Pith. sign in

REVIEW 5 major objections 4 minor 10 references

Artificial Intelligence as an Opportunity for the Science of Consciousness: A Dual-Resolution Framework

T0 review · 5 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper argues that consciousness arises when an informationally autonomous system continuously updates its own history, making it "the epistemic resolution of life."

desk verdict A clearly written theoretical synthesis of ITI and MtM that reframes AI as a testbed for consciousness research, but the central claim is stipulated and the proposed experiment's marker-to-consciousness mapping is unvalidated, making the test weaker than the paper claims. read the letter →

arxiv 2509.07001 v3 pith:AWGD4DFI submitted 2025-09-05 q-bio.NC

Shahar Dror , Dafna Bergerbest , Moti Salti This is my paper
classification q-bio.NC
keywords consciousnessartificialintelligencedual-resolutionframeworkInformationTheoryofIndividualityMoment-to-Momenthysteresisinformationalautonomybiologicalnaturalism
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 paper argues that the deadlock between biological naturalism and computational functionalism can be broken by treating consciousness as a layered informational phenomenon. An entity must first qualify as an informationally autonomous individual (the ontological condition), and only then can moment-to-moment updating with hysteresis turn that individuated existence into subjective experience (the epistemic condition). Consciousness is "the epistemic resolution of life": life is the unit that persists, and consciousness is that unit experiencing itself through continuous updating. The payoff is a substrate-independent theory with concrete tests: if the two conditions are engineered into an AI system, characteristics of consciousness such as self/other body boundaries should spontaneously emerge. That makes AI not just a challenge to consciousness science but an experimental probe for it.

What carries the argument

The load-bearing mechanism is hysteresis—the persistence of a system's own history in its present updating. In the Moment-to-Moment component, each new stimulus is integrated against the backdrop of prior updates; this creates an internal trace unique to the system, so the new input is recognized as its own, and a boundary forms between self and environment. Hysteresis is what turns an informationally autonomous unit (the ITI component) into a subject with a perspective. The dual-resolution pairing matters as much as the mechanism itself: ITI supplies the ontological condition that a candidate entity exists as an individual, MtM supplies the epistemic condition that this individual appears t

What would settle it

Run the proposed experiment: give modified reinforcement-learning agents informational autonomy (persistent self-governed memory, intrinsic state variables to maintain) and enriched hysteresis (context-, goal-, and saliency-weighted updating), then place them in an environment with manipulable objects, hazards, and self-repair. If, despite satisfying both criteria, the agents show no difference in response to self-caused versus externally imposed perturbations—or if agents lacking both criteria show the same self/other distinction—the framework's claim that these conditions are sufficient for

Watch

Extended reading notes

Core claim

The paper's central claim is that consciousness is not a property of brains, computations, or information integration per se, but of a dual relationship between individuation and updating. The Information Theory of Individuality fixes what counts as an individual: a system that propagates information from its past to its future while maintaining temporal integrity, thereby securing informational autonomy against its environment. The Moment-to-Moment theory fixes what it is like to be such an individual: a continuous process in which incoming stimuli are updated and recoded against a history-laden, context-sensitive, saliency-weighted stream, producing hysteresis and a self/other boundary. Be

Load-bearing premise

Everything depends on treating an agent's spontaneous ability to distinguish self-caused from externally imposed perturbations as evidence that consciousness-like characteristics have emerged, yet the paper does not independently show that this behavioral marker tracks consciousness rather than adaptive self-modeling.

Editorial extensions

If this is right

  • Consciousness ceases to be a human- or brain-specific category: any substrate that realizes informational autonomy plus history-dependent updating is a candidate.
  • Biological naturalism and computational functionalism become partial truths: life in the informational sense is required, but biological biochemistry is not.
  • Current AI systems fail the framework for different reasons: LLMs on individuation, RL agents on the richness of hysteresis; targeted architectural changes could move them toward the criteria.
  • The proposed RL testbed gives an empirical way out of circularity: emergence or absence of body-boundary behaviors can raise or lower confidence in the framework.
  • Consciousness science can use AI as a constructive laboratory, updating theory from experiments rather than declaring verdicts about existing systems.

