Intelligent Systems And Their Societies Walter Fritz

Overview of the General Learner


The General Learner (GL) can be mentally divided into three parts; The senses (input), the brain (main cognitive architecture) and the actuators (output). The senses get information from the environment and send them to the brain as elementary sensations (expressed in integers, two bytes). The actuators get elementary actions from the brain (also expressed as integers) and perform them.
The type of elementary sensations and elementary actions available determine in a large measure the intelligence the system can show. In the present GL, the elementary sensations are keystrokes and the mouse. Elementary actuators are the drawing on the screen of characters, straight lines, curved lines and erasing. The brain has no way to know what the input bytes and output bytes mean (for human beings). So its functioning is independent of this (human) meaning. Any senses or actuators could be attached. (Anything learnable with these senses could be learned).

The brain of the GL has as main objective the pleasing of the person, operating the computer. The person expresses approval and disapproval by pressing the up or down arrow on the keyboard. When the person disapproves an action just done, the weights attached to each concept of the situation part of a rule become lowered, and with that its future use becomes less likely.

The brain takes the input bytes and assigns to each group of a repetitive pattern, an (elementary) concept. The first concept is assigned 10001 the next 10002 and so on until 20000 (in the GL'95). These numbers do not represent concepts, they ARE concepts because they are used whenever a manipulation of the concept is needed.

The concept has as content all the integers of the group (integers above 10000). We call it also a total concept because it has parts. Later, abstract concepts are created and given a number. Its contents are the numbers of the concrete examples (also concepts). Furthermore higher level total concepts are build up, having lower level concepts as its contents.

Now the brain expresses the present situation as a series of concepts. If it already has response rules in its memory, and any is applicable to the present situation, it sends the action part of that response rule to the actuators. This action part is also expressed in concepts. If it has no response rules, or none is applicable, it does nothing.

It continuously observes its environment (its incoming integers) and what actions the person does. In the GL'95 the representation of sensations and actions is symmetrical. The sensation of a word produces a concept. Sending that concept to the actuator, produces the word on the screen. Likewise the sensation of a line produces a concept and sending out that concept produces a line on the screen.

In future GL's, like a hand - eye robot, that would not be the case. The brain would have to try out actions by chance, and observe the resulting change in the environment. This is very similar to what a baby does: It makes haphazard movements and observes. It makes chance sounds and observes the effect on the environment. Pretty soon movements and sound become purposeful.

The GL stores the situation expressed in concepts, and the action, also expressed in concepts, as a response rule. In moments of external inactivity it generalizes concepts and response rules, to make them more universally applicable.


For continuous reading, like a book - continue here
Jump to the General Learner document /Artificial IS document /e-book Contents.

Last Edited 11 April 2013 / Walter Fritz
Copyright © New Horizons Press