 | | | GROSOF |
SEATTLE, Wash. - The pace of the marketplace is going to pick up - as soon as we get those pesky humans out of the way.
Artificial intelligence researchers are building computer programs that will act as online agents. They will be able to buy stock, trade securities, bid in auctions and buy widgets for a company at the best possible price.
``There will be billions of agents . . . performing automated interactions between software systems,'' said Benjamin Grosof, a speaker at the 17th International Joint Conference on Artificial Intelligence being held this week in Seattle.
Grosof studies artificial intelligence at Massachusetts Institute of Technology's Sloan School of Management. As a part of his research, he helped develop an AI agent to negotiate contracts for business suppliers in manufacturing.
The $29 million (U.S.) project, conducted over three years, was funded by such high-tech firms as International Business Machines Corp. and Lucent Technologies Inc., as well as by the National Institute of Science and Technology, a U.S. government agency.
The challenge to make the research commercially viable, he told the conference Sunday, is to build a shared language for agents from different industries. If Company A needs to find someone to mow its lawn, its agent will need to scour the Web for lawn-mowing companies, then communicate with a company's agent to negotiate a contract.
The terms of the contract - price and number of cuts per year - need to be defined using shared terminology. Without any thought on the part of human owners, AI agents would modify a contract until they arrived at terms both parties could agree to.
``This is the idea of the semantic Web,'' Grosof said, ``the idea of having data on Web sites defined and linked in a way that can be read by all machines.''
Setting up such a language is a daunting task but it could one day make businesses far more efficient than they are now.
``A key difference between an economy with lots and lots of agents and an economy with lots and lots of people is that the speed with which things can happen is thousands of times faster,'' Grosof said.
``In a world full of agents, the time scale for market swings could get radically compressed down.''
The problem is that such swings could cause market supply squeezes, shortages and gluts.
How AI agents might affect electronic markets has been the research subject of another speaker, Tuomas Sandholm, an associate professor at Carnegie Mellon University.
``This field is becoming increasingly important due to the rise of the Web, e-commerce and an industrial trend toward outsourcing,'' he said at the conference on Sunday.
Instead of outsourcing some jobs to other companies, he said, a firm could hire an AI agent.
As part of his research, Sandholm has been analyzing various types of transactions and rendering each into a mathematical formula called an algorithm. The algorithms tell the AI agent how to perform.
The human using the AI agent sets the parameters of the transaction - what to buy at an auction, for example, and at what price. The agent could then work out the strategies of everybody else bidding on the same item by looking at their strategies in past auctions.
Sound more rational than most business negotiations? It is. Unlike AI agents, humans have a tendency to behave irrationally. Simple versions of AI bidding agents, called proxy bidders, are already used online. But like computers from the 1980s, today's agents are primitive compared with what lies ahead.
Things could really get interesting when AI buyers and AI sellers start to interact.
Grosof said that long before that happens, there should be open discussions on public policy to govern the behaviour of AI agents.
Agents need to have a finite number of resources to work with, he said, and a definite human owner who is accountable for their actions.
``These issues loom large,'' particularly with the chaotic growth of the Internet, Grosof said. ``In the next two to seven years, we could see a lot of these agents out in the wild.''
|