Reading between the lines

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

  • The paper leaves open how to distinguish genuine phenomenality from a system that merely models itself; its body-boundary test may conflate self-modeling with consciousness. A stronger test would require the agent to use its self/non-self distinction to reorganize its own goals, not just optimize externally given rewards.
  • If the framework is right, "what it is like to be a system" becomes a comparative, quantitative question: two systems with similar temporal and structural constraints should have communicable experiences, which could be probed by aligning their internal state dynamics.
  • A natural extension is to vary the temporal resolution of updating across agents, from slow to fast hysteresis, and ask whether the resulting self-boundary behaviors shift, turning the sloth-versus-falcon intuition into a testable manipulation.
  • The framework could be read as implying a continuum rather than a threshold: degrees of informational autonomy and hysteresis would yield degrees of consciousness, which the paper does not fully embrace.
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.

Referee Report

5 major / 4 minor

Summary. This paper proposes a 'dual-resolution framework' for the science of consciousness, combining the Information Theory of Individuality (ITI) with the Moment-to-Moment (MtM) theory. ITI is said to specify the ontological conditions of individuation and self-maintenance; MtM specifies the epistemic conditions of temporal updating and hysteresis. The central claim is that 'consciousness is the epistemic resolution of life': consciousness arises when an individuated, informationally autonomous system exhibits moment-to-moment updating with hysteresis. The paper applies this framework to current AI systems, arguing that LLMs lack individuation while RL agents are at least minimally individuated but lack rich hysteresis. It then proposes a modified RL-agent experiment in which agents with both individuation and enriched hysteresis are tested for spontaneous acquisition of 'operational analogues of body boundaries' (self vs. externally caused perturbations), claiming this would provide empirical support or refutation of the framework's sufficiency claim. The paper is a theoretical/perspective piece with no new experimental data.

Significance. If the framework could be made precise and its proposed test deployed, it would offer a genuinely substrate-neutral account of consciousness that promises to reconcile biological naturalism and computational functionalism, and it would position AI as a constructive experimental platform rather than merely a target of attribution. The paper is clearly written, engages seriously with recent literature (Seth 2024; Butlin et al. 2023), and explicitly acknowledges the risk of circularity—an unusual and honest feature. The proposed RL paradigm is an interesting attempt to derive falsifiable predictions. However, in its current form the central claim is stipulated rather than derived, the key empirical markers are not independently validated, and the proposed experiment is only sketched. The paper would be strengthened by formal operationalization and an explicit account of how markers track consciousness rather than merely self-modeling or adaptive control.

