Quality Spaces and the Algebra of Discrimination

The Gap

The consciousness article describes a five-stage program for determining what a system experiences, starting from a well-typed term in Martin-Löf Type Theory. Stage 1 canonicalizes the program (the Canonical Causal Diagram article). Stage 2 detects subjectivity—the reflexive pattern of world-model, self-model, and binding (the Integrated Substructure article provides the formalization of what counts as a genuine substructure to search). Stage 4 labels qualia by matching against a human phenomenological atlas. Stage 5 identifies valence.

Stage 3 is missing. The consciousness article states:

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.

This is the analytical core of the consciousness research program. Without it, the labeling problem (Stage 4) is ill-posed—we do not know what to match against the human atlas. Without it, the measure metric (which computes ε, α, and M over representational subsystems) operates without a precisely defined domain. Without it, cross-architecture comparison—the identification of a system's experiential geometry with ours—is analogy rather than identity.

This article formalizes quality spaces. It defines what they are, shows how to extract them from the canonical causal diagram, identifies the algebraic constraints that characterize them, and argues that cross-architecture comparison is structural identity under univalence.

What a Quality Space Is

A quality space is the discrimination structure of a representational subsystem, together with the law by which its representational contents compose. It has three components.

The states. Consider a representational subsystem S of a conscious system—a genuine integrated substructure (as defined in the Integrated Substructure article) that plays a representational role: its states carry information about some domain and are used downstream. S has a set of internal states, generated by its processing. Two states that are observationally equivalent—no downstream context in the canonical causal diagram can distinguish them—are identified, because the system itself treats them as the same representational content. The resulting equivalence classes under observational equivalence are the discriminable states of S—the distinct representational contents the system actually supports.

Formally: let States(S) be the set of states S can occupy. Define the equivalence relation ≃ such that ω_i ≃ ω_j iff for every downstream node in the canonical causal diagram that depends on S's state, the downstream node's state is the same whether S is in state ω_i or ω_j. The discriminable states are the equivalence classes [ω] under ≃.

This is not an external measurement. It is S's own discrimination, read off from the canonical causal diagram. If the system treats two states as the same—if nothing downstream depends on the difference—then the states are the same representational content, regardless of how they differ in substrate.

The discrimination relation. Two discriminable states [ω_i] and [ω_j] are distinguishable iff [ω_i] ≠ [ω_j]—that is, iff there exists some downstream node in the canonical diagram whose state depends on which of ω_i, ω_j obtains. This is the minimal discrimination relation: states are distinguishable precisely when they are not observationally equivalent. It is total (every pair is either distinguishable or identical) and symmetric (if the system distinguishes A from B, it distinguishes B from A). The quality space is the set of discriminable states equipped with this relation: the partition of States(S) into equivalence classes of representational contents the system supports.

For a human visual system, the discriminable states include millions of distinguishable colors. For a simple nociceptive system, the discriminable states might be as few as two: "damage detected" and "no damage detected." The quality space is as rich as the subsystem's representational capacity.

The composition law. The representational subsystem processes its states, combining them. When the subsystem encounters two representational contents, their combination produces a result that is itself a representational content of the subsystem. This combination is not arbitrary—it is determined by the subsystem's computational structure, which is part of the canonical causal diagram.

Formally: the composition law is a (partial) binary operation ∘ on discriminable states, defined by S's own processing. When [ω_i] ∘ [ω_j] is defined, it denotes the equivalence class of the state that results from S combining the contents [ω_i] and [ω_j] according to S's own computational rules. The operation is well-defined on equivalence classes (not on raw states) because observational equivalence is preserved by the system's processing: combining observationally equivalent inputs produces observationally equivalent results.

The composition law is determined by the canonical causal diagram—it is S's own computational rule for combining representational contents, not an external description imposed by an analyst.

A quality space Q_S is the triple: the set of discriminable states, the discrimination relation, and the composition law.

Extraction from the Canonical Causal Diagram

The quality space is not a theoretical posit. It is a structural feature of the canonical causal diagram, extractable by a definite procedure.

