Automating the Process of Generalization

Abstract

As various scholars have noted, the behavior of potentially any system can be represented in terms of network graphs called state spaces. Assuming that machine learning solutions can be represented as the automation of cognitive processes, it is noteworthy that in the case of the cognitive system it has been hypothesized that all thoughts (concepts and the reasoning that connect them) can be represented in terms of one proposed state space. In the case of the physical world up to and including the entire physical universe it is hypothesized that systems can be represented in terms of another proposed state space. Biological and other systems have also been hypothesized to be representable this way. Through simple, easily verifiable geometrical arguments, this paper explores how general problem-solving ability relates to the volume of this cognitive state space an intelligent system is able to navigate, or the volume it can navigate in the state space of any system the intelligent system might be targeted at solving problems about. Through those same simple geometrical arguments in state space this article explores why modeling problems in state space, in addition to increasing the ability to generalize until it spans the entire state space, is predicted to lead to an exponential increase in general problem-solving ability for intelligent systems, and this paper explores the patterns through which this increase in ability to generalize have been automated so that they can be more broadly replicated to increase narrow problem-solving ability in a sufficient range of areas for an exponential increase in general problem-solving ability to be achieved.