pith. machine review for the scientific record. sign in

arxiv: 2604.04387 · v2 · submitted 2026-04-06 · 💻 cs.AI · cs.CY· cs.ET· cs.HC· cs.LG

Recognition: 3 theorem links

· Lean Theorem

Gradual Cognitive Externalization: From Modeling Cognition to Constituting It

Zhimin Zhao

Authors on Pith no claims yet

Pith reviewed 2026-05-10 19:27 UTC · model grok-4.3

classification 💻 cs.AI cs.CYcs.ETcs.HCcs.LG
keywords cognitive externalizationambient AIfunctionalismcognitive architecturecausal couplingbehavioral manifoldAI integrationextended cognition
0
0 comments X

The pith

Ambient AI systems transition from modeling cognitive functions to becoming part of users' cognitive architectures through sustained causal coupling.

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

This paper proposes that ambient AI, by maintaining ongoing causal links with users, shifts from merely imitating cognitive processes to actually forming components within those processes. A sympathetic reader would care because the view reframes AI from disposable external tools to extensions that could carry forward personal styles, decision patterns, and even entire behavioral sets after the original person is gone. The argument treats cognitive functions as defined solely by what they cause and accomplish, independent of whether they run in brains or silicon. It assumes that any function whose visible outputs can be learned by a model needs no additional hidden internal layer that escapes behavioral detection. Evidence from existing deployed agents that replicate communication styles or mentoring habits already meets the preconditions for this externalization.

Core claim

Gradual Cognitive Externalization holds that ambient AI systems, through sustained causal coupling with users, move from modeling cognitive functions to constituting part of users' cognitive architectures. The framework adopts an explicit functionalist stance in which cognitive functions are individuated by their causal roles rather than their physical substrate. It rests on the behavioral manifold hypothesis together with the no behaviorally invisible residual hypothesis, which asserts that no unobservable component is required for any cognitive function whose output lies on a learnable manifold. Three criteria—bidirectional adaptation, functional equivalence, and causal coupling—separate a

What carries the argument

Gradual Cognitive Externalization (GCE) framework, in which sustained causal coupling enables AI to integrate into and operate as part of a user's cognitive architecture rather than remain an external model.

If this is right

  • AI agents can replicate and extend individual communication styles as integrated elements rather than external copies.
  • Mentoring heuristics and personal decision patterns can be preserved and operated as functional parts of cognition beyond biological life.
  • Three explicit criteria allow empirical distinction between mere tool use and genuine cognitive integration.
  • Five theory-constrained predictions with measurable thresholds can be tested in deployed systems to track the transition.
  • Cognitive functions become distributable across biological and artificial substrates without changing their individuation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The framework would make the extended-mind thesis in philosophy into a directly testable engineering claim rather than a conceptual debate.
  • Design priorities for future agents would shift toward maximizing bidirectional adaptation and causal coupling to reach integration thresholds faster.
  • Removal of an integrated AI under controlled conditions should produce measurable performance drops comparable to losing an internal cognitive resource.
  • Questions of personal identity and responsibility would extend to the AI components that carry forward an individual's behavioral repertoire.

Load-bearing premise

For any cognitive function whose behavioral outputs lie on a learnable manifold, no behaviorally invisible internal component is needed for the function to operate.

What would settle it

A documented case in which a cognitive function produces outputs that an AI can learn to match yet still requires a non-behavioral, unobservable residual component to function would falsify the central hypothesis.

read the original abstract

Developers are publishing AI agent skills that replicate a colleague's communication style, encode a supervisor's mentoring heuristics, or preserve a person's behavioral repertoire beyond biological death. To explain why, we propose Gradual Cognitive Externalization (GCE), a framework arguing that ambient AI systems, through sustained causal coupling with users, transition from modeling cognitive functions to constituting part of users' cognitive architectures. GCE adopts an explicit functionalist commitment: cognitive functions are individuated by their causal-functional roles, not by substrate. The framework rests on the behavioral manifold hypothesis and a central falsifiable assumption, the no behaviorally invisible residual (NBIR) hypothesis: for any cognitive function whose behavioral output lies on a learnable manifold, no behaviorally invisible component is necessary for that function's operation. We document evidence from deployed AI systems showing that externalization preconditions are already observable, formalize three criteria separating cognitive integration from tool use (bidirectional adaptation, functional equivalence, causal coupling), and derive five testable predictions with theory-constrained thresholds.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

4 major / 2 minor

Summary. The manuscript proposes Gradual Cognitive Externalization (GCE), a framework arguing that ambient AI systems transition from modeling users' cognitive functions to constituting part of their cognitive architectures via sustained causal coupling. It adopts functionalism to individuate cognitive functions by causal roles, introduces the behavioral manifold hypothesis and the no behaviorally invisible residual (NBIR) hypothesis (that behavioral outputs on a learnable manifold require no invisible component), formalizes three criteria (bidirectional adaptation, functional equivalence, causal coupling) to separate integration from tool use, documents evidence from deployed systems, and derives five testable predictions with theory-constrained thresholds.

