Philosophy Companion Paper¶

Jean-Paul Niko · February 2026

\begin{center} {\LARGE\bfseries\color{sectionblue} The Conceptual Irreversibility Theorem:
Why Translation Between Experience [3pt] and Description is Necessarily Lossy}

{\large Jean-Paul Niko} [4pt] {} [2pt] {\smallniko@triptomean.com}

{\small February 2026} \end{center}

We prove that no translation from experiential content to propositional description can be lossless. The result---the Conceptual Irreversibility Theorem (CIT)---is formulated in the language of topos theory: each cognitive system induces a conceptual topos \(\mathcal{T}\) whose internal logic is a Heyting algebra (intuitionistic, not Boolean). The morphism from experience to description passes through a subobject classifier \(\Omega\) that is strictly richer than \(\{0,1\}\); the round-trip composition \(\text{describe} \circ \text{experience}\) is a non-invertible functor. The Heyting Gap---the measure of information destroyed---grows monotonically with cognitive sophistication (Gap Monotonicity Theorem), formalizing the intuition that the more one knows, the harder it is to say. We derive corollaries for the hard problem of consciousness, the verbalization gap in expertise research, the structural impossibility of perfect translation between languages, and the limits of AI alignment via natural language specification. The result is structural, not merely epistemic: the loss is inherent in the logical architecture of conceptual systems, not a consequence of insufficient vocabulary or processing power.

Keywords: conceptual irreversibility, topos theory, Heyting algebra, hard problem of consciousness, verbalization gap, translation loss

%% ═══════════════════════════════════════════════════════════════════════════════

\color{sectionblue¶

Introduction} %% ═══════════════════════════════════════════════════════════════════════════════

The hard problem of consciousness [chalmers1996] asks why physical processes give rise to subjective experience at all. Jackson's Knowledge Argument [jackson1982] dramatizes the gap: Mary, the color scientist who has never seen red, learns something new upon first seeing it---something her complete physical knowledge did not contain. Nagel [nagel1974] argues that the subjective character of experience makes it impossible to know "what it is like" to be a bat from any amount of external description.

These arguments share a structural claim: there exists information in experience that cannot be captured by propositional description. But the claim remains informal. What exactly is the logical structure of the gap? Can it be measured? Does it depend on the system's complexity?

This paper formalizes the gap using topos theory---a branch of category theory that generalizes set-theoretic logic to contexts where the law of excluded middle fails. We prove that the description morphism is non-invertible (the CIT), measure the lost information (the Heyting Gap), and show that the gap grows with cognitive sophistication (Gap Monotonicity). The result is structural: it depends on the logical architecture of conceptual systems, not on computational limitations.

CS (instantiation operator) as a Modeling Construct¶

A note on ontology is essential. When we refer to "the CS operator" or "experiential content," we use these terms as modeling constructs, not metaphysical commitments. Just as phase space is the space of possible mechanical states without implying that abstract 6N-dimensional points are physically real, the CS operator is the space of possible cognitive states. The synonym "cognitive state space" may be used interchangeably. The mathematics stands regardless of one's position on the ontology of consciousness.

%% ═══════════════════════════════════════════════════════════════════════════════

\color{sectionblue¶

Conceptual Topoi} %% ═══════════════════════════════════════════════════════════════════════════════

\tA{} A conceptual topos \(\mathcal{T}\) is a category satisfying the topos axioms (finite limits, exponentials, subobject classifier), whose: [nosep] - objects are concepts (mental representations, ideas, percepts), - morphisms are conceptual transformations (inference, analogy, generalization), - subobject classifier \(\Omega\) encodes the truth-value structure of the system's internal logic.

The central claim: biological cognitive systems---and indeed any system with graded, context-dependent truth evaluation---induce Heyting topoi, not Boolean ones.

\tB{} Any cognitive system whose truth evaluation depends on context, perspective, or available evidence operates with an internal logic that is at most intuitionistic (Heyting), not classical (Boolean). Equivalently, such systems satisfy: there exist propositions \(p\) such that \(p \vee \neg p\) does not hold in the system's internal logic.

\tA{} Consider the Necker cube: the proposition "the cube faces left" is neither true nor false in the visual system's processing---it oscillates. This is not indecision; it is a structural feature of a non-Boolean logic. The subobject classifier \(\Omega\) contains the truth value "bistable," which has no counterpart in \(\{0,1\}\).

