A General Theory of Intelligence

a developing eBook by Pei Wang


The preface explains the subject and format of this book.

This eBook is an attempt to establish a theory that identifies the commonality behind various forms intelligence, including human intelligence, computer intelligence, animal intelligence, alien intelligence, group intelligence, etc.

This theory is part of an on-going AI project, together with a formal model built according to the theory, and a computer implementation of the formal model.

This eBook consists of a main text and many sidebars. The main text is organized into webpages at three levels: book, chapter, and section, where each topic is summarized in a high-level file, as well as discussed with more details in a low-level file. A sidebar covers a special topic, and is linked from the main text.

Chapter 1. Information System

This chapter introduces the basic terminology used in this book to describe various systems.

To treat a system as an information system means to describe it at an abstract level, so as to omit the concrete processes underneath.

The internal structure of every information system can be analyzed in terms of its goals, actions, and beliefs.

An information system carries out its actions to achieve its goals, following the relations among them as provided by its beliefs.

Chapter 2. Intelligent System

This chapter clarifies the central concept of this theory — intelligence.

Information systems can be divided into instinctive systems and intelligent systems.

In an instinctive system, all major components and their relations are determined when the system is formed, and remain unchanged afterwards.

In an intelligent system, all major components and their relations are adaptive to the environment. The system learns new beliefs, organize actions into skills, establish new goals, all as attempts to improve its goal-achieving capability, under the assumption that in general the future will be similar to the past.

Chapter 3. Inference System

This chapter describes the structures and procedures of an inference system, as a concrete model of intelligent systems.

To discuss an information systems in a more accurate manner, it is necessary to put it into a formalization framework. For advanced intelligent systems, the framework of an inference system is more suitable than the alternatives, because of its expressing and processing power.

An inference system can be pure-axiomatic, semi-axiomatic, or non-axiomatic. Many problems in the traditional logistic AI can be attributed to their axiomatic nature, and therefore become solvable in a non-axiomatic system.

As a model of intelligent systems, a concrete inference system, NARS, is designed to be non-axiomatic. Consequently, its goals, actions, and beliefs are represented and processed in a way that is fundamentally different from how they are handled in traditional inference systems.

Chapter 4. Self-Organizing Process

This chapter describes the running and evolving processes in an intelligent reasoning system.

For a system like NARS, its running process is a self-organizing process, in which the system reorganizes its goals, actions, and beliefs, according to its experience.

Self-organization of goals forms and evolves the system's goal complex. In the process, goals are derived, evaluated, prioritized, and removed.

Self-organization of actions means to acquire new skills as "programs" of existing actions and skills, as well as to build and apply tools.

Self-organization of beliefs produces more reliable, useful, and compact summary for the system's experience by deriving and selecting beliefs.

Self-organization of concepts provides an intermediate structure between the whole memory and the individual items to improve the efficiency of the system.

Chapter 5. Experience and Socialization

This chapter describes the interaction between an intelligent system and its environment.

The sensorimotor mechanism implements the actions of an information system, and handles the direct interaction between a system and its environment. Based on it, advanced systems also have a communication mechanism, which uses a language, and is carried out by the cooperation between the system and some other systems in the environment. Sensorimotor and language processing are both based on the beliefs learned by the general-purpose intelligence of the system, though the content of the beliefs is modality-specific.

Advanced intelligent systems have a separate sensorimotor mechanism on the system itself, which works in the same way as the sensorimotor mechanism on the outside world, though using different sensors and actuators. Self-monitor and self-control provide the foundation of consciousness.

Socialization is the process for a system to adapt to a society, which shapes the system's goals, actions, and beliefs. Education is a special form of socialization, in which the society has some control over the social experience of the systems.

Chapter 6. Community and Science

This chapter treats a community of systems as an information system, and analyses its intelligence.

The goals, actions, and beliefs of a community come from those of its members, though also depend on the social structure of the system.

Community beliefs take the form of common sense, custom, religion, or science. Science is organized common experience that can be used to guide the members in the future. The development of science follows the same logic as the self-organization of beliefs in an individual intelligent system.

A scientific theory shows properties of an information system, and its internal logic is similar to the logic followed by an intelligent system.

A science of intelligence should properly clarify the notion of intelligence, so as to cover various types of intelligent systems. The theory should contribute to our understanding of the human mind, as well as to guide the building of thinking machines.

Topic List