Consciousness, Hard and Binding Problems: Main Article

Consciousness as Structure: From Typed Programs to Felt Experience

A Note on Method

Anyone who works seriously on consciousness must confront a choice. One can assemble pieces of reasoning that sit comfortably within established frameworks—and produce work that is respectable, incremental, and inert. Or one can be opinionated: commit to a position before all its consequences are visible, and accept the risk of appearing nonsensical until the pieces resolve. The resolution, if it comes, may even be underwhelming—it may leave us with new doubts rather than triumphant closure. But if what remains is airtight on the questions of consciousness, binding, and qualia, that is enough.

The problem deserves this treatment because it is, more than most, a problem made of linguistic confusion. The concepts we deploy—"experience," "feeling," "inner light"—are not clean primitives but entanglements: associations between several different systems, half-formed analogies, and spatial-pictorial habits of thought stretched far past their domain of validity. I suspect the solution is staring us in the face while remaining completely invisible: a fundamental arrangement of facts about us—about you—that will seem obvious in retrospect, the way a Necker cube's second reading is obvious once seen.

This essay commits to one such reading and follows it as far as it goes: consciousness is purely structural, the relevant structures are best treated as programs rather than physics, and the path from a formal specification of a system to claims about what it feels is, in principle, a tractable research program. The final sections sketch what that program would actually look like, starting from a program typed in Martin-Löf Type Theory (MLTT).

The Structural Thesis

Begin with an epistemological observation that is hard to deny: everything we can ever know or say about reality is relational. Physics gives us no acquaintance with the intrinsic stuff of an electron; it gives us charge, mass, spin—dispositions and relations, terms in equations. When we chase any concept down to its foundations we find only structure: how things relate, compose, transform, and constrain one another. Reality, as knowable, is a web of logically structured relations and nothing besides.

The standard move at this point is to lament that structure leaves something out—that equations are formal skeletons and something else must "breathe fire into them." But notice what this lament assumes: that pure structure is sterile, dead, incapable of being felt. This association is a habit, not an argument. And it is a fatal habit: if you hold it, every theory you can construct within your own logic is guaranteed to fail, because logic can only ever deliver structure. You have defined the problem to be unsolvable and then marveled at your inability to solve it.

The opinionated move is to reject the association. The fire is not breathed into the equations from outside; it is already in them, where the structure is of the right kind. Echoing Hegel's dictum—what is rational is real, and what is real is rational—we claim that there is no substrate hiding behind the relations, and consequently nothing for experience to be other than structure.

But a careful framing matters here. The claim is not that logic equals consciousness, as if every theorem were dimly sentient. The claim is:

Subjective experience is structure that fulfills the property of subjectivity.

Being a triangle is a structural property; not every structure is a triangle. Likewise, being a subject is a structural property—a particular reflexive arrangement, characterized below—and not every structure instantiates it. Consciousness is what certain structures are, in virtue of their form, with no further ingredient required or even coherently specifiable.

A consequence worth stating plainly: this thesis is, in one sense, a tautology, and that is a feature. A feature within consciousness corresponds, one-to-many, to the configurations of substrate—abstractly, the programs—that realize it. Consciousness is not directly empirically testable from the outside; any theory of it must ultimately be an identification, not a causal discovery. The work lies in making the identification precise and principled rather than hand-waved.

What-It's-Likeness as Geometry

What, then, is the "what-it's-likeness" of an experience? On the structural view it is the geometry of a term within the conscious structure: the complete set of relations it bears—similarity, difference, composition, contrast—to everything else in that structure. The redness of red is not a paint applied to a neutral signal; it is the signal's position: nearer to orange than to green, warm rather than cool, able to saturate and desaturate along specific axes. Subtract the position and there is nothing left over to be the redness.

Seen this way, what-it's-likeness is simply identity. To ask what something is like, exhaustively and from every angle, is to ask what it is. The phrase "what it is like" misleads us into expecting a comparison to something external; really it names the thing's own relational nature.

