Living in the computer age, we find that the "information highway" is truly a high-speed information center making some "travelers" dizzy. Intelligent software agents can help alleviate the problems with the amount of information received. What exactly is an intelligent software agent? They are autonomous computer programs, where their environment dynamically affects their behavior and strategy for problem solving.

Classes of Intelligent Agents

There are three distinct classes of agents, which can be easily identified. The first level of intelligent agents is the gopher agent, which function by executing straightforward tasks based on predetermined rules and assumptions. The second level of sophistication is where the service-performing agents execute a well-defined task at the request of the user. Finally, predictive agents volunteer information or services to a user without being explicitly directed, whenever it is deemed appropriate.

Characteristics of Intelligent Agents

A typical intelligent software agent (ISA) is should contain the following main characteristics:

"At a higher level, agents are also expected to perform other functions such as mobility, rationality, adaptivity, benevolence, and collaboration. All these characteristics enable an ISA to better represent its creator in the typically volatile WWW environment in which it operates." (Khoo et al, 1998)

As can be seen, "this technology combines artificial intelligence (reasoning, planning, natural language processing, etc.) with system development techniques (object-oriented programming, scripting languages, human-machine interface, distributed processing, etc.) to produce a new generation of software that can, based on the user’s preference, perform tasks for that user." (Roesler & Hawkins, 1994)

Applications

There are many applications that make use of intelligent agents. This range includes personalized information management (such as filtering email), electronic commerce (such as locating information for purchasing and buying), and management of complex commercial and industrial processes (such as scheduling appointments and air traffic control).

These tasks/applications can generally be grouped into five categories:

Information Processing

In the age of computers, information is readily available on the Internet, whether they are useful or useless. There is so much data available that we often claim to be "overloaded with information". Having too much data can cause as much trouble as having no data, as we must shift through so much information to get what we need. We can categorize this information overload problem into two divisions:

In this paper, we shall focus more on the information-filtering problem. As was described previously, users can be easily overloaded with information. Users need a way of filtering this data into a more manageable situation.

"Knowledge workers" (such as managers, technical professionals, and marketing personnel) need information in a timely manner as it can greatly affect their success.

Tasks performed by administrative and clerical personnel can be automated thereby greatly reducing the labor costs and increase office productivity. Today, labor costs are estimated to be as much as 60 percent of the total information delivery costs of sales, service, and support organizations.

Tasks that are redundant or routine need to be minimized by some individuals that can otherwise spend their time more productively. (Roesler & Hawkins, 1994)

Some companies receive so much email that they have to employ clerical worker to sift through the flood of e-mail, answering basic queries and forward others to specialized workers. A recent survey conducted by the Institute for the Future in Menlo Park, Calif., showed that 71% of 972 Fortune 1,000 workers interviewed felt overwhelmed by the number of messages they send and receive each day. Some companies have spent $50,000 to set up an intelligent agent server to filter email. Although these systems cost a lot, users are confident that they will recoup costs by lowering head count. A software company that uses the Aptex product estimates its return on investment to be one year or so because it will avoid hiring several highly paid support personnel. Amoco is using GrapeVine for Lotus Notes from GrapeVine Technologies Ltd. In Troy, Mich., which filters and routes much of the data coming in off the wires. "It helps us get through the information pollution," Joe Jesson, a staff consultant at Amoco said. The software reads a pre-established "knowledge chart" to determine who should receive what mail. "We think we can cut those 200 messages down to a more manageable dozen," he said. (Cole-Gomolski, 1997)

‘There’s a sea of information, but it’s a sea of information that you feel like you’re drowning in.’ As information becomes more available, ‘It becomes more and more crucial to have strong editors filter that information’, Seiken said." (Webb, 1995)

Even the CEO of Verity, Philippe Courtot agrees.

"There is so much content out there," said Courtot, "that the tools that filter content are going to be as important as the content itself." (Wingfield, 1995)

An end user, required to constantly direct the management process, is the contributing factor to information overload. But having agents to do the tasks such as searching and filtering can ultimately reduce the information overload to a degree. The idea is so compelling that many projects are directed at doing exactly this.

Maxims

Maes (1994) describes an electronic mail filtering agent called Maxims. Maxims is a type of learning agent. The program ‘learns to prioritize, delete, forward, sort, and archive mail messages on behalf of a user’. The program monitors the user and uses the actions the user makes as a lesson on what to do. Depending upon threshold limits that are constantly updated, Maxims will guess what the user will do. Upon surpassing a degree of certainty, it will start to suggest to the user what to do.

NewT

(Maes, 1994) also describes an example of an Internet news filtering program called NewT. This program takes as input a stream of Usenet news articles, and gives as output a subset of these articles that is recommended for the user to read. The user gives NewT examples of articles that would and would not be read, and NewT will then retrieve articles. The user then gives feedback about the articles, and thus NewT will then be trained further on which articles to retrieve and which articles not to retrieve. NewT retrieves words of interest from an article by performing a full-text analysis using the vector space model for documents.

The Zuno Digital Library

A digital library is collection of data that is organized and managed together with services to assist the user in making use of this data. The Zuno Digital Library (ZDL) system is a multi-agent system that enables a user to obtain a single, coherent view of incoherent, disorganized data sources such as the World Wide Web. Agents in ZDL play one of three roles:

Consumer agents in the system are responsible for representing the user’s interests. They maintain models of users, and use these models to assist them, by proactively providing information they require, and shielding them from information that is not of interest. ZDL thus acts both as an information filter and an information gatherer. (Jennings & Wooldridge, 1997)

Personal Internet Newspaper

Bolt Beranek and Newman Inc. has developed software that filters data available on the Internet and delivers manageable information to users’ desktops. BBN’s Personal Internet Newspaper, or PIN, uses intelligent agent technology. PIN sits on a server accessible via any World Wide Web browser. Users request information, using keywords. The requests are then processed with similar inquiries and agents are dispatched to conduct searches and obtain results. (Knowles, 1995)

BBN isn’t the only company using intelligent agents to filter. Some more news filtering intelligent agents include "NewsEDGE" from Desktop Data, "PowerNews" from Pentekk Technologies, "First!" from Individual, Inc. and "HeadsUp" from Individual, Inc.

Other Developments

Other developments using intelligent agents include the "[PersonaLink.sup.SM] from AT&T, a cooperative multiagent network platform and service based on [Telescript.sup.TM] technology. Telescript is a communications-oriented programming language developed by General Magic Inc. … Programs written in Telescript are interpreted and executed by a software program called the Telescript engine. Telescript engines may reside anywhere in the PersonaLink network. A Telescript agent is a mobile program that travels between Telescript engines into, out of and around the network. Telescript agents can perform many personalized functions from filtering your electronic mail to shopping for goods and information on your behalf. To execute such tasks, Telescript agents can cooperate with other agents, clone themselves when needed and be customized by users." (Roesler & Hawkins, 1994)

Information overload is a problem of the world today, but intelligent agents help reduce this problem. Using them to filter the oncoming "traffic" of the "information highway" can help reduce cost, effort, and time. Yet the development of intelligent agents is still in its infancy. As it gains in popularity and use, we can expect to see more sophisticated and better-developed intelligent agents.