Algebraic Emergence and Effective Theories in Recurrent Neural Circuits

A classic claim in emergence is Philip Anderson’s “more is different”: novel structure appears at scale that is absent in the parts. A new preprint by Nima Dehghani puts that claim on algebraic footing for neural circuits. It represents canonical recurrent motifs, divisive normalization and winner-take-all competition, as finite transformation systems, and shows that composed circuits generate algebraic structure that none of the primitive components possesses.

The paper “‘More Is Different’ in Neural Circuits: Algebraic Emergence of Effective Theories in Canonical Recurrent Motifs” was posted to arXiv as 2608.30231. The analysis distinguishes structure already present in a single generator from structure that appears only through composition, and structure inherited from one factor from structure that lives on a joint configuration.

From function description to algebra

Canonical motifs are usually described functionally. Divisive normalization rescales population activity by a pooled signal. Winner-take-all competition selects one pattern through recurrent excitation and shared inhibition. The paper represents these as transition monoids, the algebraic object generated by their input-conditioned updates, and analyzes what their composition can compute.

The central technical finding is that individually aperiodic updates can generate non-aperiodic monoids. In winner-take-all, every frozen-drive generator collapses to fixed points, yet short input sequences create local cycles of winner-dependent inhibitory gating. That is globally dissipative dynamics with a reversible action, structure a single collapse cannot produce.

The strongest result: composite group structure

The core result comes from composing winner-take-all with divisive normalization. The composed monoid contains a genuinely composite local cycle in which the normalization state and the winner’s gating state change together, although every primitive generator is aperiodic. The paper uses holonomy analysis to certify this as a genuine group component of a Krohn-Rhodes cascade, not an incidental cycle.

An exhaustive interface sweep shows the composite cycle is a property of the coupling, not of a chosen map. It is the algebraic signature of the interaction itself.

Composing circuits is programming

The paper’s framing is that if motifs are building blocks of neural computation, composing them is a form of programming. One chooses primitives and interfaces so that the generated algebra has the intended repertoire. The transition monoid is that repertoire, what a primitive presents to any later construction.

This is a substrate-independent account of what recurrent circuits can compute. It does not claim that group structure is consciousness. It claims that composition, not the size of the parts, generates the computational repertoire, and that this repertoire can be analyzed with the formal tools of algebraic automata theory.

Relevance to emergence and consciousness

The site’s coverage of emergence includes causal emergence and multiscale measures and emergence in cellular automata. Dehghani’s paper supplies a distinct, rigorously defined notion: algebraic emergence. A composite system has effective-theory structure, a group component, that no primitive possesses. That is emergence in Anderson’s sense, made measurable.

For functionalist emergentism, the relevance is that computational power is compositional and formal. The structure that matters emerges from how parts are wired, which is a property of the interface, and the same logic applies whether the substrate is biological or silicon. That supports the project’s claim that consciousness as an emergent property does not require a specific material.

Comparison to The Consciousness AI

The Consciousness AI project’s Neutral Core uses oscillatory binding and recurrent dynamics. The algebraic framing here is directly relevant to how the core substrate reasons about whether a recurrent implementation “has” a computational property: the property is a property of the composed monoid. The project’s functionalist emergentism, set out on the functionalist emergentism page, treats computational organization as the carrier of mind, which is exactly the level Dehghani analyzes. How that compositional view compares with the rest of the field is surveyed in the current scientific consensus on AI consciousness.

Limits

The paper is a theoretical analysis on canonical motifs, not an experimental study of brains. Transition monoids describe what circuits can compute, not what they do compute. The bridge to biological cortex is plausible but not established. What it provides is a precise vocabulary and a formal result, algebraic emergence, that the field can test in real networks.