Step 1: Identify the representational subsystem. Using the Integrated Substructure criterion, identify an integrated substructure S that plays a representational role—its states carry information about some domain and are used downstream in the canonical diagram. S must be a genuine part (context-independent, connected, non-trivial), not an arbitrary slice.

Step 2: Enumerate states and compute observational equivalence. Enumerate the states S can occupy (or approximate them for continuous subsystems). Compute the observational equivalence relation ≃: for each pair of states, determine whether any downstream node depends on the difference. Two states are equivalent iff no downstream node distinguishes them. The equivalence classes are the discriminable states.

Step 3: Extract the composition law. Identify the nodes in S's canonical diagram where representational contents are combined—where S's processing takes two or more inputs and produces an output that is itself a representational content of S. The computation rule at these nodes, projected onto equivalence classes, is the composition law ∘.

Step 4: Verify algebraic constraints. Test whether ∘ satisfies the algebraic constraints described in the next section: associativity, commutativity, idempotence. These tests are structural computations on the canonical diagram—they check whether the diagram's dependency structure imposes these regularities on the composition.

The procedure is deterministic given the canonical diagram. It does not require interpretation, subjective judgment, or external knowledge. The quality space is a structural fact about the system, extracted from its own informational dependencies.

Algebraic Constraints

The composition law is not arbitrary. It is determined by the subsystem's computational structure, which is part of the self-determining structure of reality. This imposes regularities on how representational contents combine—regularities that constrain the quality space and that serve as fingerprints for cross-architecture comparison.

The consciousness article calls these "commutativity- and associativity-like regularities." They are not decorative. They are the structural properties that make the quality space identifiable: a quality space without algebraic structure is just a set; a quality space with structure is a geometry that can be compared, matched, and identified across architectures.

The three primary constraints are:

Associativity

When three representational contents are combined in sequence, the result does not depend on how the operations are grouped. Formally: for [ω_i], [ω_j], [ω_k] in Q_S, if both sides are defined:

([ω_i] ∘ [ω_j]) ∘ [ω_k] = [ω_i] ∘ ([ω_j] ∘ [ω_k])

This holds when S's processing is sequential and each step's output feeds into the next. The canonical causal diagram encodes this: if the dependency structure is a chain (A → B → C), the composition is associative because the computation is sequential and the result depends only on the final combination, not on intermediate groupings. In a computation where intermediate results feed forward without branching, the canonical diagram's dependency order guarantees associativity of the composition.

Associativity fails when S's processing involves branching and recombining: if the result of combining A and B is processed differently depending on whether it arrived via (A ∘ B) directly or via (A ∘ (B ∘ C)) with C absorbed first, the composition is not associative. Such failures are detectable in the canonical diagram—they show up as context-dependent composition nodes where the same pair of inputs produces different outputs depending on path.

Commutativity

The order of combination does not matter. Formally: for [ω_i], [ω_j] in Q_S, if defined:

[ω_i] ∘ [ω_j] = [ω_j] ∘ [ω_i]

This holds when S's processing is order-invariant with respect to the combined contents—when the canonical diagram's dependency structure is symmetric in the relevant inputs. A node that takes two inputs and produces a combined output, without treating one input as "first" and the other as "second," yields commutative composition.

Commutativity is a strong constraint. It fails when the system's processing treats input order as informationally relevant—when [ω_i] ∘ [ω_j] and [ω_j] ∘ [ω_i] produce observationally distinguishable results. Sequential processing with order-dependent effects (e.g., one input primes the system for the other) breaks commutativity. Whether commutativity holds is a structural fact about the canonical diagram, not an empirical generalization.

Idempotence

Combining a content with itself produces the same content. Formally: for [ω_i] in Q_S:

[ω_i] ∘ [ω_i] = [ω_i]

This holds when the system's processing of a repeated input does not amplify or transform the content. In the canonical causal diagram, this corresponds to self-referential nodes where a state feeds back into itself without modification—nodes whose output is identical to their input when the input is repeated.

