The assumptions that restrain us from understanding consciousness

Jaan Aru

REVIEW 3 major objections 5 minor 3 references

The assumptions that restrain us from understanding consciousness

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper argues that consciousness is not an abstract computation: biological matter itself does part of the work, so spiking patterns alone cannot be the neural correlate and current computationalist theories are aiming at the wrong…

desk verdict A readable, honest perspective piece that asks good questions about spiking NCCs and computational functionalism, but its central challenge depends on a simulation and a notion of 'computation' that are not fully pinned down. read the letter →

arxiv 2506.21485 v1 pith:NBVEYFRN submitted 2025-06-26 q-bio.NC

Jaan Aru This is my paper

classification q-bio.NC

keywords consciousnessneuralcorrelatesofcomputationalfunctionalismbiologicalnaturalismspikingactivityaha-momentsinsightmultiplerealizability

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This review argues that the science of consciousness is held back by assumptions it rarely examines: that the neural correlates of consciousness live at the level of spiking, that consciousness is an abstract computation that could run on any substrate, and that perception triggered by external stimuli is the right place to look. The central claim is that consciousness is not an abstract computation and that matter matters for the mind: biological tissue does part of the computational work, so the same algorithm on a different substrate would not automatically be conscious. If the author is right, computational functionalism about consciousness is false, multiple realizability is severely limited, and current theories that abstract away cellular and molecular detail are searching at the wrong level. The review points to sudden internal changes such as aha-moments as a neglected window because they show consciousness reorganizing abruptly without any new external input.

What carries the argument

The argument runs on two thought experiments and one evolutionary principle. The scattered-neurons thought experiment asks whether consciousness survives when the exact pattern of spikes is replayed by neurons dispersed in space; the paper takes the intuitive 'no' as evidence that spiking patterns are not the NCC. The degenerated-computation thought experiment, backed by a computational simulation, removes the computation while leaving neural activity identical, forcing computational functionalists to say consciousness changes with no neural change. Generative entrenchment, the principle that lower-level biological features become deeply interwoven with higher-level functions over evolution, is then used to explain why no clean abstraction layer separates computation from matter. Aha-moments add a candidate mechanism: NMDA-dependent ignition, a nonlinear transition into a self-sustaining attractor, illustrates how consciousness could reorganize abruptly and endogenously.

What would settle it

Replay the exact recorded spike pattern of a percept in neurons physically scattered in space, in a preparation where the animal can still report perception; if the percept report survives, the paper's central intuition fails. Alternatively, in the cited degenerated-computation simulation, compute several independent formal measures of computation and show that all of them are zero while neural activity and behavior are unchanged; if any measure remains nonzero, the reductio against computational functionalism does not go through.

Watch

Extended reading notes

Core claim

The central discovery, argued as a thesis, is that the prevailing computationalist picture of consciousness lacks a solid foundation. Spikes became the default neural correlate largely because they were measurable and historically prominent, not because they were shown to be the right level; the paper uses a thought experiment in which the neurons producing a percept are scattered while the same spike pattern is replayed, and takes the intuitive loss of consciousness as evidence that the pattern itself is not the correlate. A second argument, drawn from a computational simulation, shows a case where the computation is degenerated while neural activity is unchanged, so a computational functionalist must say consciousness changes while the brain does not; the paper treats this as a reductio of the idea that consciousness is an abstract computation. The positive proposal is that part of the computation is done by the physical matter itself, which is why generative entrenchment and biological specificity constrain what realizations are possible, and why the field needs a computational theory of biological naturalism that states exactly which biological processes can and cannot be captured computationally.

Load-bearing premise

The load-bearing premise is that a conscious experience cannot survive the same pattern of neural firing being replayed by neurons scattered through space, and that this intuitive implausibility is reliable evidence about the nature of consciousness; if the intuition is wrong, the attack on spiking-based NCCs and on abstract computation loses its force.