major comments (5)
  1. [AI as a testbed (pp. 17–18)] The entire empirical test rests on equating 'operational analogues of body boundaries' with 'known characteristics of consciousness.' This equation is stipulated, not derived or independently supported. The text cites Revach and Salti (2022) for the claim that body boundaries emerge from the same updating mechanism, but no such reference appears in the reference list (only Revach & Salti, 2021, is listed). Without independent evidence that the proposed behavioral markers track consciousness—rather than self-modeling, predictive control, or adaptive homeostasis—the positive result would be a false positive and the negative result would not actually test the sufficiency claim. If the markers are too liberal, self-maintaining non-conscious systems could pass; if too strict, a conscious system might fail. The paper needs to justify why these particular behaviors are the right diagnostic.
  2. ['Revisiting the challenge' / 'AI as a testbed' (pp. 17–18)] The logic of falsification is internally inconsistent as stated. The paper claims that if individuated systems with enriched hysteresis do not spontaneously acquire body boundaries, this 'provides evidence against the sufficiency of our criteria for consciousness.' This inference is valid only if the emergence of body boundaries is a necessary consequence of the proposed sufficient conditions. But the paper nowhere argues for this necessity; body boundaries are just presented as one characteristic that 'emerges in the same manner' (p. 13). The framework's own definitions do not entail that all conscious systems must exhibit exactly these operational markers, nor that the markers are uniquely diagnostic. The experiment therefore does not disambiguate the sufficiency claim; it only confirms the stipulated mapping.
  3. ['AI as a testbed' (pp. 17–18)] The proposed experimental paradigm is too underspecified to be evaluated or executed: there are no agent architectures, no environment details, no operational thresholds for 'spontaneously acquire,' no baseline or control conditions, and no specification of how one distinguishes self-caused from externally imposed perturbations in practice. As written, the paradigm cannot yield a determinate falsification. For example, any moderately good predictive-coding agent might treat self-caused sensory consequences as less surprising—so the test may not be specific to the proposed dual-resolution conditions. Concrete parameters and controls are needed before the claimed 'empirical consequences' are meaningful.
  4. [The updating approach to consciousness (pp. 13–14)] The central thesis—'consciousness is understood as the epistemic resolution of life'—is introduced as a definitional stipulation, not as a hypothesis with independent justification. The paper acknowledges this circularity in 'Revisiting the challenge' but responds only by asserting that the proposed experiment avoids it. This is not sufficient: the experiment can test whether the stipulated criteria produce particular behaviors, but unless those behaviors are independently linked to consciousness, the test remains circular. The authors need to clarify the epistemological status of their core claim: is it a definition (in which case no experiment can confirm it), or an empirical hypothesis (in which case it needs to be stated in a form that could, in principle, be disconfirmed by evidence that does not already presuppose the definition)?
  5. [Revisiting the challenge (pp. 15–16)] The classification of LLMs and RL agents relies on qualitative, non-quantified characterizations of 'informational autonomy' and 'hysteresis.' For example, LLMs are said to lack genuine individuation because they are 'statistical transformers' with 'no ontological continuity of existence,' while RL agents are 'at least minimally individuated.' But ITI is a formal information-theoretic framework; the paper does not provide any measure or formal criterion to support these judgments. Without a precise operationalization of the ITI and MtM conditions, the framework cannot be applied consistently to actual systems, and the proposed engineering modifications (e.g., 'self-governed objectives,' 'saliency weighting') are not tied to formal thresholds. This weakens both the theoretical claim and the experimental design.
minor comments (4)
  1. [References / p. 13 and p. 17] The text cites 'Revach and Salti (2022)' but the reference list contains only Revach and Salti (2021). Either the correct reference should be added or the citation corrected.
  2. [Passim] Typos and minor language issues: 'casual structure' (p. 12) should be 'causal structure'; 'Aikire' in the Mashour & Aikire reference should be 'Alkire'; 'interweaved' (p. 7) should be 'interwoven'; 'salti et al.'s' should be 'Salti et al.'s'.
  3. [p. 4] The reference to 'Krakauer et al. (2020)' is correctly cited, but the paper refers to it as 'Krakauer and colleagues' in the same paragraph; style inconsistency.
  4. [p. 12] The phrase 'specifies a casual structure very similar to that of a natural brain' is unclear because 'casual' is likely a typo; if intended, the claim is ambiguous. Please verify and clarify.

Circularity Check

3 steps flagged · score 6.0 of 10

The proposed falsification test is circular: 'body boundaries' are already claimed by the framework to emerge from the same MtM hysteresis that defines consciousness, so observing them cannot independently confirm the sufficiency claim.

  1. self definitional [Revisiting the challenge (after 'The updating approach to consciousness')]
    "Trivially, our dual-resolution framework allows us to assess whether current AI systems meet the criteria for consciousness. The first criterion, drawn from ITI, is whether an entity is individuated from an environment in a manner that secures informational autonomy and self-maintenance. The second criterion, extracted from the MtM, is whether the system exhibits hysteresis that supports epistemic resolution, namely a temporally extended process of updating that constitutes subjective experience."

    Consciousness is defined earlier as 'the epistemic resolution of life', i.e., ITI individuation plus MtM updating/hysteresis. The assessment of LLMs and RL agents then applies exactly those two criteria: LLMs are said to lack genuine individuation and therefore 'no epistemic resolution'; RL agents are said to have insufficient hysteresis. These verdicts follow from the definition by construction rather than from independent evidence.

  2. self definitional [AI as a testbed (after 'Revisiting the challenge')]
    "The central question is whether, under these conditions, the agents spontaneously acquire operational analogues of body boundaries, that is, the capacity to distinguish between self-generated and externally imposed perturbations and to allocate resources preferentially to the protection of their own integrity. If individuated systems with enriched hysteresis display novel behaviors, such as spontaneous emergence of body boundaries, then the framework gains explanatory traction."

    The predicted marker, body-boundaries, is not an independent indicator of consciousness in this framework. Earlier the paper says that body-boundaries and volition 'emerge' from the same MtM updating mechanism. Thus the experiment's positive result is entailed by the very criteria being tested; observing body-boundaries merely re-observes the definitional link. A negative result would speak against sufficiency only if one already accepts the stipulated equation between body-boundaries and consciousness, which is not independently validated.

1 more flagged steps
  1. self citation load bearing [The updating approach to consciousness (MtM paragraph)]
    "Importantly, other characteristics of conscious experience emerge, such as body-boundaries and volition in the same manner (See Revach and Salti (2022) for details)."

