What the loop carries: Comment on "Loops all the way up: Recurrency as an implementation-first primitive for consciousness" by Zefan Zheng et al.
Zenodo (CERN European Organization for Nuclear Research) October 19, 2026 preprint DOI: 10.5281/zenodo.23478413 (opens in new tab)
Study at a glance
AI-extracted from the abstract| Characteristics | Commentary |
|---|---|
| Key points | Argues that the boundary between mere recurrence and consciousness-relevant recurrence depends on what the loop represents, not closure depth alone, and proposes that self-modelling be treated as a dimension cutting across four levels of recurrency and tested by holding closure depth fixed while varying self-model write-back. Also contends that one sentence in the reply to the unfolding argument contradicts its accepted premise and that claims about von Neumann hardware simulating recurrence may conflict with the framework's perturbational tests. |
Abstract
Zheng et al. (2026) propose recurrency as the implementation-level feature shared by the major theories of consciousness, and leave open where along the continuum of causal closure it becomes sufficient, and by which measure. Central pattern generators time each next state from the circuit's own trajectory, so on the glossary's measure they are deeply closed; the article excludes them because they sustain no perceptual hypothesis and represent nothing. One formalisation the glossary names, non-trivial information closure, draws the same line by asking whether the closed dynamics encode the environment. The criterion therefore turns on what the loop represents. The Four-Model Theory adds one condition: recurrency becomes relevant to consciousness when the loop re-conditions a world-model that contains a model of the system itself, and the self-model's output changes the system's next state. Self-modelling can then be treated as a dimension that cuts across the four levels of recurrency, and tested with the article's methods by holding closure depth fixed and varying only whether a self-model's output is written back. Two further points concern the reply to the unfolding argument, where one sentence contradicts the premise it accepts and either of two rewrites restores consistency, and the claim that systems run on von Neumann hardware only simulate recurrence, which can disagree with the perturbational tests the framework proposes for artificial systems. Changelog v2 **v2 (2026-10-19)** Supersedes v1 (2026-10-19). *Corrections from a cross-check of all FMT publications.* A cross-check of the FMT corpus against the master paper and against itself led to these corrections: - Model kinds (Section 3). The sentence describing FMT's taxonomy now reads "two scopes of model, the world and the system itself, each in an implicit and an explicit mode, which gives four model kinds", so that the four are named as kinds of model, not as separate models. - Title. A stray hyphenation mark in "Recurrency" in the title was removed. No argument, claim or reference changed. The reference to the Four-Model Theory already cites its concept DOI (10.5281/zenodo.18669891), which resolves to the current version.