Conceptron: The Dawn of Conceptual AI?
Ron Pisaturo, founder of Conceptron
Inductica is a proud supporter of Conceptron, a startup seeking to revolutionize AI. Conceptron has an entirely new approach to representing knowledge in which the meanings of words are represented with measurements, not with embeddings representing statistical patterns of word usage (like in an LLM) or with connections between words (as in a knowledge-graph.)
Real world measurements, such as length, mass, color, and many more are represented as dimensions in a many dimensional hyperspace. Concepts, like “apple” or “dog,” are represented by ranges of these measurements. The Concept “apple,” has the measurements mass, hue, sweetness, sourness. The value of each of these measurements for any given apple is variable, but its particular values must fall within a certain range. As a result, the concept “apple,” is represented as a volume in this many dimensional space defined by the ranges of each of its measurements. Likewise, the concept “dog,” has measurements of mass, max movement speed, sound volume and many others. As a result, the concept dog is also represented as a volume in this same N dimensional space, but one that extends into a different set of dimensions than does the concept “apple.” Individual dogs, with their specific masses, movement speeds, sound volumes, etc are represented by points in this hyperspace. Thus individual things are represented by points in this hyperspace and categories of things, such as concepts, are represented by volumes.
By using the mathematical principles of high dimensional spaces, Conceptron is able to process the relationships between these concept volumes in a way which carries out logical operations. Because these operations are based on actual information about the concepts and individual things they represent, the ambition is that Conceptron will enable a much more flexible, context-sensitive form of machine learning than LLMs, which use statistical patterns of how words occur with one another, or knowledge graphs, which form ridged linkages between words.