction at the lowest price: Did all the clients arrive and depart when they wanted to, did they each have a place to sleep, will they be in town the night of the concert ticket, and how much did it all cost?

In addition to being a fun problem, the contest “is a good way to focus the research community on a common challenge,” Wellman says. Not that he expects all the entrants to come from traditional academic labs like his own. Hobby programmers and talented pre-college students are also expected to enter, and several artificial intelligence and e-commerce courses around the country will be using the contest as a class project, Wellman says. His lab will be offering several “dummy” agents for participants to download and improve upon, and the auctions will be run on the U-M’s Internet AuctionBot software.

Registrants should notify the organizers of their intent to participate by April 15. Details on the contest and registration information can be found on the Web at http://tac.eecs.umich.edu/cfp.html. You can learn more about the U-M Internet AuctionBot at http://auction.eecs.umich.edu/.

EDITORS: Contact Michael Wellman at (734) 764-6894 or [email protected], and Peter Wurman at (919) 515-9676 or [email protected]

Michael WellmanPeter WurmanInternational Conference on Multiagent Systemshttp://tac.eecs.umich.edu/cfp.html