|Intelligent Systems And Their Societies||Walter Fritz|
First, the General Learner (GL) creates a concrete present situation, expressed as a series of elementary concepts (Thus the present situation is a row of numbers). If it finds that these are parts of composite concepts, it creates a "composite" situation, replacing the parts (some numbers) by the number of the corresponding composite concept.
Again if it finds that a concept is a concrete case of an existing (abstract) composite concept, then it creates an abstract situation, replacing the concrete by the (abstract) composite concept.
At the start, the program's memory is empty. There are no concepts, neither elementary or composite. Only when an amount of experience exists, the GL creates composite concepts during a period of external inactivity; as we will see later on.
The present situation represents the circumstance in which the intelligent system (IS) finds itself; it represents the environment of the IS. The GL composes the present situation using concepts for words or those for drawings.
Response rules, together with concepts, are the basic building blocks of our IS. All response rules have been learned, they indicate what to do in a given situation. They contain a situation part and a response part.
The situation part is a situation, expressed in concepts. To each concept corresponds a positive or negative value. Reinforcement learning produces these values, as we will see later. The response part also consists of concepts, but each indicates an action. Both the concepts of the situation part as those of the response part, can be elementary or composite concepts.
Here is the structure of a response rule (in our program, written in the C computer language):
The GL refers to a response rule by an integer number, that is related to the location where the response rule is stored. In the present GL this is a number between -10001 and -19999. For instance a list of response rules, applicable to the present situation, is a list of these numbers.
The intelligent system learns concrete response rules by recording an occurring situation together with the action that the person made in this situation. Later, during the "sleep" period, the GL learns more general response rules.
For continuous reading, like a book - continue
Jump to the top of this document /General Learner /Artificial IS /e-book Contents.
Copyright © New Horizons Press