Katie Clark and Michelle Krupa

Katie Clark (60%) and Michelle Krupa (40%)

CSC 200-201 – Expert Paper

3/14/02

Intelligence Agents

The study of intelligence agents has emerged as a more effective, practical alternative to artificial intelligence, or AI research. Alan Kay of Apple Computers and Nicholas Negroponte of MIT first wrote the concept of intelligence agents about in the 1970’s. Although no autonomous agents had been programmed, they envisioned them as an alternative to the development of AI. Throughout the 1950’s most engineers and scientists expected that AI research would bring about artificial personal assistants, who possessed an artificial brain that literally mimicked the nature of real human intelligence. Realizing that this may too ambitious a goal, agent developers hoped to program more simple software that could, for example, manage schedules, purchase items, organize documents and screen phone calls without needing to emulate human brain waves. This goal for intelligence agents was outlined in a 1987 video presentation, “Knowledge Navigator” put together by Apple (Economist, 76-77).

Intelligent agents are also known as “knowledge robots”, “knowbots”, and currently just “bots.” An intelligent agent is a software program that can autonomously accomplish tasks for a person. They are equivalent to travel agents, such that they do the research and put everything together for a person. This software program is described to have a “trigger” that is built into it and when executed the agent carries out the functions. Agents can function either in the World Wide Web or on the local machine. Some agents work directly of the web, while others need to be downloaded and then function from the local drive. Most of these agents are free to use, and can easily be found on the web.

Many researchers have figured out the best characteristics for the intelligent agents. Even though no agent contains all of these, they contain bits and pieces of them. The characteristics are as follows (All of this information was taken form “Intelligent Agents on the Internet: Fact, Fiction, and Forecast” by Oren Etzioni and Daniel S. Weld):

Autonomy: Agent takes initiative and exercise control over its own actions.

Temporal Continuity: The agent is a continuous running process.

Personality: A well-defined, believable personality that facilitates interaction with human uses.

Communication Abilities: Agents can communicate with other agents allowing for the agent to obtain information faster for the user.

Adaptability: The agents automatically customize themselves based on the preferences of its previous interaction with the user. It also adapts automatically to changes in the environment.

Mobility: Agents can move from one machine to another and access different system architecture platforms.

Presently there two predominant types of intelligence agents, most of which are located on the Internet. The first type is rule-based, meaning that it is programmed with a set of rules it can follow. BargainFinder is a good example of a rule-based agent. It helps Internet shoppers find inexpensive CD’s. The user types in the CD she wants, and then BargainFinder searches the Internet for that CD and comes back with the lowest price. BargainFinder is a fairly simple agent, and is not able to operate outside of the name of the CD’s and low numeric prices (76-77)

There are more complicated rule-based agents, however agents that use a collaborative-filtering technique can make new and adapt existing rules as it carries out different processes. Firefly is a collaborative-filtering agent that recommends music and movies to Internet shoppers. Firefly gives surveys the user’s taste when they fist sign up, asking htem to rate different artists and films. Then, the agent compares that users response to those of other users in the system and creates a profile that enables it to recommend films and movies to the user as they become available (76-77). This agent is more refined and flexible than BargainFinder, for example.

Agents are also commonly used to conduct Internet searches and data collection. For instance, Xerox’s super-search engine can “perform searches of any database, located anyplace, and [has] the ability to understand, collect, and store present and future informational needs without prompting by individuals (Wilken, 111).

While the most common use of agents to date is Internet shopping and information searches, researchers anticipate more advanced uses agents in the future. There are now agent toolkits that enable programmers to build more intelligent agents that possess a number of cognitive processes. These more advanced agents are used in military training simulation, educational software, digital personal assistants, games and entertainment (Sloman, 71). There are two approaches to designing a more “cognitively rich” agent.

The first approach involves researching the way different agents interact with each other. This method “focuses on forms of communication, requirements for consistent collaboration, planning of coordinated behaviors to achieve collaborative goals, extensions to logics of action and belief for multiple agents, and types of emergent phenomena when many agents interact” (71). Essentially, when developers utilize this method they take simple agents and combine their individual functions to create a multi-agent system that can perform a host of functions.

The second approach “focuses on the internal architecture of individual agents required for social interaction, collaborative behaviors, complex decision making, [and] learning” (71). Unlike the first approach, this research concentrates on learning how to develop a single intelligence agent that can perform the same functions of a system of multiple simple agents. Single agents with a more complex set of commands seems to be the most feasible way of creating complex, more human-like agents. However, these agents will require programs that have not yet been successfully developed. For instance, if a company developed a complex intelligence agent that acts as a personal assistant, in order to interact effectively with humans it would need “mechanisms for copying with resource limits and that interactions within these mechanisms will sometimes produce emotional states involving partial loss of control of attention, as already happens in human” (72).

Some preliminary models of intelligence agents possess the necessary short and long-term memory, voice recognition, gesture reading, and other various rules necessary to smoothly interact with human beings. One example is Rea, which stands for “Real Estate Agent”, who sells real estate on the Internet. Rea is a graphic interface that has been programmed to read gestures, meaning the agent can tell when the user moves his head or starts talking. Rea also possesses voice recognition, meaning she can talk to a potential buyer the same way a human would. Furthermore, she is equipped with a wide-range of reactions, meaning she knows to respond pleasantly when a customer says “hello” and grimaces as if offended if the user calls her “stupid” (Setton, 23).

MobilCom has a similar agent on their web site named Kim, who wears a suit and talk to customers the same as any real sales person would. These graphical interface agents are the most advanced to date and are examples of the collaborative filtering technique in one complex agent.

Eventually corporations will not be the only Internet users who utilize advanced intelligence agents to manipulate and control the Internet’s open environment. While regular Internet users use agents now to conduct searches and buy products, some day we all will be able to afford intelligence agent software that will act as our own personal assistants. Perhaps we will even be able to talk to them someday, the way a homebuyer speaks with Rea or cellular phone customers interact with Kim. The shy is the limit on intelligence agents because they are not required to simulate human brain waves, the way AI robots must. However, because intelligence agents developers are not concerned with recreating the human brain, intelligence agents may never be able to fully handle the tasks a human department manger or administrative assistant might, but perhaps for fallible humans, that is a good thing.

Works Cited

“Central Intelligence Agents,” Economist 33 (June 15, 1996): 76-77.

Etzioni, Oren, and Daniel S. Weld. “Intelligent Agents on the Internet: Fact, Fiction, and Forecast” Vol.10, No. 4; August 1995 Found in IEEE Expert

Setton, Dolly. “Invasion of the Virbots,” Forbes (September 11, 2000): 22-26.

Sloman, Aaron. “Building cognitively rich agents using the SIM agent toolkit,”

Communications of the ACM 42 (March 1999): 71-73.

Wilken, Earl. “Search agents ease information quest,” Graphic Arts Monthly 70

(October, 1998): 111-112.