An Agent Classification
Russell & Norvig classify agents into 4 categories, from least to most complex.
Simple Reflex (Stimulus-response) Agents
These agents react to immediate stimuli. They have no memory at all. The percept sequence is just the immediate environment sensed by the agent's sensors.
Nevertheless, they are capable of behaviour of considerable complexity, especially if they are part of a Multi-Agent system.
These agents are also referred to as stimulus-response agents [Nilsson]. We look at this type of agent in more detail later.
Note the "condition-action rules" in the diagram. These are production rules. Other representations are also possible.
Reflex agent with state
These agents have a memory of state. They can remember earlier experiences. These can be combined with current information from sensors to produce a more sophisticated response to the environment.
Goal based agents
Goal based agents can plan ahead before making their actual 'move' in their environment. They use various more or less sophisticated search methods to search a state space of potential future environments, as a chess player might do in his head before actually moving a piece. Future potential 'positions' are evaluated as to how close they are to the goals of the agent.
Utility Based Agents
At this level agents have acquired human aspects. Utility is an economic term (from micro economics) related to the concepts of happiness and personal preferences. This level of ability is beyond the present capabilities of present AI.
Информация любезно предоставлена сайтом http://www.ryerson.ca (http://www.ryerson.ca/~dgrimsha/courses/cps720/agentclass.html)