Intelligent Systems And Their Societies

Walter Fritz

The Autonomous Intelligent System

This System was described in the July '89 issue of the Robotics and Autonomous Systems: Journal. A group of persons, coordinated by the author and meeting in the SADIO Society wrote the program. The names of the people in the group were: Ramón García Martínez, Javier R. Blanqué, Rogelio Adobbati, Alberto Rama, and Mario Sarno.

This program is an evolution of the previous one. It displayed a representation on the computer screen and can:

• Sense its environment through four senses
• Recognize a situation it has met before
• Focus its attention
• Generalize from two similar situations
• Establish its own list of priorities
• Try to establish a plan of response for reaching its own objectives
• If it does not find a viable plan, act by curiosity (a built in instinct)
• If that is not possible, act by chance
• Check the proposed action for undesirable side effects
• Learn (only!) by its own successful and erroneous responses
• Remember experiences
• Step forward, turn to the right or left, and do actions that symbolize eating and sleeping (When "eating" food disappears and a variable for hunger diminishes)

The general structure of the computer program is:

Perceive

Receive information from sensors

Establish the situation

Plan

Create a stack of sub-objectives from its stored experiences

Create a tree of responses

Select a plan

Act

If no plan exists, act by curiosity

If that is not possible, act by chance

If a plan exists, take the next response and do it.

Store

Store the new experience and the new situation

This program makes up a list of possible objectives. This is a list of those experiences that had the highest positive result in the past. Here the brain expresses the experience as four numbers:

situation - response - resulting situation - result.

The result is, for instance:

present situation -> situation #745

situation #745 -> situation #43

situation #43 -> desired situation.

Experiences with the program, after running it a few thousand "instances" are as follows: (Every "instance" is a loop of the program that starts with observing the environment and ends with a response).

The amount of different situations met, does not increase as fast anymore but the amount of different experiences does.

The average hunger, using plans and generalizations, is only about one fifth of that existing when using chance alone. This shows that using plans and generalizations helps to reach one's objectives.

The average hunger, when using generalizations, chance and plans, is about one third of that existing when using only chance and plans.

Adding curiosity to all other functions decreases the average hunger by about 10%.

For continuous reading, like a book - continue The authors programs
Jump to the e-book Contents / top of this page.