Editorial extensions

If this is right

If consciousness is not an abstract computation, then building an AI that performs the right computations is not enough to make it conscious; claims of artificial consciousness need biological evidence, not just algorithmic equivalence.
If spiking patterns are not the right level for NCCs, then the search should widen to calcium waves, intracellular signaling, silent synaptic mechanisms, and interactions across scales, because the focus on spikes may be a historical artifact.
If current theories abstract away molecular and cellular detail, then they may be discarding the very components a true theory of consciousness needs; a computational theory of biological naturalism should specify which biological processes are captured and which are not.
If aha-moments are genuine endogenous reorganizations of consciousness, then any neural mechanism proposed for consciousness must be able to produce abrupt, discrete state transitions without external input, and studying insight at the cellular level becomes a test bed for such mechanisms.
If generative entrenchment applies to the brain, multiple realizability of consciousness is heavily constrained: possible realizations are narrowed to systems very close to our biology, undercutting the standard functionalist assumption.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A sharper version of the paper's case would define 'the physical part of computation' operationally; one testable extension is to look for dissociations between identical spike trains and different intracellular, metabolic, or glial contexts in conscious-report experiments.
The scattered-neurons intuition could be probed directly by building a real or simulated circuit where the same spike pattern is produced in two spatial arrangements and measuring whether downstream readers (or the system itself) treat them as the same state; the paper assumes the intuition, but it is an empirical question.
A computational functionalist could reply that the degenerated-computation simulation only shows one formalization of computation vanished; the debate would benefit from specifying which formal measure (for example, integrated information, dynamical independence, or algorithmic complexity) is the right one, and showing that all of them go to zero simultaneously.
If aha-moments are a privileged window, then designing delayed-insight tasks with intracellular or calcium imaging, where the solution emerges after minutes without new sensory input, could reveal the transition mechanism directly; the paper proposes the phenomenon but not the experimental design.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Dear colleague,

You should know this is a perspective essay, not a research preprint. Aru explicitly says he offers no answers. What the paper does well is name two implicit assumptions in consciousness science: that NCCs live at the level of spiking, and that consciousness is an abstract computation. He makes a reasonable case that both deserve scrutiny, and the suggestion to study 'aha' moments as internal transitions in consciousness is a concrete, under-explored direction. The prose is clear and the chapter is honest about its own limits.

The soft spots are in the argument against computational functionalism. The load-bearing example is a simulation by Gidon et al. (2025) that is cited but not reproduced, and the paper never defines what 'degenerated computation' means or pins down the implementation relation. As it stands, the example may only show that computational descriptions are context-relative, not that consciousness depends on the substrate. The scattered-neurons thought experiment is acknowledged to not apply to IIT, so it only targets pattern-identity views, not all NCC theories. The 'physical part of computation' remains a metaphor; the RNA-to-protein analogy suggests some physical processes are necessary for proteins, but it doesn't show which physical processes, if any, are constitutive of consciousness. The heavy reliance on the author's own prior work and forthcoming manuscripts (Milinkovic & Aru) is a limitation, though not a fatal one for a perspective piece.

None of this is a reason to desk reject. For what it is—a provocative chapter written for a Springer volume—it is a legitimate contribution to the debate. It deserves a serious referee, but the referee should push the author to either supply the Gidon simulation details or soften the claim that it shows computational functionalism fails. The chapter would also benefit from acknowledging that IIT already escapes the scattered-neurons challenge and that the burden of proof for a 'physical part' of computation needs to be stated more precisely.

I'd bring this to a reading group for discussion, but I wouldn't cite it in my own work yet; I'd cite the primary sources it draws on. My verdict: send to peer review, but expect substantial revision on the central argument.

Referee Report

3 major / 5 minor

Summary. The manuscript is an invited perspective chapter arguing that consciousness science is restrained by implicit assumptions: that neural correlates of consciousness are to be found at the level of spiking responses, that consciousness is an abstract computation, and that current theories are close to the answer. It uses a thought experiment about scattered neurons and a cited simulation by Gidon et al. (2025) to argue that computation can be "degenerated" while neural processes remain identical, concluding that consciousness is not an abstract computation and that matter matters for the mind. It then proposes studying "aha" moments as endogenous reorganizations of consciousness and calls for theories of biological naturalism. The author explicitly states that the piece offers no answers, but it does advance a negative thesis about computational functionalism.

Significance. If the central anti-computationalist claim were established, it would challenge computational functionalism and multiple realizability and would support biological naturalism. The manuscript is clearly written and usefully surfaces implicit assumptions that are rarely questioned, and the suggestion to study aha moments as a window into internal transitions of consciousness is a constructive empirical direction. Credit is due for acknowledging some limitations in the text, including the possibility that the scattered-neurons intuition may be wrong, and for engaging with the fact that IIT escapes the thought experiment by appealing to causal structure. However, the paper is an argumentative essay rather than a demonstration: the decisive simulation is not described, the notion of "degenerated computation" is not defined, and the thought experiment's force depends on an intuition that the author himself flags as uncertain. The manuscript would benefit from clearly separating what is demonstrated from what is proposed as a research program.

major comments (3)

