Abstract:
As AI models become larger, replicating long term memory structures (LTM-S) may produce the same benefits that evolution provided the human brain (efficiency, performance, and extensibility).
At the heart of this paper is the conjecture that software structures are close representations of LTM-S. If this is true, then open source can be considered a huge database of easily searchable LTM-S examples that could assist in a deeper understanding of the same.
The paper proposes a general refactoring algorithm based around two elements of LTM-S, chunks and analogies. The underlying aim is to develop mechanisms and theories to analyse the analogical and chunking structures employed in software.
At the heart of this paper is the conjecture that software structures are close representations of LTM-S. If this is true, then open source can be considered a huge database of easily searchable LTM-S examples that could assist in a deeper understanding of the same.
The paper proposes a general refactoring algorithm based around two elements of LTM-S, chunks and analogies. The underlying aim is to develop mechanisms and theories to analyse the analogical and chunking structures employed in software.