Abstract:
The evolution of modern software agent technology has given rise to an extensive overuse of the term agent. The result of this abusive usage of the word agent and the terminology that relates to it, is a great confusion as to what an agent is, what types of agents can be identified, and what are the properties and characteristics with respect to which the aforementioned identifications can be achieved. This article aims to provide a reasoned approach to these questions, both by the study of the existent proposals of agent theory, and by the analysis of potential applications of the currently developping agent technology.

• Interfacing with other similar processes across the network
• Collaboration / Negotiation with such processes to achieve a goal
• Reactivity to perceived changes in its environment
• Taking the initiative to proceed into goal-directed actions, a behaviour called proactiveness
• Supporting its operation in an autonomous manner, that is, requiring minimal or no human interaction.

• Personal Assistants : Self-Contained systems that schedule, organise and optimize personal productivity tasks. Possible applications include mail filtering and classification systems, meeting schedulers and miscellaneous information filters.
• Visitor Hosting Systems where the hosting and scheduling of activities, meetings and events participated to by visitors at a site, are organized and coordinated through the cooperation of local agents with the visitors' personal agents. An example of such a system is the Visitor Hosting System at Carnegie Mellon University.
• Data Mining where information-specific agents provide a context for data searches in vast databases or other information sources (like the web for example) from which, cooperating intelligent personal agents will extract a selection of useful information. This field is one of the fastest-evolving ones at the moment given the explosive growth of the amount of accessible information via networks and communications.
• Network Management where collaborative agents collect and exchange local information on network statistics in order to achieve automation and optimization of decisions on network administration tasks like routing, access, service provisions, monitoring and statistical evaluation, within a global view.
• Air/Land Traffic Control in which a multi-agent system is installed in both controlling sites and travelling vehicles to help resolve control decisions in routing and scheduling.

autonomy

the ability to self-support its operation without the intervention of human or others, as well as the ability to retain control over its actions and internal states

social ability

the ability to interface and interact with other agents via a communication language

reactivity

the behavior according to which perception of environmental changes results to a response by the agent in a timely fashion

proactiveness

the ability to take the initiative and thus exhibit, self-contained goal-directed behaviour

" [Agents are] software components that communicate with their peers by exhanging messages in an expressive agent communication language. "

mobility

By mobility, we mean the capability of an agent to move around some network. So we may class agents as static or mobile

behavioural model

We may also class agents considering the way they model their environment and internal states. Deliberative Agents possess an internal symbolic reasoning model and use it for planning and negotiation to achieve coordination. Reactive agents do not have any internal representation of their environment and use a stimulus-response behaviour model.

primary attributes

Other ideal primary attributes of agents can be used to classify them. Nwana proposes three: autonomy , learning and cooperation . Of course all agents will exhibit all these three at least at a minimal degree. But the level of sophistication that an agent exhibits in each of these, will allow us to classify agents as collaborative , colaborative learning , interface and smart .

roles

A last and less well-defined criterion is the role of an agent. Identifying agents by their role is a method subject to interpretation, but we can roughly identify some major roles of general nature, such as information agents (e.g. internet search bots) or control agents (e.g. train traffic control systems).

• Situations where the problem is too large for a single agent and must be distributed due to resource limitations.
• Situations where there is a need for interconnection of multiple existing systems
• Problems whose nature is itself distributed.
• Problems whose solution requires accessing a distributed set of resources
• Situations in which the problem-solving capacity is distributed
• Situations in which the enhancement of modularity is of prime interest.

• Personal Assistants/Organizers
• Preferential Information Filters
• Financial Decision Systems
• Entairtainment
• Miscellaneous Aids
• Tutorial Systems

• Situations where local resources are limited (CPU or storage for example) and there is a need to use remote resources to complete tasks.
• Situations where asynchronous distributed computation needs to be achieved.
• Situations where coordination is complex.
• Situations where the cost of bringing vast amount of remote information to the local node is big, thus necessitating the use of a mobile agent to set off and fetch it.