[Is consciousness a computation?] The central argument against computational functionalism rests on the claim that "it can be shown with a computational simulation that one can generate circumstances where computation is degenerated, while neural processes work exactly the same as before" (Gidon et al., 2025). This is load-bearing, yet the simulation is not described and the key term "degenerated computation" is never defined. To refute functionalism, the degenerated computation would have to be the computation the system genuinely implements under the functionalist's own implementation criterion; without such a criterion, the example may only show that computational descriptions are context-relative, not that consciousness depends on the substrate. Please either summarize the simulation and specify the implementation relation, or explicitly present the scenario as a challenge rather than a demonstration.
[Is consciousness simply a very complex pattern of spiking?] The scattered-neurons thought experiment assumes that if consciousness were "solely a matter of these spiking patterns," scattering the neurons while replaying the same spike pattern would leave consciousness unchanged. Most NCC theories, however, treat spiking patterns as correlates rather than identities, and the author notes that IIT escapes because it appeals to causal structure. The manuscript does not explain why the pattern reading is the correct one to test, nor whether the scattering preserves the causal interactions that many theories consider constitutive. As it stands, the argument relies on the intuition that scattered neurons would not be conscious, an intuition the author explicitly concedes may be false. The section needs a stated criterion for when a pattern is the same "neural process" before the thought experiment can support the conclusion.
[Matter matters for the mind] The inference from generative entrenchment and the RNA-to-protein analogy to "consciousness is not an abstract computation" is not justified. Those examples show that some biological outcomes require specific physical processes and that evolutionary history constrains possible realizations, but they do not show that consciousness itself lacks a multiply realizable functional description. The claim that "there is no reason to think that [the physical part] should be nonexistent" is an argument from ignorance unless a concrete physical process is identified as constitutive of conscious states rather than a contingent implementation. Please either supply such a candidate process or clearly mark the conclusion as a research hypothesis.

minor comments (5)

[Beyond perception: Sudden shifts in consciousness] The text cites Klatzmann and colleagues (2024), but the reference list contains Klatzmann et al. (2022) on bioRxiv; please reconcile the year and, if the published version appeared in 2024, update the reference accordingly.
[Is consciousness simply a very complex pattern of spiking?] The name "Shall" in the citation "Logothetis & Shall, 1989" is a typo for "Schall."
[Matter matters for the mind] The in-text citation "Rosa Cao (2002)" appears with a quoted passage, but the reference list gives Cao (2022); please correct the year.
[References] The reference list includes both Seth (2024, PsyArXiv) and Seth (2025, Behavioral and Brain Sciences) for what appears to be the same target article; please unify and cite the definitive venue.
[What type of theory of consciousness do we still need?] The phrase "Seth, 2024" in the discussion of entrenchment is inconsistent with the "Seth (2025)" citation used elsewhere for biological naturalism; please harmonize the citation.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular reduction found: the anti-computationalist argument is a perspective piece whose self-citations are load-bearing but not definitional; score 1.

full rationale

This is an opinion/perspective essay, not a derivation with equations, fitted parameters, or definitions that presuppose their conclusions. The central claim that consciousness is not an abstract computation is supported by a cluster of arguments: the Gidon et al. (2022) scattered-neurons thought experiment, the Gidon et al. (2025) degeneration simulation, generative entrenchment (Wimsatt, Cao, Seth), the RNA-to-protein physical-process analogy, and the observation that current theories abstract away biological detail. Although Gidon et al. (2022) and Gidon et al. (2025) are co-authored by the present author, they are presented as independently checkable thought experiments and computational simulations, not as definitional stipulations; the paper does not define 'consciousness' or 'computation' so as to make the conclusion true by construction. The paper explicitly disclaims derivational status ('I offer no answers'), admits empirical ignorance ('we have no clue how big the physical part is'), and labels the aha-mechanism account as 'just speculation.' These self-acknowledged limitations are honesty about support, not circularity. The weakest point is that the Gidon 2025 simulation's bearing on functionalism depends on an implementation criterion that is not specified here, but that is a rigor/underdetermination concern, not a circular reduction. No step in the paper is equivalent to its own input by definition, so the score is low.

Assumptions & free parameters 0 free parameters · 3 assumptions · 1 invented entities

No free parameters are fit because the paper presents no quantitative model. The axioms are conceptual commitments imported from prior work or asserted as phenomenological intuitions. The paper introduces no new empirically testable entities; the 'physical part of computation' is a broad explanatory concept rather than a defined mechanism.

assumptions (3)

domain assumption Consciousness is a unified and localized experience, so a scattered implementation could not have it.
Used to reject the thought experiment outcome in the section on spiking patterns; this phenomenological premise is asserted rather than argued.
domain assumption The Gidon et al. (2025) computational simulation actually demonstrates degeneration of computation with identical neural activity.
The paper cites this simulation as the basis for concluding that consciousness is not an abstract computation, but does not reproduce or describe the simulation's mechanism.
domain assumption Generative entrenchment applies to the brain in a way that makes biological matter causally relevant to consciousness.
Invoked via Cao (2022) and Seth (2025) to support the claim that physical substrate constrains possible consciousness; this is a theoretical commitment, not an empirical result.

