NARS as an AGI

Capability and limitation

1. What have been done

A full specification of NAL has been given in Non-Axiomatic Logic: A Model of Intelligent Reasoning.

The expressing power and inferential power of the logic have been under testing in various domains.

In the publications on this project, various aspects of the model have been discussed and compared to other models.

The functions covered in the lectures are all implemented, with source code, demonstration, documentation, and working examples available online.

2. What can be done

More testing on NAL are needed to fully understand this logic, which may need to extensions and revisions.

On the control part, there is still a need for a theory of "thought economy" that provides a solid foundation for the design decisions.

The model will be connected to various sources to automatically acquire knowledge.

Various NARS+ systems should be experimented to explore the integration of this model with other hardwareand software.

There is a need for educational theories and procedures to raise the system according to various needs.

Practical applications: The model will be useful to situations where adaptation under AIKR is desired.

3. What will not be done

NARS is not an attempt to accurately duplucate the structure of human brain or the behaviors of human beings.

NARS will not become a "universal problem solver" that is omniscient and omnipotent.

NARS will not necessarily provide optimum solution to any specific domain problem.

4. Toward a general theory

Common usage: normal humans are "intelligent", while the other things are not (so much).

Intelligence is a matter of degree, but there is still a boundary to draw. Too wide or too narrow are both bad.

Forms of intelligence: human, artificial, animal, collective, alien, etc.

What are there in common? What is missing in ordinary computers?

Adaptivity, flexibility, originality, robustness, ...

What is unintelligent? Not incapable, but fixed response.

A General Theory of Intelligence


Reading