Intelligent agent can understand the environment, process the perceptions and respond rationally tending to maximize an expected result. If you a bit newer with these terminologies, some previously posted article might help you to understand the depth of Intelligent Agent, like Strong Artificial Intelligence, general idea on Artificial Intelligence (AI) which might help for better understanding of Strong Artificial Intelligence (SAI), Machine Learning and Artificial Intelligence, Artificial Neural Network (ANN), Collective Intelligence and Cyborg.
Basics of the idea of Intelligent agent
In this context, Intelligent agent has rationality has the feature that more specifically, tend to maximize an expected result. This concept of rationality of Intelligent agent is more general and therefore more appropriate that intelligence (= understanding) to describe the behavior of an intelligent agent.
An intelligent agent can be a physical or virtual entity. While the term Intelligent agent refers to rational artificial agents in the field of Artificial Intelligence, it can also be animals including human.
An Intelligent agent is described schematically as a functional abstract. This is why, an intelligent agent sometimes called Abstract Intelligent Agent to distinguish them from computer systems and biological systems. Some definitions of intelligent agent emphasize autonomy and they prefer the term autonomous intelligent agent. Still others considers a goal-directed behavior as the essence of Intelligent agent and prefer a term “rational agent”. In computer science the term intelligent agent can be used to refer to a software.
The conduct of an Intelligent agent is usually not optimal
Paradoxically, the behavior of an Intelligent agent is rarely optimal. The reason is simple; calculation of the optimum criterion can reasonable be very difficult when the problem to multiple constraints. An example is the calculation of the best wing for an aircraft, where the Intelligent agent should be able to take into account criteria as diverse as aerodynamics, compatibility with other components of the aircraft, or economic criteria and constraints such as limitations own weight of the wing, the total weight of the aircraft, the applicable regulations, etc.
When the criterion is a real function of many variables and constraints, the calculations are much more complicated. Sometimes you can get a good approximation, but if the agent has to decide very quickly, it should settle for the best approximation we can calculate in the limited time available.
For human, its easy (we are taking human as Intelligent agent) as we need not to really understand our neural networks for day to activities – that is what we say reflex.