Loops All the Way Up: Why Consciousness May Live in the Architecture, Not Just the Algorithm

The Loop Beneath the Theory

For thirty years, the science of consciousness has been dominated by a search for the right algorithm. Integrated Information Theory asks how much information a system integrates. Global Workspace Theory asks which contents get broadcast. Higher-Order Theories ask whether a state is represented by another state. Each proposes a different computational signature for experience — but each, in its own way, treats the brain’s wiring as incidental, a mere substrate on which the “real” theory of mind runs.

A new theoretical review published in Physics of Life Reviews by Zefan Zheng, Roman Chis-Ciure, Peter Thestrup Waade, Annika Eiserbeck, Jaan Aru, and a formidable group of co-authors including Thomas Andrillon, Lucia Melloni, Georg Northoff, and Fernando Rosas, challenges that hierarchy directly. Their argument, distilled: recurrency — the physical looping of neural signals back onto themselves — is not an implementation detail of consciousness. It may be the primitive from which consciousness is built.

This is a quietly radical move. It says: stop asking only what computation the brain performs, and start asking what physical architecture makes any such computation conscious at all. The paper’s title captures the stance with wit — “loops all the way up” — echoing the old philosophical joke about turtles, except here the loops aren’t infinite regress. They’re the actual mechanism.

What “Implementation-First” Means

Most consciousness theories operate at what computer scientists call the algorithmic level — the level of information processing, independent of hardware. The implementation level is different: it’s the physical stuff doing the computing. Marr’s classic three-level framework (computational, algorithmic, implementational) has always allowed that the same algorithm could, in principle, run on different hardware and produce the same output.

Zheng and colleagues argue that for consciousness specifically, this substrate-independence assumption may be false. Drawing on empirical work from anesthesia research, split-brain studies, and cortical microcircuit physiology, they marshal evidence that feedforward-only processing — no matter how computationally sophisticated — does not generate subjective experience. Consciousness appears reliably where recurrent, re-entrant signaling exists: cortico-cortical feedback, thalamocortical loops, and local recurrent microcircuits. It disappears or degrades when these loops are interrupted, even when feedforward throughput remains intact.

This lines up with decades of work by researchers like Victor Lamme, whose recurrent processing theory has long held that feedforward sweep alone produces unconscious processing, while conscious perception requires re-entrant signal exchange between higher and lower cortical areas. What the current paper does is elevate this specific mechanism — recurrency — above the level of “one candidate mechanism among several” to something closer to a necessary implementation-level condition, potentially common to whatever the correct algorithmic theory turns out to be.

The Evidence Base

The authors synthesize converging lines of data:

• Anesthesia studies consistently show that propofol, ketamine, and other general anesthetics disrupt long-range and feedback cortical connectivity well before they eliminate feedforward sensory responses — the brain can still register stimuli, but the loops that would normally re-process and integrate that information collapse.
• Cortical layer physiology: feedback projections predominantly terminate in different laminar compartments than feedforward projections, giving the cortex a physically distinct circuit for “looping back” versus simply relaying information forward.
• Perturbational complexity approaches — the kind of work associated with Giulio Tononi’s lab and clinical consciousness assessment — show that the brain’s capacity to sustain complex, differentiated, yet integrated responses to a perturbation (a rough proxy for consciousness) tracks recurrent connectivity, not raw processing speed or feedforward bandwidth.
• Comparative and developmental data, including work referencing Béchir Jarraya’s clinical neuromodulation research, suggest recurrent architecture emerges and can be selectively disabled or restored, correlating tightly with reported awareness in ways that feedforward metrics do not.

None of this proves recurrency is consciousness. But it does suggest that whatever the correct high-level theory of consciousness ends up being — integration, higher-order representation, predictive modeling — it likely needs recurrent loops as its physical carrier. The algorithm may vary; the loop may not.

Predictive Coding and the Loop as Inference Engine

The paper draws heavily on active inference and predictive coding frameworks associated with Andy Clark and the broader Free Energy Principle community — fitting, given Peter Thestrup Waade’s affiliation with the Active Inference Institute. In predictive coding architectures, the brain is constantly generating top-down predictions and comparing them against bottom-up prediction errors. This is, structurally, a loop: prediction flows down, error flows up, prediction updates, and the cycle repeats.

What Zheng and colleagues add is the claim that this inferential loop isn’t just a useful computational metaphor — it may be doing double duty as the actual physical mechanism generating experience. The recurrent exchange between predictive models and sensory evidence isn’t merely how the brain minimizes surprise; it may be how the brain generates a felt, unified now. Fernando Rosas’s work on integrated information and complexity science provides some of the formal tools to quantify how looped, re-entrant dynamics generate the kind of high-dimensional, temporally extended integration that experience seems to require.

What This Means for the Meditating Brain

Here the paper’s implications extend naturally into contemplative science, even though the authors don’t frame it this way themselves.

If recurrency is the physical substrate of conscious content, then the brain’s most persistent recurrent loop — the self-referential narrative network anchored in the default mode network — deserves special scrutiny. The DMN’s characteristic activity isn’t simple feedforward relay; it’s densely recurrent, looping self-models, autobiographical retrieval, and predictive narrative generation back through medial prefrontal and posterior cingulate hubs. This is, structurally, a loop generating the loop we call “me.”

Contemplative practice, in this light, isn’t a vague act of “quieting the mind.” It may be a direct intervention on recurrent architecture. Meditators trained by researchers like Richard Davidson, Sara Lazar, and Judson Brewer show measurable changes in DMN connectivity and self-referential processing with sustained practice — changes that, under this framework, look like literal reconfiguration of the brain’s looping circuitry, not just a shift in mental content.

This also offers a mechanistic angle on ego dissolution under psychedelics, a phenomenon extensively studied by Robin Carhart-Harris and Robert Carhart-Harris. If the 5-HT2A receptor-mediated disruption of DMN connectivity works partly by desynchronizing recurrent self-referential loops, then ego dissolution isn’t the erasure of consciousness — it’s the disruption of one specific, habitual loop, potentially freeing recurrent processing to organize around different, less self-centered attractors. The felt sense of boundarylessness reported in these states may correspond to a literal reduction in the dominance of the self-model’s recurrent loop relative to other looping systems in the brain.

Similarly, practices producing states of nonduality or flow — studied by Mihaly Csikszentmihalyi and others — might represent recurrent architecture reorganizing away from strong self-other, subject-object loop structures toward more distributed or transiently unified re-entrant dynamics.

Implications for the Field

For consciousness researchers, this paper is a call to stop treating implementation as an afterthought. If Zheng, Aru, Melloni, Northoff, and colleagues are right, any complete theory of consciousness will need to specify not just what information gets integrated or broadcast, but how the physical loop carrying that process is built — and that architecture may be far more constraining than algorithmic theories have assumed.

For contemplatives and researchers of meditative states, it offers something rarer: a plausible bridge between subjective reports of “loops of thought” dissolving or reorganizing during deep practice, and an actual, falsifiable neural mechanism. The felt experience of a quieting, softening, or restructuring self may not be metaphorical. It may be recurrency itself, changing shape.