THEORY AND PRACTICE

Agents Gain Insight

by Lowell Rapaport

A couple of years ago when knowledge management was traveling the buzzword circuit, one of the new technologies bandied about was the intelligent agent. Intelligent agents are still with us: They function so transparently that most people don't even know they're using them.

For some of you, knowledge robots, or "bots" as they are more commonly known, may come to mind, but intelligent agents are different. Bots are programs designed to search for data according to predetermined rules created by a user. Bots are used in search engines, peer-to-peer clients and the personalization servers that are part of many portal and content management systems. A bot can also include task-specific engines that form part of many enterprise workflows. While bots are relatively simple tools that operate according to a set of rules they follow by rote, an intelligent agent is capable of modifying its behavior to suit changing environments. For example, a bot will search and index HTML text on a Web site and ignore everything else. In contrast, an agent will search and index HTML text on a Web site and then proceed to search and index non-HTML files such as PDFs and other content also contained within the Web site.

Raj Rao, senior vice president of product management at BroadVision, Redwood City, CA, describes agents as "expert systems that operate independently of users, [requiring] input only for exceptions or special situations."

As an example, Rao points to customer service applications where the agent can take care of routine questions automatically and refer to a human expert in special cases. This approach puts agents square in the path of any workflow where they can take the place of humans for routine decisions.

As an example, Rao points to customer service applications where the agent can take care of routine questions automatically and refer to a human expert in special cases. This approach puts agents square in the path of any workflow where they can take the place of humans for routine decisions.

"Agents ensure that tasks will be carried out with a specified quality of service according to standardized business practices," Rao says.

The long-term trend for agents is toward greater levels of autonomy and intelligence. In the case of data aggregation, agents will learn to differentiate relevant information from the irrelevant. Currently, if you tell an agent to search the Web, it will search the entire Web and won't be able to separate the really useful information from the useless. Expert systems will improve their reasoning abilities and accomplish more tasks autonomously.

Rao predicts that agents with high-level capabilities are only 12 to 18 months away, but Marcus Zillman, CEO of Bottechnology, San Francisco, contends it will be three to five years before agents offer true reasoning capabilities. As these technologies emerge, Zillman predicts political roadblocks may arise. "Agents may threaten to displace workers, which may lead to strong criticism of the technology," he warns.

While controversy may arise, the history of technology suggests that humans will embrace and adapt. As agents become more intelligent, they will increase productivity and create new opportunities.