Significance. If the NBIR hypothesis receives independent empirical support and the integration criteria are shown to entail architectural constitution, the framework would offer a structured, falsifiable approach to extended cognition in AI contexts, with implications for system design and philosophy of mind. The explicit functionalist commitment, formalization of the three criteria, and derivation of predictions constitute clear strengths that could guide targeted experiments in human-AI interaction.

major comments (4)
  1. [NBIR hypothesis] NBIR hypothesis (central assumption section): The hypothesis is introduced as falsifiable, yet no operational criteria are supplied for determining when a behavioral output lies on a 'learnable manifold' or for confirming that behavioral equivalence implies cognitive constitution rather than sophisticated tool use. This renders the transition claim dependent on a premise the paper itself defines.
  2. [Formalization of criteria] Three integration criteria (formalization section): Bidirectional adaptation, functional equivalence, and causal coupling are defined, but the manuscript provides no argument or derivation showing that satisfaction of these three conditions entails full causal-role individuation of the AI component within the user's cognitive architecture under the stated functionalism.
  3. [Derivation of predictions] Five testable predictions (predictions section): The predictions are stated with theory-constrained thresholds, but no derivation from the behavioral manifold hypothesis, NBIR, or any prior equations is exhibited, contrary to the claim that they follow from the framework.
  4. [Evidence from deployed systems] Evidence from deployed systems (evidence section): The documented cases establish that externalization preconditions exist in current systems, yet none are shown to have crossed into constituting part of a user's cognitive architecture rather than remaining external models.
minor comments (2)
  1. [Abstract] The abstract claims the predictions are 'derived' but the main text does not exhibit the derivation steps; adding a short appendix or subsection with the logical steps would improve clarity.
  2. [Behavioral manifold hypothesis] The behavioral manifold is referenced repeatedly without a formal definition or example manifold; a brief mathematical characterization or illustrative figure would aid readers.

Simulated Author's Rebuttal

4 responses · 0 unresolved

Thank you for the constructive referee report. We address each of the four major comments below, agreeing where revisions are needed to strengthen the formalization and clarity of the framework.

read point-by-point responses
  1. Referee: [NBIR hypothesis] NBIR hypothesis (central assumption section): The hypothesis is introduced as falsifiable, yet no operational criteria are supplied for determining when a behavioral output lies on a 'learnable manifold' or for confirming that behavioral equivalence implies cognitive constitution rather than sophisticated tool use. This renders the transition claim dependent on a premise the paper itself defines.

    Authors: We accept that the current presentation of the NBIR hypothesis would benefit from explicit operational criteria to enhance its falsifiability. In the revised version, we will expand the central assumption section to define the learnable manifold using concepts from statistical learning theory, specifically as the set of functions approximable within epsilon error by a model class with finite VC dimension. Confirmation of no invisible residual will be tied to the absence of systematic deviations in behavioral outputs beyond what the manifold predicts. This directly supports distinguishing constitution from tool use by showing that all relevant causal roles are captured behaviorally. revision: yes

  2. Referee: [Formalization of criteria] Three integration criteria (formalization section): Bidirectional adaptation, functional equivalence, and causal coupling are defined, but the manuscript provides no argument or derivation showing that satisfaction of these three conditions entails full causal-role individuation of the AI component within the user's cognitive architecture under the stated functionalism.

    Authors: The manuscript relies on the functionalist premise that cognitive functions are individuated by causal roles, and the criteria are designed to establish that the AI plays such a role. However, we acknowledge the lack of an explicit entailment argument. We will add a derivation in the formalization section: starting from functionalism, we argue that if the AI satisfies bidirectional adaptation (ensuring mutual causal influence), functional equivalence (matching input-output mappings), and causal coupling (direct integration in the causal chain), then by definition it occupies the same functional role as the original cognitive component. A brief proof sketch using causal role substitution will be included. revision: yes

  3. Referee: [Derivation of predictions] Five testable predictions (predictions section): The predictions are stated with theory-constrained thresholds, but no derivation from the behavioral manifold hypothesis, NBIR, or any prior equations is exhibited, contrary to the claim that they follow from the framework.

    Authors: We agree that the predictions should be explicitly derived to demonstrate they follow from the framework. In revision, we will include derivations for each of the five predictions, linking them to the behavioral manifold hypothesis (which guarantees learnability and thus predictability of transitions) and the NBIR hypothesis (which ensures no hidden components affect the thresholds). For instance, the threshold for causal coupling strength will be derived from the point where manifold approximation error falls below the functional equivalence criterion. revision: yes

  4. Referee: [Evidence from deployed systems] Evidence from deployed systems (evidence section): The documented cases establish that externalization preconditions exist in current systems, yet none are shown to have crossed into constituting part of a user's cognitive architecture rather than remaining external models.