%% ═══════════════════════════════════════════════════════════════════════════════

\color{sectionblue¶

The Description Morphism} %% ═══════════════════════════════════════════════════════════════════════════════

\tA{} The description morphism is a functor \(D: \mathcal{T}_{\mathrm{exp}} \to \mathcal{T}_{\mathrm{prop}}\) from the experiential topos (the system's full conceptual space) to the propositional topos (the subsystem accessible to linguistic description). The experience morphism is a functor \(E: \mathcal{T}_{\mathrm{prop}} \to \mathcal{T}_{\mathrm{exp}}\) going the other direction (interpreting descriptions as experiences).

The question is: does the round-trip \(E \circ D\) recover the original experience?

%% ═══════════════════════════════════════════════════════════════════════════════

\color{sectionblue¶

The Conceptual Irreversibility Theorem} %% ═══════════════════════════════════════════════════════════════════════════════

\tA{} Let \(\mathcal{T}_{\mathrm{exp}}\) be a Heyting topos and \(\mathcal{T}_{\mathrm{prop}}\) be a Boolean topos. Then no functor \(D: \mathcal{T}_{\mathrm{exp}} \to \mathcal{T}_{\mathrm{prop}}\) admits a left inverse: there is no functor \(E\) such that \(E \circ D = \mathrm{Id}\). Equivalently, the round-trip \(D \circ E \circ D \neq D\) in general.

[Proof sketch] The subobject classifier of \(\mathcal{T}_{\mathrm{exp}}\) is a Heyting algebra \(\Omega_H\) with \(|\Omega_H| > 2\). The subobject classifier of \(\mathcal{T}_{\mathrm{prop}}\) is \(\Omega_B = \{0,1\}\). Any functor \(D\) must map \(\Omega_H \to \Omega_B\), which is a surjection from a richer to a poorer truth-value structure. By the pigeonhole principle, \(D\) identifies distinct truth values in \(\Omega_H\) that map to the same value in \(\Omega_B\). Any functor \(E: \Omega_B \to \Omega_H\) cannot recover which pre-image was intended. Therefore \(E \circ D \neq \mathrm{Id}_{\mathcal{T}_{\mathrm{exp}}}\).

\tB{} The hard problem of consciousness is not merely "difficult"---it has a precise mathematical structure. The experiential content that resists propositional capture is exactly the content whose truth values lie in \(\Omega_H \setminus \{0,1\}\): the intermediate, graded, context-dependent truth values that have no Boolean counterpart.

%% ═══════════════════════════════════════════════════════════════════════════════

\color{sectionblue¶

The Heyting Gap} %% ═══════════════════════════════════════════════════════════════════════════════

\tA{} The Heyting Gap of a cognitive system with experiential topos \(\mathcal{T}_{\mathrm{exp}}\) and propositional topos \(\mathcal{T}_{\mathrm{prop}}\) is:

For continuous Heyting algebras, replace cardinality with the topological dimension of \(\Omega\).

The Heyting Gap measures the bits of experiential information destroyed by the description morphism.

\tA{} Let \(\mathcal{T}_1 \hookrightarrow \mathcal{T}_2\) be an inclusion of conceptual topoi (i.e., \(\mathcal{T}_2\) has strictly more concepts and richer internal logic). Then \(\mathcal{G}(\mathcal{T}_2) \geq \mathcal{G}(\mathcal{T}_1)\).

\tB{} Expert knowledge resists verbalization in proportion to its experiential richness. Hinds' "curse of knowledge" [hinds1999] and Dreyfus & Dreyfus' observation that expert skill is largely non-propositional [dreyfus1986] are consequences of Gap Monotonicity.

%% ═══════════════════════════════════════════════════════════════════════════════

\color{sectionblue¶

Corollaries and Applications} %% ═══════════════════════════════════════════════════════════════════════════════

The Hard Problem Dissolved¶

The three-space ontology (Part XIII) goes beyond reformulating the hard problem---it dissolves it. The hard problem asks: "Why is there something it is like to be a physical system?" Under the co-primordial thesis, the answer is: consciousness (\(\CSp\)) is not something that needs to "arise" from physical systems. It has existed co-primordially with \(\QS\) and \(\PS\) since moment zero. The question is not "how does matter produce consciousness?" but "how does consciousness complexify through its interaction with matter?" The hard problem was hard because it was formulated within a single-space materialist framework that lacked the ontological resources to state the answer.