    The load-bearing bridge that lets the testbed treat spontaneous body-boundaries as evidence for or against the sufficiency claim is justified by a citation to Revach and Salti (2022), whose author (Salti) overlaps with the present paper. No independent, machine-checked, or externally falsifiable support is provided for the claim that body-boundaries emerge from hysteresis and are a characteristic of consciousness. Notably, the reference list contains Revach & Salti (2021) but not Revach & Salti (2022), so the cited support is not even locatable in the manuscript.

full rationale

The paper explicitly acknowledges the circularity risk: 'At the same time, one might object that this line of reasoning risks circularity. If we begin by defining consciousness in terms of individuation and hysteresis, then it follows that any system engineered to satisfy those constraints will, by definition, count as conscious.' It attempts to escape by proposing an experimental testbed in which AI agents are engineered to have individuation and enriched hysteresis and then checked for 'known characteristics of consciousness'. That escape fails because the chosen characteristic, body-boundaries, is itself derived from the same MtM hysteresis mechanism via a self-citation (Revach & Salti 2022, which is also missing from the reference list). Therefore the experiment's predicted outcome is internal to the definition, not an independent empirical consequence. The general philosophical synthesis of ITI and MtM is a coherent proposal rather than an empirical derivation, but the paper's central testbed claim—that the framework 'yields empirical consequences' and 'avoids circularity'—does not hold. This is partial circularity: the central test reduces to its own definitions, but the paper is transparent about the definitional starting point and does not hide the structure.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The framework rests on several stipulated domain assumptions, chief among them the validity of ITI's informational individuality as the ontological ground and MtM's hysteresis as the epistemic ground. The central claim that consciousness is the epistemic resolution of life is an ad hoc definition introduced by the authors. No empirical evidence is provided for these assumptions, and the proposed experiment would test only a subset of them.

assumptions (6)
  • domain assumption ITI's definition of individuality (informational autonomy) is a valid basis for the ontological condition of consciousness.
    The paper adopts Krakauer et al.'s (2020) framework without independent justification for its sufficiency for consciousness; it is used as the foundation for the first criterion.
  • domain assumption MtM's hysteresis/updating process constitutes subjective experience.
    The epistemic criterion relies on the authors' own MtM theory (Salti et al., 2019) and self-cited extensions (Revach & Salti, 2021, 2022), with no independent validation cited.
  • ad hoc to paper Consciousness is the epistemic resolution of life (stipulated).
    This central definitional assertion appears in the section 'The ITI-MtM dual resolution perspective'; it is not derived from the two component theories but is presented as their synthesis.
  • domain assumption Intelligence and consciousness are interweaved.
    Argued from Descartes and Dennett in the section 'Redefining the relations between consciousness, intelligence and life'; it is a philosophical stance, not an empirically established fact.
  • domain assumption Behavioral markers (body boundaries) can indicate consciousness characteristics in artificial agents.
    In 'AI as a testbed', the authors assume that distinguishing self-generated from external perturbations maps onto a functional boundary relevant to consciousness, without independent evidence.
  • domain assumption Biological embodiment is not necessary for consciousness.
    The entire framework assumes substrate independence, drawing on ITI's broadening of life to non-biological systems; this is stated as a premise, not tested.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Artificial Intelligence as an Opportunity for the Science of Consciousness: A Dual-Resolution Framework." pith.science (2026). https://pith.science/paper/AWGD4DFI

@misc{pith2026250907001,
  author       = {Pith},
  title        = {Pith review of: Artificial Intelligence as an Opportunity for the Science of Consciousness: A Dual-Resolution Framework},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AWGD4DFI}},
  note         = {Machine review of arXiv:2509.07001}
}
read the original abstract

The encounter of artificial intelligence with consciousness research is often framed as a challenge: could this science determine whether such systems are conscious? We suggest it is equally an opportunity to expand and test the scope of existing theories of consciousness. Current approaches remain polarized. Computational functionalism emphasizes abstract organization, often realized through neural correlates of consciousness, while biological naturalism insists that consciousness is tied to living embodiment. Both positions risk anthropocentrism and limit the possibility of recognizing non-biological forms of subjectivity. To move beyond this impasse, we propose a dual-resolution framework that defines the ontological and epistemic conditions for consciousness. This approach combines the Information Theory of Individuality, which defines the ontological conditions of informational autonomy and self-maintenance, with the Moment-to-Moment theory, which specifies the epistemic conditions of temporal updating and phenomenological unfolding. This integration reframes consciousness as the epistemic expression of individuated systems in substrate-independent informational terms, offering a generalizable theory of consciousness and positioning AI as a promising testbed for its emergence.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

