Simplification and integration in computing and cognition: the SP theory and the multiple alignment concept

The main purpose of this article is to describe potential benefits and applications of the SP theory, a unique attempt to simplify and integrate ideas across artificial intelligence, mainstream computing and human cognition, with information compression as a unifying theme. The theory, including a concept of multiple alignment, combines conceptual simplicity with descriptive and explanatory power in several areas including representation of knowledge, natural language processing, pattern recognition, several kinds of reasoning, the storage and retrieval of information, planning and problem solving, unsupervised learning, information compression, and human perception and cognition. In the SP machine – an expression of the SP theory which is currently realised in the form of computer models – there is potential for an overall simplification of computing systems, including software. As a theory with a broad base of support, the SP theory promises useful insights in many areas and the integration of structures and functions, both within a given area and amongst different areas. There are potential benefits in natural language processing (with potential for the understanding and translation of natural languages), the need for a versatile intelligence in autonomous robots, computer vision, intelligent databases, maintaining multiple versions of documents or web pages, software engineering, criminal investigations, the management of big data and gaining benefits from it, the semantic web, medical diagnosis, the detection of computer viruses, the economical transmission of data, and data fusion. Further development of these ideas would be facilitated by the creation of a high-parallel, web-based, open-source version of the SP machine, with a good user interface. This would provide a means for researchers to explore what can be done with the system and to refine it.

💡 Research Summary

The paper presents the SP theory, a bold attempt to unify artificial intelligence, mainstream computing, and human cognition under a single framework whose cornerstone is information compression. The author argues that a good scientific theory should combine simplicity with explanatory power—a principle often expressed as “Occam’s Razor.” In the SP theory, simplicity is achieved by representing all knowledge as arrays of atomic symbols (patterns) in one or two dimensions, while explanatory power is retained through the rich structure that these patterns can encode.

The theory distinguishes between “New” patterns (incoming data) and “Old” patterns (stored knowledge). Learning consists of compressing New patterns by matching and unifying them with existing Old patterns, thereby creating more compact representations. The central computational mechanism is the multiple‑alignment process, adapted from bioinformatics but repurposed so that a New pattern is aligned with one or more Old patterns to achieve maximal compression. This alignment simultaneously yields parsing, pattern recognition, and inference, and because the compression is probabilistic (based on pattern frequencies and sizes), the system can assign probabilities to its conclusions, handling uncertainty and errors gracefully.

Two software models embody the theory: SP62, which can construct multiple alignments but lacks unsupervised learning, and SP70, which adds a learning component that incrementally builds a repository of Old patterns from compressed New patterns. Both models rely on a dynamic‑programming‑style matching algorithm that is flexible yet computationally tractable, and they calculate probabilities for all derived structures.

The paper also sketches a “SP machine” – a virtual or hardware platform that would run these models. The author advocates the creation of a high‑parallel, web‑based, open‑source implementation with a user‑friendly interface, and even speculates about dedicated hardware accelerators for multiple‑alignment construction.

A tentative neural interpretation is offered: the two‑dimensional cortical sheet could be viewed as a substrate on which SP patterns are “written,” with neural connections representing the links between patterns. This provides a speculative bridge between the abstract theory and biological cognition.

In the discussion of related work, the SP theory is positioned alongside unified cognitive architectures such as SOAR and ACT‑R, but it claims a unique emphasis on compression as the unifying principle. The author acknowledges several current shortcomings: the present models handle only one‑ and two‑dimensional patterns, they lack built‑in preprocessing for low‑level perceptual features needed in speech and vision, large‑scale learning efficiency has not been demonstrated, and comparative benchmarks against existing AI methods are missing.

Despite these gaps, the author enumerates a wide range of potential applications: natural‑language understanding and translation, versatile robot intelligence, computer vision, intelligent databases, version control for documents and web pages, software engineering, criminal investigation, big‑data management, the semantic web, medical diagnosis, virus detection, economical data transmission, and data fusion. The overarching claim is that by grounding diverse AI tasks in a single compression‑driven mechanism, the SP theory can simplify system architectures, reduce redundancy, and foster deeper integration across traditionally fragmented research areas.

In summary, the paper introduces a comprehensive, compression‑centric framework, demonstrates its operation through concrete examples (sentence parsing and hierarchical object recognition), outlines existing implementations, identifies current limitations, and proposes a roadmap for open‑source, high‑performance development. If realized, the SP theory could serve as a unifying scaffold for future advances in AI, computing, and cognitive science.