Logical agents

Here is an introductory description of knowledge-based agents (or simply knowledge-based agents). The concepts considered here (knowledge representation and reasoning processes that link knowledge with reality) are central to the entire field of artificial intelligence.

It is quite obvious that people know a lot and have the ability to reason. In addition, knowledge and reasoning are very important for artificial agents, since they ensure the formation of successful behaviors that would be very difficult to achieve in a different way. In this chapter, it will be shown that knowledge of the results of actions allows agents who solve problems to successfully operate in complex variants of the environment. In contrast, reflex agents were able to find a path from Arad to Bucharest only thanks to blind luck.

However, knowledge of agents solving problems is very specific and not flexible enough. A chess program is able to calculate the allowable moves for a king of the color for which she plays, but does not possess many other useful information, for example, that no single piece can stand on two different cells at the same time. Knowledge-based agents are able to take advantage of knowledge expressed in very general forms, combining and recombining information in accordance with a myriad of external conditions. Often this process can be very far from the needs of the current moment; This can be compared to the way a mathematician proves an abstract theorem or an astronomer calculates the expected duration of the existence of the Earth.

In addition, knowledge and reasoning play a crucial role when you have to act in partially observable variants of the environment. A knowledge-based agent is able to combine general knowledge with the results of current perceptions in order to be able to identify hidden aspects of the current state before choosing actions. For example, a therapist diagnoses a patient (that is, identifies a disease that is not directly observable) before choosing a treatment. Some of the knowledge used by the therapist is in the form of rules derived from textbooks and teachers, and another part is presented in the form of associative images that the therapist cannot always describe in words. But if these associations are in the therapist’s mind, they are also in the field of knowledge.

Understanding natural language also requires the identification of hidden aspects of the state, in particular the speaker's intentions. Hearing the phrase: "John saw the diamond through the window and was eager to get it," we know that the word "him" refers to a diamond, and not to a window; we reason, perhaps even unconsciously, with the help of our knowledge of the relative value of these subjects. Similarly, having heard the phrase: "John threw a brick out the window and broke it," we understand that the word "him" refers to a window. Reasoning allows us to cope with almost an infinite number of forms of expression of thought, using the finite stock of everyday knowledge. Facing this kind of ambiguity, problem solvers have difficulty because the method of representing problems with unforeseen situations used in them causes an exponential increase in the number of options considered.

An equally important reason for studying knowledge-based agents is that such agents are characterized by considerable flexibility. They are able to accept new tasks in the form of clearly defined goals, they can quickly achieve competence by receiving instructions or learning new knowledge from their own environment, and they are also able to adapt to changes in their environment, updating relevant knowledge.