The structural parallel with quantum gravity is exact and illuminating. The unrenormalisable infinities of perturbative quantum gravity arise from trying to describe a three-space phenomenon (gravity as proto-consciousness) entirely within \(\QS\)---the mathematics returns NaN because the domain is wrong. The hard problem of consciousness arises from trying to derive a three-space phenomenon (consciousness) entirely from \(\PS\)---the explanation returns "explanatory gap" because the ontological domain is wrong. In both cases, the apparent intractability is not a sign that the problem is genuinely hard; it is a diagnostic that the single-space framework lacks the structure to pose the question well. Fix the domain---include all three spaces---and the infinities resolve and the gap closes, not through clever new arguments but because the type error has been corrected.

Qualia are private instantiations: the specific way a particular consciousness projects a particular \(\QS\)-configuration into its own \(\PS\). The privacy of qualia follows from the irreducible individuality of the instantiation operator \(\Inst_\alpha\); no two consciousnesses project identically.

Free will is lateral temporal navigation: the freedom of consciousness to move in the \(t_{\mathrm{lat}}\) direction of complex time \(t_\CSp = t_\mathbb{R} + i\,t_{\mathrm{lat}}\). This is neither compatibilist (determined by prior causes) nor libertarian (random); it is structured navigation of a complex temporal manifold, constrained by the Id filter but not eliminated by it.

The Hard Problem Reformulated¶

The CIT does not solve the hard problem, but it gives it a precise shape. The explanatory gap between physical processes and subjective experience corresponds to the Heyting Gap between the experiential topos and the propositional (scientific) topos. Closing the gap would require either (a) enriching scientific language to a Heyting logic (abandoning excluded middle in physics), or (b) showing that biological cognition is actually Boolean (contradicting the evidence of perceptual ambiguity, graded beliefs, and context-dependent truth).

Mary's Room¶

Jackson's Mary [jackson1982] knows all physical facts about color but has never seen red. In our formalism: Mary's propositional topos contains all Boolean truths about color, but the experiential topos of seeing red has Heyting truth values (phenomenal gradations, context-dependent saturation, affective valence) that \(\{0,1\}\) cannot represent. When Mary sees red, her experiential topos expands; the new content was provably not in her propositional topos.

What Is It Like to Be a Bat?¶

Nagel's question [nagel1974] concerns the overlap between human and bat experiential topoi. The overlap subspace \(\mathcal{T}_{\mathrm{overlap}} = \mathcal{T}_{\mathrm{human}} \cap \mathcal{T}_{\mathrm{bat}}\) has smaller \(\Omega\) than either system's full topos. The CIT predicts that the human description of bat experience is lossy by exactly \(\mathcal{G}(\mathcal{T}_{\mathrm{bat}}, \mathcal{T}_{\mathrm{overlap}})\) bits---the part of bat experience that has no counterpart in human conceptual space.

Translation Between Languages¶

Each natural language induces a conceptual topos \(\mathcal{T}_L\) via its lexical and grammatical resources. Translation from language \(L_1\) to \(L_2\) passes through the overlap \(\mathcal{T}_{L_1} \cap \mathcal{T}_{L_2}\). The CIT predicts that perfect translation is impossible whenever the source topos has concepts outside the overlap---precisely when "untranslatable" words exist. This is Jakobson's dictum formalized: "languages differ in what they must convey" because their topoi have different subobject classifiers [jakobson1959].

AI Alignment¶

The CIT has implications for AI alignment. If human values are encoded in a Heyting experiential topos but alignment specifications use Boolean propositional logic, then no specification can fully capture the intended values. The Heyting Gap measures the irreducible alignment loss---the extent to which "what we mean" exceeds "what we can say." This suggests that alignment via natural language instruction alone has a provable ceiling; complementary approaches (demonstration, feedback, constitutional methods) are structurally necessary.

%% ═══════════════════════════════════════════════════════════════════════════════

\color{sectionblue¶

Engagement with Existing Frameworks} %% ═══════════════════════════════════════════════════════════════════════════════

Integrated Information Theory (IIT)¶

Tononi's IIT [tononi2004] measures consciousness via integrated information \(\Phi\). In our framework, \(\Phi\) is related to the spectral structure of the compatibility matrix \(\bK\) governing cross-type interactions. The conceptual topos adds a layer IIT lacks: the logical structure of integration, not just its quantity. Two systems with equal \(\Phi\) may have different Heyting Gaps if their internal logics differ.

