Just What is an Intelligent Agent?
Software agents, Intelligent agents, autonomous agents, information filtering agents, matchmaking agents and buying and selling agents--these are just a subset of terms used to identify a user defined program that automates our interactions on the web. The expanding array of terms and the possibilities they bring to mind is almost dizzying. They also capture our imagination with the notion of being able to instruct a piece of software to automatically carry out user-specified tasks and report back with a useable solution. But what are intelligent agents, why are they becoming more important and how might they assist in many of our mundane, and not so mundane everyday tasks?
Intelligent Agent--Definitions
MIT Media Laboratory's Software Agents Group defines agents as "computer systems to which one can delegate tasks" noting that they "differ from conventional software in that they are long-lived, semi-autonomous, proactive, and adaptive."1 To flesh-out MIT's definition of agent technology think about some of the tasks an agent might carry out: They can make airline reservations, order new books from an online store, find out about the latest song from a favorite musician, or monitor stock portfolios. They can scour the Internet to find information for us. Some of the more sophisticated software agents under development can negotiate the purchase of raw materials for a factory, schedule factory production, negotiate delivery schedules with a customer's software agent, or automate the billing process.
Intelligent software agents act on behalf of the user to find and filter information, negotiate for services, easily automate complex tasks, or collaborate with other software agents to solve complex problems. Pattie Maes, associate professor at MIT's Media Laboratory and founder of the Software Agents Group, uses the metaphor of the personal assistant to describe these assistants. The agent collaborates with the user in the same work environment becoming gradually more effective as it learns the user's interests, habits and preferences (as well as those of his or her community).2 She also makes the distinction that, as with human-to-human interaction, the assistant isn't always initially helpful to the employer. This is largely due to the assistant's unfamiliarity with the work habits and preferences of the employer. Over time, and with additional exposure to these fundamentals, the assistant gradually becomes more competent and therefore useful.
What is it that Makes Them Intelligent?
In its white paper on Intelligent Software Agents, Reticular Systems states that software constructions must meet several criteria in order to be considered an agent. They must be autonomous--free to, and capable of, executing without user intervention. Additionally, they must be able to communicate with other software or human agents and they must have the ability to perceive and monitor the environment in which they reside.
With the understanding of what an agent is, we can now look to what makes them intelligent. Researchers in the field have different views as to what makes an agent intelligent. Most agree though that to be intelligent, agents must include the ability to operate in real-time and communicate using natural language. Along with this, they must be able to learn from their environment and be capable of adaptive goal-oriented behavior. In other words, intelligent agents need to work together on a user-specified problem when told to do so and they must be able to do this successfully in a dynamic environment. Importantly, the agent must communicate to the user, in a language he or she understands, that the task has been successfully completed or that it has been otherwise terminated.
Current Trends and Applications
When many of us hear about agents, we associate these with robots and from there it is an easy jump to robots and search engines. Generally, we know that search engines use software robots to survey the web and build their databases. There are many who believe agents, and their ability to automate search processes, are the future of search engines. With agent technology, a user might run a search and instruct the search engine's agent to alert the user when new indexed items are added to the database. This effectively takes a static system and makes it a more dynamic, user driven information resource.
There are other applications though which also lend themselves to execution by intelligent agents. For instance, agents can automate some of the more mundane tasks we perform hundreds of times a day. E-mail is an excellent example of where agents can automate work in an everyday environment. Intelligent agents can learn to prioritize, delete, forward, sort, and archive mail messages on behalf of the user. By monitoring how the user interacts with email, the agent can learn how to handle similar mail in the future.
Patti Maes expanded on this scenario by noting that if the user saves a particular electronic mail message after having read it, the mail agent adds a description of this situation and the action taken by the user to its memory of examples. Keeping track of the sender and receiver of a message, keywords in the Subject: line, whether the message has been read and so on are other tasks to track. When a new situation occurs, such as a message arriving, the agent tries to predict the user's actions based on the examples stored in its memory. The agent compares the new situation with the memorized situations and acts on this mail in a manner consistent with what the user has done in the past with similar mail.
Another interesting futuristic application is the Cartalk system described by Alexandros Moukas and Giorgos Zacharia of MIT. This use of agents would help drivers with their shopping needs and interests. Moukas and Zacharia envision automobiles as broadcast centers programmed with a set of user interests ranging from food likes to shopping preferences. Local businesses would have agents programmed to receive these broadcasts. Both consumer's and vendor's agents would act in concert to provide just-in-time shopping information to the driver. For example, the car may broadcast that the driver is hoping to find an Italian restaurant in the neighborhood, or that the driver is interested in antique shops. An appropriate vendor agent response might be: "Take the next exit, turn right on Route 111, proceed one mile and you will find an antique store that is currently open. The Blue Grotto Italian restaurant is one quarter mile south of the store."
Carnegie Mellon's Matchmaker program would seem to work hand-in-hand with a system like Cartalk. This system is being developed to help agents find each other. Matchmaker makes connections between agents that request services and agents that provide services. When the Matchmaker agent receives a query from a user or another software agent, it searches its database for a registered agent that can fulfill the incoming request. In this way, Matchmaker serves as a liaison between an agent that requests service and an agent that can fulfill requests for services.
The Internet's Future Driven by Agent Interaction
The future of intelligent agents, and our interactions with them, creates a dichotomy of promise and discomfort. On one hand, as agents become more sophisticated, larger numbers of people will have access to agent driven support staffs. This type of support might today be most commonly associated with society's elite. With the aid of intelligent agent technology, more of us will be empowered with the ability to find and use information than at any time in the past. On the other hand, the past few years have seen the proliferation of malicious viruses and worms that wreak havoc on the Internet. Will agents and agencies be developed to facilitate the continuation of this troublesome conduct and will other agencies arise to combat them? Agents also store information about their users so a natural concern is the protection of privacy. The list of positives and negatives can grow to extraordinary lengths. The one constant though is that, for better or worse, intelligent agents will play a significant role in the development of the Internet and our interactions in that environment.
1 Software Agents Group of the MIT Media Laboratory. Web page accessed: 13 May 2000. http://agents.www.media.mit.edu/groups/agents/
2 Maes, Pattie. Agents that Reduce Work and Information Overload. Web page accessed: 13 May 2000.
http://pattie.www.media.mit.edu/people/pattie/CACM-94/CACM-94.p1.html
Information Outlook Main Page | This Issue's Table of Contents | Back Issues of Information Outlook
SLA Home Page | Join SLA Now | Feedback | Search
RSS
Feedback form
