How does an agent work ?

Without going into too much technical detail, it is nonetheless interesting to understand the basic logic.

An agent doesn't work spontaneously. It is created to link a number of tasks based on a programming system which can be compared to that of a tape recorder. The difference is that the tasks to be carried out are more complex and various stimuli can activate the agent : firstly, it can be made to work by a person, which is the simplest case. It can also start working at regular intervals (for example, every two hours). Finally, it can react to exterior events, on condition that it has been given the data to allow it to understand the exterior events (for example, when someone connects to Internet in France, check whether it's a man or a woman ; if it's a woman, display advertising for Lancôme eau de toilette, etc).

The degree of complexity of the tasks carried out by the agent depend on its intelligence, which is itself divided into three types of intelligence :

Static intelligence

When it is active, the agent can carry out a series of standard tasks, which is possible due to its intelligence, which I have defined as static : each agent presents certain characteristics or attributes which allow it to take on a series of tasks by following a method of working. It also needs a certain amount of information to complete these tasks in different contexts (initial knowledge).

Dynamic intelligence

The agent's initial programme, created by the software publisher, must be completed by a process of personalisation by the user, who must be able to select a series of tasks which correspond to his particular needs, enter information which enables the agent to carry out new tasks (depending on its original characteristics) and improve its knowledge. Finally, the agent may be equipped with an automatic learning process (again, if its attributes allow it) to carry out new tasks or modify the execution of existing tasks. An automatic update is increasingly incorporated into the agents, as the associated technologies evolve very fast.

Contextual intelligence

When the agent has to carry out a task, a certain number of unforeseen events can take place and prevent the work being done. The more intelligent the agent, the more it will be able to adapt to the context, by analysing the data received. The most sophisticated agents will be able to improvise (by adopting an operating mode which corresponds to the situation).

Thanks to its intelligence, the agent generally provokes three types of reactions when it is activated : most of the time, it simply carries out the tasks for which it was created (analysing the traffic on a website, referencing a website with several search engines, proposing an offer aimed at an individual ...). Sometimes it can go as far as learning automatically : thus, having observed an individual's reaction to an advertising display, it draws its own conclusions as to the possibility of displaying the same type of information again. Finally, when the agent encounters problems (unforeseen or critical circumstances) it asks for human help.

next