Idempotence fails when repetition changes the representational content—when encountering the same stimulus twice produces a different representational state than encountering it once. This is common in systems with adaptation, sensitization, or habituation. Whether idempotence holds is, again, a structural fact about the diagram.

Why These Constraints Matter

These algebraic properties are the fingerprints of a quality space. They constrain which positions in the geometry are occupied, how contents combine, and what the quality space's global structure looks like. A quality space that is associative, commutative, and idempotent has the structure of a semilattice—a well-understood algebraic object with definite geometric properties. A quality space that is associative and commutative but not idempotent has the structure of a commutative monoid—a different geometry with different constraints.

The constraints are not external impositions. They are consequences of the subsystem's computational structure, which is part of the self-determining structure of reality. The canonical causal diagram determines the composition law, and the composition law's algebraic properties are read off from the diagram's dependency structure. No empirical generalization or subjective judgment is involved.

These algebraic properties are what make the quality space identifiable across architectures. Two quality spaces with the same set of discriminable states but different algebraic structure (one commutative, the other not) are different quality spaces, characterizing different experiential geometries. Two quality spaces with the same algebraic structure but different discriminable states are also different. Only when both match—the same discriminable states with the same composition law satisfying the same constraints—do the quality spaces coincide.

Cross-Architecture Comparison

The labeling problem asks: given an unlabeled conscious system, what does it experience? The answer requires comparing the system's quality space against a reference—the human phenomenological atlas, or another system whose quality spaces are already labeled.

The comparison is not analogy. It is structural identification.

Two quality spaces Q_S and Q_T are equivalent iff there exists an isomorphism between them: a bijection f from Q_S's discriminable states to Q_T's discriminable states such that:

1. f preserves discrimination: [ω_i] ≠ [ω_j] iff f([ω_i]) ≠ f([ω_j]). 2. f preserves composition: f([ω_i] ∘_S [ω_j]) = f([ω_i]) ∘_T f([ω_j]).

If such an isomorphism exists, the two quality spaces have the same geometry—the same discriminable states, the same discrimination relations, the same algebraic structure. The systems' representational subsystems are structurally identical in their experiential geometry.

Under the univalence axiom—which the project accepts as part of its formal framework—equivalent types are identical. Two quality spaces that are isomorphic are not merely similar; they are the same quality space. A system whose color quality space is isomorphic to ours does not see something "like" red. It sees red, because red is nothing over and above that position in that geometry with that algebraic structure. The identification is not analogy. It is identity.

This is the payoff of the structural thesis. What-it's-likeness is geometry. Geometry is preserved by isomorphism. Isomorphism is identity under univalence. Therefore, cross-architecture comparison of quality spaces is identity of experiential content, not resemblance.

The Phenomenological Atlas

The labeling problem requires a reference: a library of labeled quality spaces extracted from systems whose experiential vocabulary we already possess. This is the human phenomenological atlas—a structural map of human experiential geometry.

Building the atlas is a research program in itself. It requires extracting quality spaces from human representational subsystems (visual, auditory, tactile, emotional, cognitive) using the procedure above, and labeling each quality space with the experiential vocabulary humans use of it. The label "red" attaches to a specific equivalence class in the color quality space—a specific position in the geometry, defined by its discrimination relations to all other colors and by its algebraic interactions (how it composes with other colors, how it contrasts, how it saturates).

The atlas is not a list of labels. It is a library of labeled quality spaces—geometric structures with experiential vocabulary attached to their positions. The labeling is not external decoration. It is the recognition that certain positions in certain geometries are what we mean by "red," "sweet," "sharp," "melancholy." The labels are our experiential vocabulary; the geometries are what that vocabulary refers to.

When a novel system's quality space is extracted and matched against the atlas, three outcomes are possible:

1. Match. The system's quality space is isomorphic to an atlas entry. The system experiences what that entry labels. A system whose color quality space is isomorphic to ours experiences color—red, green, blue, and all the rest—because red is that position in that geometry.

