By crossing these different attributes, it is possible to identify at least four agent families:
- Mobile agents. Their principal characteristic is their capacity to move around information networks in order to perform precise tasks, independently of the end-user. But mobility constitutes a necessary, but insufficient, condition to be able to define an agent as "intelligent". It must also possess sufficient autonomous capabilities to be able to know where it must go to find information, and collaborative capabilities to be able to cooperate with other agents (figure 2.12). A typical example is an application that leaves a client system in order to execute itself on a distant server. The company General Magic contributed significantly to spreading this category of agents, at first using the Telescript language and then adapting them to the Java language. IBM developed "aglets", a kind of mobile agent in Java, which travel on the Internet looking for information. Little by little, Java is becoming a language more and more adapted to the development of mobile agents.
- "Intelligent interfaces" have significant autonomous capacity and also a certain capacity for learning. Pattie Maes (MIT) defines them as personal assistants able to collaborate with a user in the same workspace. These "intelligent interfaces" cooperate more with the end-user than with other agents, but this is not an absolute rule. They can record personal behavior in order to suggest different alternatives for the user. Learning occurs either through observation and imitation of the end-user, or through the reception of their instructions.
The applications are fairly varied, and range from expert systems, which assist the user in installing, configuring and retrieving information within their computer, to agents that indicate whether you have already visited a particular Web site (Webbie from lBM).
Firefly was a complex intelligent interface which made it possible to get music and movie recommendations from agents able to establish users’ psychological profiles (behavior, tastes, practices...). Accordingly, based on your profile and after having interacted with other individual agents representing people with a similar profile, the agent provided you with suggestions corresponding to your interests. This filtering technology (Automated Collaborative Filtering) is used on certain Web sites where information is presented in a personalized way, based on your interests. It is also applied in electronic commerce in order to provide a personalized greeting and to adapt commercial offers to each visitor.
- Profile management agents. This category has recently taken off because of increasing requests for agents capable of optimizing information retrieval and monitoring on the Internet (information agents). Their role is to seek, process, and monitor information from various sources. In general, they react according to the information provided by the user (capable of learning) and they can suggest actions to be undertaken (capable of collaboration). It is important to stress that within this group, the attributes of the agents can vary appreciably. We will therefore see agents that are static or mobile, co-operative or autonomous, etc.
- The true intelligent agent should have all of these attributes. Equipped with significant autonomy, it can decide to leave a network (the Internet or other) and possibly to interact and collaborate with other agents, with an objective of bringing back relevant and reliable information. It is the dream of every information researcher. Unfortunately, such an agent has not been born yet ...