Agents, or software systems that act rationally, intrigue Amy Greenwald, assistant professor of computer science. She's exploring pricebots, which let companies search for the lowest prices, then automatically set their own prices lower
The market surrounding the Internet is competitive, but software is also getting competitive. Electronic shoppers can use Web sites - like mySimon.com or evenbetter.com - that seek the lowest prices for the item they want.
These e-businesses use what are known in computer science as agents, software systems that act rationally. Shopbots like mySimon.com search the Internet and come back with the lowest prices, but there also are things called pricebots that let companies search for the lowest prices, then automatically set their own prices lower.
This link between computers and strategic thinking is what intrigues Amy Greenwald, assistant professor of computer science. Her exploration of what will happen next in e-business leads her to believe that these lurking agents will keep spying on each other, setting lower prices, creating price wars that will eventually raise prices and start the cycle again.
Greenwald's interest is in using modeling and simulation to understand the agents' behavior. Then software could be designed so that cyclical price wars won't break out. These and other topics will be covered in her computer science course, "Topics in Internet Agent Economics," next semester. This summer, she is teaching on campus through a mentoring program of the Computing Research Association.
Greenwald's interest started when she was working on her doctoral degree. She audited a course at New York University on game theory - the strategic interactions between humans, as in , "If he does this, then I might do this."
"It occurred to me in 1995 that this strategic reasoning could be built into computers," Greenwald said. At the time, interest in the Web and e-commerce was beginning to grow.
Greenwald looked at the impact of shopbots on the economy from the seller's perspective: If a seller knows that shopbots are going to come, how can the seller respond, knowing that others are going to respond in turn?
She and her collaborator Jeff Kephart at IBM's Institute for Advanced Commerce in Yorktown Heights, N.Y., theorized about price-comparison software that would automatically allow a seller to undercut prices set by competitors. They called the software agent a pricebot, believing they were first to coin the word. Soon afterward, someone built the software.
Greenwald presents her research at conferences on artificial intelligence and game theory.
To create agents, Greenwald spends time at the whiteboard, working out algorithms to see whether they might learn game theory.
When she is not theorizing about agents, Greenwald teaches her students how to write agents. She recently directed her students to design auction agents for a mock competition. The agent would have to bid high enough to win a travel package that included flights, hotels and entertainment - but without paying too much.
The winning team consisted of Hyun-Sik Byun '00, Sze Lok Audrey Yau '01 and Wolfgang Bardorf '02. Their agent will compete against others on July 7 in a competition organized by the University of Michigan.
"We have some pretty stiff competition from AT&T Research and North Carolina State," Greenwald said.