Nomen Meum Earl: Yet Another Route to Intelligent Machine Behavior

Abstract

A unifying machine learning algorithm is proposed, in which the same processes, data structures and memory management can be used simultaneously in divergent realms, including conversation, musical composition, and robotics. A central aspect of the project is setting up learning modules that can absorb and use relationships in the input – relationships that are neither analyzed and predicted nor perceived by the programmer. The data, internal objects, and actions available to the program exist as points in a quasi-Cartesian, multi-dimensional knowledge space. The geometry of this space is determined by semantic content. Within a combination of this space and a kind of production system one can take advantage of content-addressable memory to replace all search, one can implement table-driven program control that allows the program to change itself while running and distributed massively-parallel processing can be used.