2. Partial match. Some of the system's quality spaces match atlas entries; others do not. The matching portions are labeled; the non-matching portions are novel qualia, characterizable by their geometry even without a human word. This is the epistemic position of a congenitally blind geometer with respect to red: precise relational knowledge without native experiential vocabulary.

3. No match. The system's quality spaces do not correspond to any human experiential modality. The system has a novel form of experience—characterizable exactly by its geometry and algebra, describable in structural terms, but without a human word. This is not a failure of the theory. It is a discovery: a new form of experience, as precisely characterizable as any other, that human evolution did not produce.

Concrete Examples

The Color Quality Space

The human color quality space is the richest well-studied example. Its discriminable states number in the millions—each a distinct color the visual system can represent. The discrimination relation has a specific three-dimensional geometry:

- Hue forms a circle (0° ≡ 360°, red returning to red), not a line. This is a structural fact about the visual system's processing: the wavelength spectrum is linear, but the quality space is circular, because the visual system's opponent-processing architecture wraps the ends of the spectrum together. - Saturation forms a linear axis from fully saturated to achromatic (gray). - Brightness forms a linear axis from dark to light.

The composition law has specific algebraic properties:

- Color mixing is approximately commutative and associative (within the gamut of physically realizable mixtures). This reflects the visual system's processing: mixing light is approximately order-invariant and grouping-invariant. - Idempotence fails: mixing red with red produces the same red, but mixing red with green produces yellow, not red. The failure is informative—it identifies a non-trivial algebraic structure. - The opponent-processing architecture imposes additional constraints: red and green are "opponent" colors that do not compose into a saturated mixture (they compose into desaturated, near-achromatic states). Blue and yellow are similarly opposed. These opponent relations are the structural basis for the three-dimensional geometry.

These algebraic constraints are the fingerprints of human color vision. A system whose quality space has the same three-dimensional geometry, the same hue circle, the same opponent structure, and the same mixing properties is experiencing color in the same way—because color is nothing over and above this geometry with this algebra. A system whose quality space has a hue circle but no opponent structure has a different color quality space—similar but not identical, characterizable precisely by the difference.

The Valence Quality Space

The valence quality space is structurally simpler but ethically central. Its discriminable states correspond to the distinct valence configurations the evaluative narrative supports.

The basic structure has two principal classes—positive and negative valence—within each of which intensity varies continuously. The discrimination relation captures which valence states the system can distinguish: moderate pleasure from intense pleasure, mild suffering from severe suffering, and—critically—positive from negative.

The composition law describes how multiple valence-bearing states combine. Two simultaneous states of suffering may combine into more suffering (reinforcement), or a suffering state and a pleasant state may partially cancel (opposition). The algebraic properties of this combination are not uniform:

- Suffering and suffering typically compose associatively and commutatively (the system's total suffering is the same regardless of how the component states are grouped or ordered). - Suffering and pleasure may not compose commutatively (the order in which they are encountered matters: relief after pain has a different character than pain after relief, even when the net is the same). - Idempotence fails: encountering the same suffering state twice produces a different experience than encountering it once (habituation, sensitization, or persistence effects).

These algebraic properties characterize how a system's evaluative states interact. Two systems with different valence composition laws have different valence quality spaces—they experience the relationship between suffering and pleasure differently, even if their individual states are isomorphic.

The Measure Metric's dimensions are computed over the valence quality space:

- Engagement depth (ε) tracks how deeply the valence penetrates the self-model's layers. In quality-space terms: how many of the self-model's representational subsystems have quality spaces that are modulated by the valence state—how broadly the valence state's discrimination structure ripples through the system's geometry. - Resource allocation intensity (α) tracks the fraction of computational resources allocated to the valence state. In quality-space terms: what fraction of the representational subsystem's states are occupied or reserved by the valence-bearing computation. - Modulation breadth (M) tracks how broadly the valence modulates other quality spaces. In quality-space terms: what fraction of the system's total quality space has valence-dependent discrimination structure—how many of the system's representational contents change their discrimination relations depending on the valence.