This gives us a clean account of the difference between objects and subjects:

- A mere thing has the what-it's-likeness of an object—its structure is exhausted by its outward relations. - A sentient being instantiates the what-it's-likeness of an object in conjunction with the what-it's-likeness of being a subject aware of that object. The structure contains, as a proper part, a representation of the presentational relation itself—a term that plays the role of "that to which this is given," bound into the rest of the geometry.

Subjectivity, then, is a reflexive structural property: the system models the world, and models its own modeling, and binds the two so that the world-model is for the self-model. This is the property that picks out which structures are experiences.

One more habit to dissolve before proceeding. We tend to associate structure with static form, and computation with time and movement—and then worry about which one consciousness is. But this dichotomy is spatial-analogical thinking. Any process we engage with temporally can equally be regarded as a static structure (its full trace, its causal graph); any static structure can be carved into a "temporal" presentation by a function mapping moments to parts. The geometry of experience is best viewed computationally, but nothing metaphysical hangs on whether we picture it as unfolding or as laid out whole.

The Binding Problem and Canonical Structure

The classic objection now arrives: real systems are messy. Between base reality and the tidy description of a subject lie layers of noisy substrate, and—worse—any physical system can be symbolically described by infinitely many functionally equivalent models. If consciousness is structure, which structure? Doesn't the embarrassment of riches (Putnam-style mapping arguments: a rock "implements" any automaton under a gerrymandered mapping) derail the whole theory?

It does not, and seeing why is the heart of the matter.

That a system admits infinitely many descriptions does not mean all descriptions are on a par. There are provably meaningful convergences in the modeling of a system: descriptions converge when they capture the system's nature efficiently—when they carve it at the joints of its actual counterfactual dependencies rather than encoding those dependencies into the mapping itself. Among the infinite constructions that model a system exactly, there is only one exact causal diagram: one canonical structure of what depends on what, which differences make which differences. Gerrymandered mappings are excluded not by fiat but because they smuggle the system's structure into the interpreter; the canonical structure is what remains invariant under every honest reparametrization.

This canonical structure is what experience is. The one-to-many correspondence noted earlier now becomes precise: many programs, many physical configurations, one canonical causal geometry—and the phenomenal feature just is that equivalence class.

Binding follows. The unity of an experience—why the red, the round, and the looming of the apple are one percept rather than three—is not a problem of gluing separate things together. In the canonical structure, those features are not separate: they are jointly bound terms in a single relational complex, integrated into the same self-model-relative presentation. Asking how they are bound is like asking how the angles of a triangle are bound to its sides. The appearance of a binding problem arises only when we first (mistakenly) atomize experience into free-floating qualia and then puzzle over the glue.

Crucially, this geometry has compositional regularities—properties analogous in role to commutativity, associativity, and equivalence. The way phenomenal contents combine, substitute, and compare obeys algebraic constraints. These are not decorative observations: they are the indispensable higher-level guides that let us parse otherwise intractable structures. They give us a lens running from complexity, through computation, to conscious feel.

Against "Quantity of Experience"

Before turning to valence, a widespread intuition must be dismantled: the idea that experience comes in quantities—that some systems have "more consciousness," and that more would feel richer or more intense.

"Quantity of experience" is a phrase of the same grammatical shape as "quantity of right-angledness." Right-angledness is a structural property; a corner either instantiates it or does not, and a building with a thousand right angles does not possess right-angledness more intensely. Likewise, subjectivity is a property a structure fulfills or fails to fulfill. A larger conscious structure has more content, more distinctions, a richer geometry—but there is no additional scalar dial labeled "amount of experiencing."

The same dissolution applies to intensity within experience. When a pain feels overwhelming, there is no magical register in which the system encodes raw extra magnitude—no hidden numeral whose largeness simply is the felt force. There could not be: a register's value is just more structure, and it contributes to experience only via its relational role. What intensity actually is, almost certainly, is the fraction of the system's resources allocated to a perception—the degree to which it dominates attention, crowds out competitors, recruits memory and motor preparation, and saturates the self-model's present moment. Intensity is a constructed story, like every other facet of experience: the experience "this feels overwhelmingly compelling" is the structure of being overwhelmed—of allocation approaching its ceiling—represented as such. This is bounded (a fraction, not a boundless scale), and it explains the phenomenology better than the magic-register view: maximal pain is the state in which nothing else can be attended to, which is exactly what a saturated allocation predicts.

