The booming world of electronic commerce now provides bidders using Internet auction houses such as eBay and OnSale with virtual agents to do their bidding. These agents are programs that literally do a buyer's bidding as he or she goes about another task, from cleaning the house to closing a big corporate deal. The agent reports back to the buyer on his or her PC every five minutes or so and is instructed not to bid above a certain price.
Virtual agents have great advantages over human agents. They don't charge a 15 percent commission, for one. And they don't put you on hold when you call, for another. But the agents in current auction houses are limited by a lack of mobility -- they can't go to other auction houses to compare prices and enter the bidding there -- and capability -- they only perform within the narrow parameters of the auction at a particular Web site.
Now, a computer scientist at Washington University in St. Louis has created a group of versatile virtual agents for electronic commerce that runs the gamut from Swifty Lazarr to Jerry McGuire.
Tuomas Sandholm, Ph.D., assistant professor of computer science in the School of Engineering and Applied Science, has developed a prototype, or model, e-auction house that provides users with "mobile" agents that can monitor auctions, bid, set up auctions and move to other auctions, among other capabilities. The user has the full flexibility of Java programming language at his or her disposal when designing an agent. Also, an html interface is provided for non-programmers, allowing the user to specify what he or she wants the agent to do. The system automatically generates the Java code for the corresponding mobile agent and launches it.
The server presently has five different kinds of predesigned mobile agents, each tailored to the precise rules of a particular auction and each able to zoom through cyberspace to enter multiple auctions and return to the user's computer. The prototype server is called eMediator, and, though additional features are continuously being added, it already has features that go beyond those of any of the top eight on-line auction houses now on the Net.
Sandholm is a specialist in artificial intelligence and the mastermind behind eMediator. He has been working on agent-mediated e-commerce for nine years and has more than 85 publications on the topic. The prototype e-commerce server uses a mathematical tool called game theory and elements of expert systems to provide a comprehensive market place where virtual agents and real people alike can compete in a real marketplace. EMediator, though a research prototype, is an actual auction house, where people can both buy and sell, with noted restrictions. Its URL is: http://ecommerce.cs.wustl.edu/emediator/.
"To my knowledge, we are the first Internet auction house with mobile agents," says Sandholm. "In essence, we let the user create his or her own agent. In many auction houses the best bidding strategy is quite intricate and difficult for a user to grasp. Based on game theoretic analysis, we compiled the optimal bidding strategies into the different agents. This puts a novice user and a game theorist/expert bidder on equal footing for electronic commerce."
In a recently published technical report, "eMediator: A Next Generation Electronic Commerce Server," Sandholm described the five predesigned agents:
* Information agent, a standard workhorse that e-mails the user price information and asks whether to bid higher or not, freeing the user from actually monitoring the auction;
* Incrementor agent, employed in single-item, single-unit auctions, it bids slightly higher than the current highest price and stops if the user?s top price is reached;
* N-agent deliberately underbids optimally in an auction where the number of bidders is known (the underbid is based on the worth of the item to the bidder);
* Control agent adds an air of espionage to an auction. This agent submits very low, noncompetitive bids to artificially increase the number of bidders to mislead others, like the N agent, for instance. It is in the seller's best interest to submit control agents so that N-agents would bid higher; and
* Discover agent, based on a learning model, tries to reason or discover if it makes sense to bet slightly higher based on the pattern of other bids coming in.
An expert system in eMediator determines which agent or agents are best for you based on the kind of auction you're entering. Tired of your agent? You can do what disgruntled creative artists have dreamed about since the beginning of the "middleman" profession.
"You can kill your agent, too," says Sandholm. "Each user can view her list of agents and press the kill button when she no longer needs one of them."
While agent mobility is a fascinating aspect of eMediator, there are other features that are as equally outstanding. It is currently the only Internet auction house that offers combinatorial auctions, where multiple items of multiple goods can be gathered and bid on. It also offers millions of auction types and has built-in expertise to help the user choose an appropriate one for his or her situation. EMediator provides a price-quantity graph that allows buyers and sellers a number of options, such as buying in larger quantities for a lower unit price, and tradeoffs.
Applications for eMediator extend beyond e-auction houses. Electronic equity trading, transportation routing, telecommunications and many other industries are turning toward electronic planning, buying and scheduling. Sandholm, for instance, has performed an experiment with five trucking companies bidding for contracts. All of them used virtual agents, and they realized 17 percent savings in costs by being able to form coalitions and achieve efficiency by reassigning transportation tasks among the dispatch centers.
To enhance eMediator and other Net-based auction houses, Sandholm also has developed an algorithm that determines the winner in combinatorial e-auctions orders of magnitude faster than previous approaches. He has filed for a patent, and will present a paper this summer on the algorithm at the premiere artificial intelligence conference in Stockholm, Sweden.
"Today, everything from toasters to beanie babies, telecommunications hardware to sporting goods and jewelry are being auctioned on-line, and it appears that e-commerce will be expanding in all sorts of areas," says Sandholm. "We think eMediator offers very attractive components for more sophisticated e-commerce."
Materials provided by Washington University In St. Louis. Note: Content may be edited for style and length.
Cite This Page: