Intelligent Systems And Their Societies

The General Learner

This was a program written by the author, in Lightspeed C in 1995. We will look at this program in great detail because it is a good example that shows most of the functions of a brain. This permits us to be concrete instead of having to talk in generalizations.

We describe the system as it was. Our final objective is to create a completely general learning system. To reach it, we will have to add a number of further functions. We work by trying out the system on a number of dissimilar tasks, observing how it performs, and then adding the function that the system seems to need, to gain an additional capability.

There were two reasons to write this program:

To leave the "toy world" on the computer screen by making the computer interact with the real world: with people.
To concentrate on the learning function. This program has no concepts or response rules at the start of the run. None are programmed in. All concepts and response rules are learned during its existence.

(Enter each for continuous reading, like a book) .

Learning with no prior knowledge
of the environment

What the Program Can Learn

Comparison With Previous Programs

Details of The Program
The following sections are included for those interested in writing artificial intelligence programs themselves or wishing to know exactly how such a program works and produces its results. They clarify what we have explained about intelligent systems (IS) on a more abstract level. (See Intelligent Systems (For continuous reading, like a book - do not enter here now).)

(Enter each for continuous reading, like a book) .

Overview of the General Learner

Summary Of Processes

Objective and Senses

Concepts

Present Situation and Response Rules

Selecting the Action

Acting and Reinforcement

Storing and Generalizing and Forgetting

Results and Experiments

Some Technical Details

We have built a small, but complete, system that shows how amazingly far we can go toward creating an IS on a relatively small computer with a sophisticated stimulus - response method and with reinforcement learning.
This IS is complete with senses, elaborates the situation using concepts, chooses a response rule, and performs its chosen responses. It is completely generalized in the sense that there is nothing in the General Learner that is specific to a problem space. It can also learn a wide range of relations between the parts of a stimulus and the parts of a good response.

(For continuous reading, like a book - enter).

Limitations of The System

For continuous reading, like a book - continue here
Jump to the book Contents / Artificial IS document /
Robot Brain / Robby / top of this page.