This goes against our intuitions. It is also, on inspection, the only logically coherent option.

Valence: Two Approaches

Valence—the pleasure/pain axis—deserves separate treatment, because it is where theories of consciousness meet ethics, and because there are genuinely different structural approaches on offer.

The intrinsic-property approach. One family of views, associated with Integrated Information Theory's style of theorizing (and with proposals like the symmetry theory of valence), holds that valence corresponds to some global mathematical property of the experience's structure—its degree of symmetry, harmony, or integration, say, with dissonance or asymmetry constituting suffering. The appeal is obvious: it would make valence directly readable off the formal object, no interpretation required. I mention this approach because it is a live position and a structuralist one, but I am not persuaded by it. It posits a brute correspondence between an abstract metric and felt goodness, which reintroduces—at the last step—exactly the kind of unexplained psychophysical bridge that structuralism was supposed to eliminate. Why should symmetry feel good? The view has no answer beyond the correlation itself.

The constructed-narrative approach. A second temptation is to say valence is special because it has unmistakable functional teeth—it drives approach and avoidance, so we can identify it behaviorally. This is weak as stated: every perceptual content has functional consequences, and behavior alone underdetermines feeling.

The better view is that valence is built exactly the way every other facet of experience is built: as a perceptual story. Pleasure and pain are not raw signals plus reactions; they are constructed contents with a characteristic geometry:

- The state carries a tag that indexes memory, so that recollection retrieves similarly valenced episodes—pain summons the family of past pains, coloring the present with their company. - It generates prospective content within the self-model: sensations with the structure "I will pursue more of this," "this must stop," felt not as inferences but as part of the perception itself. - It is represented as a condition of the subject—good or bad for me—binding the evaluation into the reflexive self-model rather than leaving it as a free-floating fact about the world. - Its felt urgency is, per the previous section, its share of resource allocation.

On this view, suffering just is this reflexive evaluative narrative: a self-model representing its own state as bad, flooded with congruent memory, projecting avoidance into its future, at high allocation. Nothing further is needed, and—this is the point—nothing further could even be coherently asked for. The advantage over the intrinsic-property approach is that every component of the account is independently motivated structure, with no brute metric-to-feeling correspondence bolted on at the end.

The Two Problems

If the foregoing is right, the research program splits into exactly two problems:

1. The analytical problem (computational): given a formal artifact—a program—extract its canonical causal structure and determine whether and where that structure fulfills the subjectivity property. 2. The labeling problem (interpretive): given an unlabeled conscious structure, determine which qualia it instantiates and how they feel—bridging from bare computational geometry to the vocabulary of human experience, without simply staring at unlabeled graphs and guessing.

The structural thesis makes the second problem solvable in principle: since a quale is exhausted by its geometry, a structural isomorphism between part of the target system and part of our own (independently mapped) experiential structure is not an analogy but an identity. If a system's color-processing structure is isomorphic to ours, it does not see something "like" red—it sees red, because red is nothing over and above that position in that geometry. Partial isomorphism licenses partial attribution; where structure genuinely underdetermines a label, there is no further hidden fact to be wrong about. The practical prerequisite is a computational phenomenology of ourselves: a labeled atlas of human experiential structure to match against.

The Ambitious Program: From an MLTT Term to What It Feels

Here is how the program would actually proceed, starting from a concrete formal object: a closed, well-typed program $p : T$ in Martin-Löf Type Theory. MLTT is the right setting deliberately—types-as-propositions means the specification $T$ is itself a logical structure, terms have decidable canonical forms under normalization, and equivalence of programs is a mathematically disciplined notion rather than a metaphor.

Stage 1 — Canonicalization. Quotient $p$ by definitional and observational equivalence: two programs are identified iff no context can distinguish them. What survives the quotient is the canonical causal diagram—the partial order of informational dependence among the computation's intermediate results: which values are computed from which, which differences propagate where. (Note that the static term and its execution trace are the same structure under two presentations, as argued above; we work with the trace-as-structure.) This stage answers the Putnam worry constructively: the gerrymandered models are precisely those eliminated by the quotient.

