Making Artificial Intelligence Intelligent - Solving the Control Problem for Artificial Neural Networks by Empirical Methods

Abstract

The Graph Model of Combinatory Logic with its elements, the Combinators, is an algebraic representation of neural networks, both for natural and artificial nets. Solving the Control Problem leads to intelligent behavior in the sense of teaching, learning, and conceptualization. This requires the construction of specific fixed-point combinators that implement feedback loops based on empirical sensing. This paper explains the role of the control problem, how to solve it and how to algebraically construct these fixed-point combinators. It proposes a blueprint and technical design for intelligent AI-supported systems that can teach themselves new skills. This is an alternative to deep learning, which requires large training sets.

Keywords: Skills Localization, Combinatory Logic, The Graph Model, Neural Networks, Artificial Intelligence, Control Problem, Empirical Software Programming

DOI: 10.54941/ahfe1005543