ARTIFICIAL INTELLIGENCE
The Smartest Agents Will Learn to Be Team Players.
As their tasks become more complex, the smartest agents will learn to be team players
By Christopher Locke
January 9, 2002

You would trust them completely. They'd become your closest confidants. But you wouldn't be able to see or touch them, and unlike some friends or family members, they would never betray you.

Welcome to the future of "smart agents". This new breed of technology uses small software programs built with artificial intelligence to make independent decisions, like automatically searching for and purchasing specific kinds of products on the Web, or deciding what stocks to buy and sell in your financial portfolio.

Most of the so-called smart agents commonly used today--those in commercial applications--are actually pretty dumb Web crawlers that seek out requested material over the Internet. The leading-edge research on agent technology is still being conducted at only a handful of universities and companies. Next-generation agents are still several years away.

A number of things are holding back progress. First, there's no clear agreement on a universal standard or language, making communication between disparate agents difficult. In addition, there's the issue of security--protecting agents as they zip from one Web site to another, collecting and interpreting data.

Smart-agent technology also has a history of shortcomings. One notable example is the Mycin system, which was developed at Stanford University in the '70s to diagnose and recommend treatment for certain blood infections. Scientists fed it a list of symptoms: the patient had blotchy patches and raised sections that were reddish in color. Mycin diagnosed the patient as having measles. The only problem was that the patient in question was a rusted-out Volkswagen.

Katia Sycara, principal research scientist at Carnegie Mellon University, is unfazed by the naysayers. A prominent figure in the field, she has spent 15 years working on intelligent agents and is convinced that the technology will play a crucial role in the future of computing.

Her most interesting work is focused on creating multi-agent systems. In these systems, several agents work together, much as a team of people would, with individual members working on small portions of a larger problem.

An example of how the system would work is a crisis situation--say, planning the evacuation of American citizens from a foreign city during a conflict. In this scenario, each agent in a multi-agent team would be assigned a specific task, like plotting an evacuation route, forecasting the weather, or laying out a flight plan. All of the agents would coöperate and gather data to determine the optimal choices.

But without standard definitions for how data is shared and presented, intelligent agents won't be able to reach the levels of coöperation and intelligence that Ms. Sycara's vision requires. What's more, agents need to communicate with one another so the overall system won't go haywire if an agent abandons the task at hand or another joins the group.

Moreover, security issues haven't been ironed out. That's especially important in the distributed world of networks, where data flows unhindered from server to server across the Internet. With agents flowing around the public networks, there's the potential for cyberthugs to intercept them and decipher their data or reprogram them to collect data for illegal purposes.

IBM's director of research, Paul Horn, has a slightly different vision than that of Ms. Sycara. He says computing should work like a human autonomic nervous system--monitoring temperature, checking oxygen levels in the blood, and telling the heart how often to beat. Mr. Horn says the next evolutionary step is to have computers run themselves and adjust to varying circumstances. The autonomic-computer scenario is currently the No. 1 priority for IBM's $5.6 billion research and development division.

It's clear that intelligent-agent technology will play an important role in both interpreting and acting on information across networks. But in order to succeed, coöperation among and protection of these new confidants will have to be ensured. Looks like Ms. Sycara has her work cut out for her.

Write to Christopher Locke.