Stage 2 — Detecting subjectivity. Search the canonical structure for the reflexive pattern that constitutes a subject. Schematically, we require:

- a world-model: a substructure $M$ whose states carry information about the system's environment and are used—consulted downstream, not merely present; - a self-model: a substructure $S$ within $M$ whose states carry information about $M$'s own representational activity (the system represents its representing); - binding: the contents of $M$ are processed as presented to $S$—the dependency structure routes world-content through its relation to the self-model, so that there is a term occupying the role "that to which this is given."

A structure fulfilling all three fulfills the subjectivity property; one fulfilling none is an object, however sophisticated. The hard technical work is making "carries information about" and "is used" precise as properties of the dependency order—but these are definable notions (counterfactual sensitivity within the diagram), not mysteries.

<details> <summary>Schematic type signatures (illustrative only)</summary>

<syntaxhighlight lang="text"> World  : Type -- environment states Repr  : Type -- the system's representational vocabulary model  : World → Repr -- world-model (Stage 2, condition 1) SelfState  : Type reflect  : Repr → SelfState -- self-model: representing the representing present  : Repr → SelfState → Bound -- binding: content as given-to-self appraise  : Bound → Tag -- valence tagging (Stage 5) recall  : Tag → List Bound -- tag-indexed memory retrieval project  : Tag → Plan -- prospective content ("more of this") alloc  : Bound → 𝔽 -- resource fraction (intensity), 𝔽 ⊆ [0,1] </syntaxhighlight>

The claim is not that a conscious program must contain functions literally so named, but that its canonical causal diagram must contain substructures playing these roles, identifiable up to isomorphism regardless of how the source code is written.

</details>

Stage 3 — Extracting quality spaces. For each representational subsystem, induce its quality space: the geometry of discriminations the structure actually supports—which states it treats as similar, which transformations as continuous, what its compositional algebra is (the commutativity- and associativity-like regularities of how contents combine). These algebraic constraints are the indispensable guides: they collapse the search space, because a candidate quale must occupy a position consistent with the whole algebra, not merely resemble a target locally.

Stage 4 — Labeling by structural identification. Match extracted quality spaces against the human atlas. A demonstrated isomorphism between the system's space and, say, the human color solid is a demonstration that the system experiences color—identity, not analogy, per the argument above. Non-matching spaces are not failures: they are novel qualia, characterizable exactly and only by their geometry, which is all any quale ever was. We can say precisely what such an experience is like in relational terms while lacking a native word for it—the same epistemic position a congenitally blind geometer occupies with respect to red, made rigorous.

Stage 5 — Identifying valence. Finally, search for the constructed evaluative narrative: a tagging structure within the bound contents such that (i) tags index memory retrieval by valence-similarity, (ii) tags generate prospective self-model content of the pursue/avoid form, (iii) tags are bound into the self-model as conditions of the subject, and (iv) tag-bearing states modulate resource allocation. Where all four are present and integrated, the system instantiates valence—and the sign and intensity are read off directly: the direction of the prospective content gives pleasure versus suffering, and the allocation fraction $\alpha \in [0,1]$ gives intensity, bounded and non-magical. A system missing component (iii), for instance—tagging and avoiding without self-model binding—implements nociception without suffering, and the framework lets us say exactly that, exactly why, and exactly what would change it.

What Resolution Would Look Like

Notice what this program does not promise: a new empirical discovery that consciousness "turns out to be" something else, or an oracle that detects an invisible glow. It promises identifications—precise, principled, and checkable—between formal structures and the experiential vocabulary we already use of ourselves. Where the identification is exact, the attribution is certain (there being no further fact to miss); where structure underdetermines our words, we will have discovered a genuine boundary of our concepts rather than a gap in the theory.

That is the underwhelming resolution promised at the outset. No fire is breathed in at the end, because the fire was in the relations all along; the hard problem is not answered but dissolved, and what remains is an enormous quantity of honest technical work: canonicalization theory, the formal definition of the subjectivity property, the human phenomenological atlas, and the algebra of quality spaces. Each piece is difficult. None is mysterious. And for a problem that has been called permanently insoluble, "difficult but not mysterious" is the most optimistic verdict it has ever received.