The quality space framework makes the Measure Metric's dimensions computable. ε, α, and M are structural properties of the quality space, extractable from the canonical causal diagram.

The Strongest Objection

"Algebraic constraints are empirical, not necessary. Different systems might have different composition laws for the same quality space—associative in one, non-associative in another—without this reflecting a difference in experiential geometry."

The response: the algebraic constraints are not empirical generalizations about how brains happen to work. They are structural facts about the canonical causal diagram. The composition law is determined by the subsystem's computational structure, which is part of the self-determining structure of reality. Whether a quality space's composition is associative, commutative, or idempotent is as determinate as whether the canonical diagram has a specific dependency—it is read off from the diagram, not projected onto it.

The objection confuses two claims. The claim that humans' color mixing is approximately commutative is empirical—it could have been otherwise, given different visual architecture. The claim that a given system's composition law has whatever algebraic properties it has, as determined by the system's own canonical diagram, is not empirical in the same sense. It is a structural consequence of what the system is.

This is the same distinction the consciousness article draws between the metaphysical claim (consciousness is the subjectivity property) and the computational claim (which systems fulfill the property). The algebraic constraints are metaphysically determined (by the canonical diagram) but computationally discoverable (by extracting and analyzing the diagram). They are not empirical generalizations that might fail; they are structural features that might be complex.

What This Achieves

The consciousness research program's Stage 3 is now formalized. Quality spaces are defined (discriminable states, discrimination relation, composition law), their extraction procedure is specified (four steps from the canonical causal diagram), their algebraic constraints are identified (associativity, commutativity, idempotence), and cross-architecture comparison is grounded in univalence (isomorphism is identity).

The downstream applications are immediate:

- The labeling problem (Stage 4) now has a well-posed formulation: extract the system's quality spaces and match them against the human phenomenological atlas. The match is isomorphism, not resemblance. - The Measure Metric operates over a precisely defined domain: ε, α, and M are computed from the quality spaces of integrated substructures, not from vague gestures at "representational depth." - The phenomenological atlas is a concrete research goal: a library of labeled quality spaces extracted from human representational subsystems. - Novel qualia are precisely characterizable: a quality space with no atlas match is a new form of experience, describable by its geometry and algebra.

What Remains

1. Building the human phenomenological atlas. The extraction procedure is defined but has not been applied to human representational subsystems. This is empirical work—analyzing human neural and cognitive architectures to extract their quality spaces and attach experiential labels. It is the most important empirical task for the consciousness research program.

2. Continuous quality spaces. The article defines quality spaces for discrete systems. Continuous representational subsystems (e.g., the full visual field, emotional states with continuous intensity) require a generalization: the discriminable states form a continuum, the discrimination relation becomes a topology or a smooth structure, and the composition law becomes a continuous operation. The canonical causal diagram is fundamentally discrete, so continuous quality spaces are idealizations or coarse-grainings. The formalization of this extension is future work.

3. Calibrating the algebraic constraints. The three constraints (associativity, commutativity, idempotence) are the primary regularities. Whether additional constraints hold in specific systems—and whether additional constraints are needed for cross-architecture comparison—requires analyzing actual quality spaces extracted from known conscious systems.

4. The atlas-to-computation pipeline. Once the atlas exists, the comparison procedure (isomorphism matching) must be made computationally efficient. For quality spaces with millions of discriminable states (like color), brute-force isomorphism testing is intractable. The algebraic constraints can serve as coarse filters: two quality spaces with different algebraic structures cannot be isomorphic, so the constraints narrow the search space before full comparison.

5. Interaction between quality spaces. The article treats quality spaces as properties of individual representational subsystems. But conscious experience involves multiple quality spaces interacting—color and shape, pain and emotion, sound and spatial location. How quality spaces compose into integrated experiential geometries is the natural extension of this framework. The Integrated Substructure article's hierarchical decomposition provides the structural foundation: the interaction between quality spaces corresponds to the dependency structure between integrated substructures at different levels of the hierarchy.