invented entities (1)

The physical part of computation
purpose: Explains why abstract computational descriptions cannot capture consciousness: some of the computation is performed by physical matter itself, not by an algorithm.
The paper posits this as an explanatory posit to make biological naturalism more concrete, but offers no falsifiable handle or independent evidence for where or how this physical computation occurs.

how reviews work

0 comments

Cite this review

Pith. "Pith review of The assumptions that restrain us from understanding consciousness." pith.science (2026). https://pith.science/paper/NBVEYFRN

@misc{pith2026250621485,
  author       = {Pith},
  title        = {Pith review of: The assumptions that restrain us from understanding consciousness},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NBVEYFRN}},
  note         = {Machine review of arXiv:2506.21485}
}

read the original abstract

The science of consciousness has been successful over the last decades. Yet, it seems that some of the key questions remain unanswered. Perhaps, as a science of consciousness, we cannot move forward using the same theoretical commitments that brought us here. It might be necessary to revise some assumptions we have made along the way. In this piece, I offer no answers, but I will question some of these fundamental assumptions. We will try to take a fresh look at the classical question about the neural and explanatory correlates of consciousness. A key assumption is that neural correlates are to be found at the level of spiking responses. However, perhaps we should not simply take it for granted that this assumption holds true. Another common assumption is that we are close to understanding the computations underlying consciousness. I will try to show that computations related to consciousness might be far more complex than our current theories envision. There is little reason to think that consciousness is an abstract computation, as traditionally believed. Furthermore, I will try to demonstrate that consciousness research could benefit from investigating internal changes of consciousness, such as aha-moments. Finally, I will ask which theories the science of consciousness really needs.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

3 extracted references · 2 canonical work pages

[1]

Adrian, E.D. (1928). The basis of sensation . WW Norton & Co Anderson, P. W. (1972). More Is Different: Broken symmetry and the nature of the hierarchical structure of science. Science , 177 (4047), 393-396. Aru, J., Bachmann, T., Singer, W., & Melloni, L. (2012). Distilling the neural correlates of consciousness. Neuroscience & Biobehavioral Reviews , 36...

work page 1928
[42]

K., Laukkonen, R., & Aru, J

Tulver, K., Kaup, K. K., Laukkonen, R., & Aru, J. (2023). Restructuring insight: An integrative review of insight in problem-solving, meditation, psychotherapy, delusions and psychedelics. Consciousness and cognition , 110 , 103494. Tulver, K., Kaup, K. K., & Aru, J. (2025). The road to Aha: a recipe for mental breakthroughs. Cognition Varela, F., Lachaux...

work page 2023
[506]

& Melloni, L

Cogitate Consortium, Ferrante, O., Gorska-Klimowska, U., Henin, S., Hirschhorn, R., Khalaf, A., ... & Melloni, L. (2025). Adversarial testing of global neuronal workspace and integrated information theories of consciousness. Nature , 1-10. Crick, F., & Koch, C. (1990). Towards a neurobiological theory of consciousness. In Seminars in the Neurosciences (Vo...

arXiv 2025

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Adrian, E.D. (1928). The basis of sensation . WW Norton & Co Anderson, P. W. (1972). More Is Different: Broken symmetry and the nature of the hierarchical structure of science. Science , 177 (4047), 393-396. Aru, J., Bachmann, T., Singer, W., & Melloni, L. (2012). Distilling the neural correlates of consciousness. Neuroscience & Biobehavioral Reviews , 36...

work page 1928

[2] [42]

K., Laukkonen, R., & Aru, J

Tulver, K., Kaup, K. K., Laukkonen, R., & Aru, J. (2023). Restructuring insight: An integrative review of insight in problem-solving, meditation, psychotherapy, delusions and psychedelics. Consciousness and cognition , 110 , 103494. Tulver, K., Kaup, K. K., & Aru, J. (2025). The road to Aha: a recipe for mental breakthroughs. Cognition Varela, F., Lachaux...

work page 2023

[3] [506]

& Melloni, L

Cogitate Consortium, Ferrante, O., Gorska-Klimowska, U., Henin, S., Hirschhorn, R., Khalaf, A., ... & Melloni, L. (2025). Adversarial testing of global neuronal workspace and integrated information theories of consciousness. Nature , 1-10. Crick, F., & Koch, C. (1990). Towards a neurobiological theory of consciousness. In Seminars in the Neurosciences (Vo...

arXiv 2025