Information Association as the Foundation of Intelligence: A Constructive Reduction at the Level of Cognitive Architecture
Year: 2026
Type: preprint
Abstract: The formation, strengthening, selection, composition, and variation of information associations are the foundation of the functional characteristics of intelligence. These five operations are sufficient to learn those characteristics from finite experience within finite resource budgets, and we conjecture that all intelligence depends on them. We develop this hypothesis as a claim about cognitive architecture and give it a boundary. An operation is associative when it acts through graded strengths that experience modifies by local co-occurrence or error-driven updates, and when it selects by graded similarity or context. It is non-associative when its effect is fixed by the exact identity or structure of symbols, independently of any learned strength, as in unification, variable binding by exact name, and address arithmetic. With this boundary the hypothesis has definite rivals: classical symbol-processing architectures and nativism about domain-specific structure. The fact that any finite computation can be implemented associatively therefore does not settle it. We give a constructive reduction of twenty functional characteristics. Each is graded separately for evidence that a mechanism exists and for evidence that the characteristic is acquired from experience, in brains and in machines, and the priors of the systems cited as evidence are counted. We answer the systematicity, propositional-structure, read–write memory, and triviality objections, and derive predictions that separate the hypothesis from its rivals. We state falsification conditions that a finding could actually meet, and we audit candidate counterexamples, including non-synaptic memory and program induction. We also report computational illustrations, including predictions that failed. Preprint. Not peer reviewed. This record contains the author manuscript submitted to Frontiers in Psychology (Theoretical and Philosophical Psychology), together with Supplementary Material 1 (technical report) and Supplementary Material 2 (technical appendix). Submission does not imply acceptance or endorsement by the journal. Supporting data: https://doi.org/10.5281/zenodo.24040550.
Author: Yang, Dekun
Institution: Zhejiang University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225582970