Agents

This chapter should give a description of a special kind of Agents, the Software Agents. Since this work is just about this kind of Agent, all the use of the term Agent is referred to Software Agent.

Keeping in mind that the Agent technology is a very new part of computer science research, there is no overall accepted definition of an Agent. The most accepted and therefore very basically designed definition states, that an Agent is

an autonomous software entity that can interact with its environment. [OMG00, page 9]

Autonom means in the case of an Agent that it can act on its behalf. An Agent gets a special task to process and then there is no need to interact with this Agent anymore. It can take over tasks which would have to be done by the user, but which do not require a human to act. This gives the user the possibility to transfer complete tasks to the software, thus helping in automation parts of the process which formerly required interaction of human users.

Instead of continuously interacting with the user, the Agent is capable of interacting with its environment and to change its way to accomplish its tasks depending on what it has at its disposal. It should be able to react to the changes of its environment, which is very important when considering mobility.

At the same time, it is able to collect experience in doing its tasks, in order to improve its work for the next time. So it is possible that an Agent processes the same tasks in different ways depending on its former experience and on changes in the environment. An Agent can adapt itself to a new environment if necessary.

Another aspect of interaction is the exchange of information or experience with other Agents executing in the same environment or the fact of work together on one special task. This is done by the exchange of knowledge between the Agents and the use of special functions others Agents might offer. Some tasks benefit a lot when they are addressed by more then one Agent taking the advantages of distributed processing.

The use of these possibilities of interaction with the environment and with other Agents is based on the level of intelligence the Agent has. Since no user interaction is necessary, it should nevertheless be possible for the (human) user to ask for the current status of the Agent and control the Agent's further behavior. Also an Agent should provide a log book covering all actions it took to process its task.