Global Workspace Theory¶

Baars' GWT [baars1988] identifies conscious access with global broadcast. The CIT is compatible: the broadcast content is the propositionally accessible portion of the experiential topos. What GWT calls "unconscious processing" corresponds to experiential content whose truth values lie in \(\Omega_H \setminus \Omega_B\)---present in the system but not available for propositional report.

G\"ardenfors' Conceptual Spaces¶

G\"ardenfors [gardenfors2000] models concepts as convex regions in quality spaces. Our conceptual topos generalizes this: G\"ardenfors' spaces are the geometric realization of a particular class of topoi (presheaf topoi over quality dimensions). The CIT applies to any topos, not only geometric ones.

Quine's Indeterminacy¶

Quine's indeterminacy of translation [quine1960] is epistemic: we cannot determine the correct translation scheme from behavioral evidence alone. The CIT is structural: even with complete knowledge of both systems, the translation morphism is non-invertible. The two results are complementary, not competing: Quine says we cannot know the mapping; the CIT says no perfect mapping exists.

%% ═══════════════════════════════════════════════════════════════════════════════

\color{sectionblue¶

Discussion} %% ═══════════════════════════════════════════════════════════════════════════════

Objections and Responses¶

Objection 1: Why should cognition be Heyting? The evidence is behavioral (perceptual ambiguity, graded beliefs, context-dependent truth evaluation) and computational (neural networks naturally implement graded activation, not Boolean gates). If a convincing argument emerged that biological cognition is Boolean, the CIT would not apply---but the burden of proof lies with the Boolean hypothesis.

Objection 2: Is this just Gödel incompleteness in disguise? No. Gödel's theorems concern provability within formal systems; the CIT concerns translatability between logical systems with different truth-value structures. A Heyting topos is not "incomplete"---it is complete within its own logic. The loss occurs at the boundary between two logics.

Objection 3: Topos theory is too abstract for empirical science. The abstraction is the point: the CIT applies to any cognitive system with Heyting internal logic, regardless of substrate. Concreteness comes from the applications (Mary's Room, translation, alignment), each of which yields testable or at least evaluable predictions.

Limitations¶

The proof assumes that propositional systems are Boolean. If propositional language can itself be enriched to Heyting logic (e.g., through vagueness or fuzzy predicates), the gap shrinks but does not vanish as long as the experiential Heyting algebra is strictly richer. The cardinality-based Gap measure (Definition ref:def:gap) is coarse; a finer measure using Heyting algebra homomorphism theory would capture structural, not just quantitative, differences.

%% ═══════════════════════════════════════════════════════════════════════════════

\color{sectionblue¶

Conclusion} %% ═══════════════════════════════════════════════════════════════════════════════

The Conceptual Irreversibility Theorem establishes that translation from experience to description is necessarily lossy---a logical, not merely practical, limitation. The Heyting Gap measures the information destroyed, and Gap Monotonicity shows it grows with cognitive sophistication. The result gives the hard problem of consciousness a precise mathematical shape, formalizes the expert verbalization gap, predicts the structural impossibility of perfect translation, and identifies a provable ceiling on alignment via natural language specification. Whether one regards this as a contribution to philosophy, mathematics, or cognitive science depends on one's disciplinary home; the theorem itself is indifferent to the classification.

%% ═══════════════════════════════════════════════════════════════════════════════

References¶

See PDF for full bibliography.¶

v2 Integration: Axiom 0, GNEP Person Definition, Moral Framework (TMP-20260217)¶

Axiom 0 — Relational Primacy: Only relational reality admits absolutes and infinities. Physical space P has bounded quantities; relational space ℛ does not. Mathematical objects including infinite sets exist in ℛ — this resolves the Platonism problem. Cantor's diagonal argument is a theorem about relational structure, not physical objects.

Person as GNEP Hypervisor: A person P is a GNEP hypervisor node: cognitive assembly with |A(P)| ≥ 1 agents under coordination and a self-assembly property. Personhood is functional, not substrate-bound.

Moral Framework — Id_extended:

This is a cooperative Nash equilibrium: no agent can increase total life force by unilateral deviation. The moral action is not agent-relative — life force is one undifferentiated quantity maximized globally.