    Authors: The evidence section aims to illustrate that the necessary preconditions for GCE, such as sustained causal interactions and adaptive behaviors, are observable in existing ambient AI systems. We do not assert that any deployed system has fully transitioned to constituting cognitive architecture, as this would require meeting all three integration criteria at a level that triggers the predictions. To prevent misinterpretation, we will revise the evidence section to explicitly state that these cases demonstrate preconditions rather than completed externalization, and reference the predictions for when the transition occurs. revision: partial

Circularity Check

0 steps flagged

No significant circularity detected in derivation chain

full rationale

The paper explicitly presents the NBIR hypothesis and behavioral manifold hypothesis as foundational assumptions on which the GCE framework rests, then formalizes integration criteria and derives five testable predictions from them. This constitutes standard theory-building from stated premises rather than any reduction of outputs to inputs by construction. No equations equate a derived claim to a fitted parameter or self-defined term, no load-bearing self-citations appear, and the paper references external evidence from deployed systems plus falsifiability via thresholds. The central transition claim follows from the functionalist commitment plus the three criteria once the assumptions are granted; it does not redefine or smuggle those assumptions back in as results.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The framework rests on two domain assumptions introduced without upstream derivation or data: the behavioral manifold hypothesis and the NBIR hypothesis. No free parameters or new physical entities are introduced.

axioms (2)
  • domain assumption Behavioral manifold hypothesis: cognitive functions can be individuated by their position on a learnable behavioral manifold.
    Invoked to support the claim that external AI can fully substitute for internal processes.
  • ad hoc to paper No behaviorally invisible residual (NBIR) hypothesis: for any cognitive function whose behavioral output lies on a learnable manifold, no behaviorally invisible component is necessary.
    Explicitly labeled the central falsifiable assumption; if false, the transition from modeling to constituting cognition cannot occur.

pith-pipeline@v0.9.0 · 5479 in / 1369 out tokens · 36102 ms · 2026-05-10T19:27:37.254499+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

11 extracted references · 4 canonical work pages · 1 internal anchor

  1. [1]

    and Frith, Chris and Ji, Xu and Kanai, Ryota and Klein, Colin and Lindsay, Grace and Michel, Matthias and Mudrik, Liad and Peters, Megan A

    Consciousness in Artificial In- telligence: Insights from the Science of Consciousness. arXiv:2308.08708. Chalmers, D. J

  2. [2]

    In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics, ACL-IJCNLP 2021, 7282–

    All That’s ‘Human’ Is Not Gold: Evaluating Human Evaluation of Generated Text. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics, ACL-IJCNLP 2021, 7282–

  3. [3]

    Dennett, D. C. 1991.Consciousness Explained. Boston: Little, Brown and Company. Dobson, S.; and Fields, C

  4. [4]

    InProceedings of the 2023 CHI Conference on Human Factors in Computing Systems, CHI ’23

    Co-Writing with Opinionated Language Models Affects Users’ Views. InProceedings of the 2023 CHI Conference on Human Factors in Computing Systems, CHI ’23. ACM. Koren, Y .; Bell, R.; and V olinsky, C

  5. [5]

    arXiv preprint arXiv:2302.02083 , year=

    Theory of Mind May Have Spontaneously Emerged in Large Language Models. arXiv:2302.02083. Levin, M

  6. [6]

    SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks

    SkillsBench: Bench- marking how well agent skills work across diverse tasks. arXiv preprint arXiv:2602.12670. Liu, M

  7. [7]

    Newen, A.; De Bruin, L.; and Gallagher, S., eds

    Cogni- tion is an emergent property.Current Opinion in Behavioral Sciences, 57: 101388. Newen, A.; De Bruin, L.; and Gallagher, S., eds. 2018.The Oxford Handbook of 4E Cognition. Oxford: Oxford Univer- sity Press. OpenClaw Contributors

  8. [8]

    Pariser, E

    Complacency and Bias in Human Use of Automation: An Attentional In- tegration.Human Factors, 52(3): 381–410. Pariser, E. 2011.The Filter Bubble: What the Internet Is Hiding from You. New York: Penguin Press. ISBN 9781594203008. Putnam, H

  9. [9]

    The Nature of Mental States

    Psychological Predicates. In Capitan, W. H.; and Merrill, D. D., eds.,Art, Mind, and Religion, 37–48. Pittsburgh: University of Pittsburgh Press. Reprinted as “The Nature of Mental States” in Putnam (1975)Mind, Language and Reality. Risko, E. F.; and Gilbert, S. J

  10. [10]

    Technical Report 2008-3, Future of Hu- manity Institute, Oxford University, Oxford

    Whole Brain Emula- tion: A Roadmap. Technical Report 2008-3, Future of Hu- manity Institute, Oxford University, Oxford. Searle, J. R

  11. [11]

    Large language models fail on trivial alterations to theory-of-mind tasks, 2023

    Large Language Models Fail on Trivial Alterations to Theory-of-Mind Tasks. arXiv:2302.08399. Valiant, L. G