Researched Works

What Are Intelligent Agents

Percept Sequence Intelligent Agent

A basic type of intelligent agent is the agent which keeps in memory its entire percept sequence and uses it to index to a table with the actions listed for all possible percept sequences. There are drawbacks to this intelligent agent such as a large number of entries being needed in the table because there are so many possible sequences. Also there is no autonomy in the intelligent agent at all and therefore if the environment changed unexpectedly it could not handle it.

Reflex Intelligent Agent

Another type of intelligent agent is the reflex. It does not explicitly list all sequences in a table but has commonly occuring input/output conditions. It uses condition-action rules. For example in a driving intelligent agent: If car in front is braking, then initiate braking. However this type of intelligent agent has a narrow range of application because of its simplicity.

Keep Track of World Intelligent Agent

Similar to reflex agent is keep track of world agents, however they also have an internal state. This it to distinguish between world that give same perceptual input but require different actions to be taken. For example, some cars may have brake lights in different positions making it hard to tell for a reflex agent if the car in front is braking. However if the intelligent agent has an internal state which in this case keeps track of the previous frame from the camera then it can tell if two red light go on simultaneously and therefore detect braking. To update the internal state, the intelligent agent needs to know information about how the world evolves independently and how the intelligent agent's actions affect the world.

Goal Based Intelligent Agent

Another intelligent agent is the goal based agent. It has goal information describing desirable situations. It will choose actions by using knowledge of the results of possible actions. Finding action sequences to achieve the goals involve "search" and "planning", which are subfields of AI. This differs from condition-action using agents because it considers the future rather than just doing pre-computed actions. Hence it is more flexible since it can alter behaviours to achieve goals under differing conditions. And if the goals change then it can adapt whereas reflex agents would need to rewrite its rules.

Utility Intelligent Agent

The utility agent is another intelligent agent. There may be many ways of achieving a goal yet some may be preferrable. If one world state is more preferred than another, then it has a higher utility. The utility agent has a utility function which maps states to real numbers describing degrees of happiness. In cases where there are conflicting goals where trade offs must be made utility can specify trade offs, or in cases where there are several goals that can be aimed for then utility finds a way to weigh likelihood of success against the importance of goals.

Back to Home