10 extracted references · 8 canonical work pages

  1. [1]

    2School of Behavioral Sciences, The Academic College of Tel-Aviv Yaffo; Tel-Aviv Yaffo, 61083, Israel

    1 Artificial Intelligence as an Opportunity for the Science of Consciousness: A Dual-Resolution Framework Shahar Dror1, Dafna Bergerbest2, Moti Salti1 1Department of Cognitive and Brain Sciences, Ben-Gurion University of the Negev; Beer Sheva, 84105, Israel. 2School of Behavioral Sciences, The Academic College of Tel-Aviv Yaffo; Tel-Aviv Yaffo, 61083, Isr...

    work page 2023

  2. [2]

    Revach, D., & Salti, M. (2021). Expanding the discussion: Revision of the fundamental assumptions framing the study of the neural correlates of consciousness. Consciousness and Cognition, 96(October), 103229. https://doi.org/10.1016/j.concog.2021.103229 Salti, M., Harel, A., & Marti, S. (2019). Conscious perception: Time for an update? Journal of Cognitiv...

    work page arXiv 2021

  3. [3]

    Cogito, ergo sum

    fundamental assertion, “Cogito, ergo sum”, positions thinking as the very foundation of being conscious (Husserl, 2012; Sartre, 1957). Thinking is also a core concept in the study of intelligence, from Jean Piaget’s (Piaget,

    work page 2012

  4. [35]

    https://doi.org/doi.org/10.1098/rstb.2014.0167 Turing, A. M. (1950). Computing Machinery and Intelligence. Brain Physiology & Psychology, 212–240. Varela, F. J., Thompson, E., Rosch, E., & Kabat-Zinn, J. (1993). The embodied mind: Cognitive science and human experience [Book]. In E. Thompson & E. Rosch (Eds.), The Embodied Mind: Cognitive Science and Huma...

    work page arXiv 2014

  5. [450]

    https://doi.org/10.4159/harvard.9780674594623.c15 Piaget, J. (1950). Perception et Intelligence [Article]. Bulletin du Groupe d’études de psychologie de l’Université de Paris, 4(1), 25–34. https://doi.org/10.3406/bupsy.1950.5685 Putnam, H. (2011). Minds and machines [Bookitem]. Philosophical Papers, Volume 2,

    work page arXiv 1950

  6. [1950]

    rule books

    focus on cognitive development to contemporary theories that emphasize thinking as the key to intelligence. Dennett, in his collection of essays, Brainstorms (1978), treats thinking as the fundamental mechanism that any candidate intelligence must embody. Accordingly, an intelligent system must “think” in the sense of representing, manipulating, and updat...

    work page 1978

  7. [1974]

    What is it like to be an AI system?

    fundamental question: "What is it like to be an AI system?", a perspective essential for deeper insights into AI consciousness. This challenge is not unique to Butlin et al. (2023) but reflects a broader issue in consciousness science. The field has long been constrained by its anthropocentric focus, primarily examining human consciousness while avoiding ...

    work page 2023

  8. [1993]

    consciousness

    argue for continuity between life and mind, they remain committed to biological organization as the locus of subjectivity. This dual-resolution perspective allows for a reconciliation between biological naturalism and 15 computational functionalism. Biological naturalism is satisfied as consciousness is grounded in the individuated, self-maintaining dynam...

    arXiv 1990

Show all 10 references
  1. [2013]

    talk" seemingly positions them as excellent candidates to test for generalization. Unlike animals, LLMs can articulate their own

    and the research on consciousness in non-human animals generally falls into two main categories. The first category uses non-human primates, as a model for studying consciousness under the assumption that these animals possess consciousness. The second category seeks to determ...

    2009

  2. [2023]

    http://arxiv.org/abs/2303.07103 Crick, F., & Koch, C. (1990). Towards a neurobiological theory of consciousness. Seminars in the Neurosciences, 2, 263–275. http://papers.klab.caltech.edu/22/1/148.pdf Dehaene, S., & Changeux, J. P. (2011). Experimental and Theoretical Approache...

    1990 arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.