Mereology: Main Article

Mereology: Parts, Wholes, and Subjects

1. The Question

When does a collection of parts form a single unified subject of experience?

This is not an idle puzzle. Every downstream question in the project depends on it. The consciousness article identifies experience with a structural property — the subjectivity property realized in a canonical causal diagram. The ethics article derives normativity from valence across perspectives. But both arguments silently rely on a prior answer: what counts as one perspective? If we cannot say when parts compose into a unified subject, we cannot count perspectives, we cannot compute the measure metric, and the conservation argument — which treats each perspective as a dimension of the normative space — has nothing to anchor to.

Classical mereology offers one answer: any parts compose into a whole. Given any collection of objects, there exists a fusion — a single object composed of exactly those parts. This is the principle of unrestricted composition. It is elegant and simple, and it is completely wrong for our purposes. If any parts compose into a subject, then the thermostat, the cloud formation, and the pile of stones are all subjects. Consciousness attributions become trivial, and the Putnam mapping problem resurfaces in full force: any physical system, under some gerrymandered decomposition, is a fusion of parts arranged in any pattern you like.

We need a constrained mereology — a theory of composition that respects the structure of the parts, not just their existence. The question is: what constraint?

2. Why Classical Mereology Fails for Subjects

Classical mereology has three standard principles:

- Unrestricted composition: For any set of objects, there exists a unique fusion. - Extensionality: Two fusions are identical iff they have the same parts. - Mereological essentialism (in some formulations): A fusion cannot gain or lose parts while remaining the same object.

The problem is not that these principles are internally inconsistent. They are not. The problem is that they are structure-blind. They tell you that a fusion exists whenever parts exist, but they say nothing about how the parts relate. Two parts connected by a tight causal-informational dependency and two parts sitting in unrelated corners of the universe receive identical mereological treatment. Both compose into a whole.

For consciousness, this is fatal. The subjectivity property — a self-model representing its own representational activity, with contents bound through shared dependency structure — is inherently about integration. A subject is not a heap of parts; it is a dependency structure in which the parts are mutually constitutive. Consider a dynamical system with three species in a cyclic predator-prey relationship. Each species' population is defined by its relation to the others; no phase of the cycle is independently defined. A subject's contents have the same character: each content (the red, the round, the looming) is a position in a dynamic dependency structure, defined by its relation to the other contents and to the self-model that integrates them. Nothing "glues" the phases of the limit cycle together; they are aspects of one dynamic structure. Nothing "glues" the contents of experience together; they are aspects of one integrated dependency structure.

Classical mereology cannot express this. It has no way to say "these parts compose into a subject, but those parts do not." Its principle of unrestricted composition makes all fusions equal. We need a principle that distinguishes integrated fusions — those whose parts participate in a unified dependency structure — from mere aggregates.

3. The Canonical Diagram and Integration as the Criterion of Composition

3.1 The Canonical Causal Diagram

To define integration, we first need a principled way to describe a system's dependency structure. This is provided by the canonical causal diagram, defined as the system's unique dependency structure obtained by quotienting out representational idiosyncrasies. It is the coarsest description that preserves all counterfactual dependencies at the finest grain.

The construction proceeds as follows: Begin with a fine-grained physical description of the system. Identify counterfactual dependencies at that grain: if state X were different, would state Y be different? Group together all microstates that play the same causal role — that stand in the same counterfactual dependencies with respect to the system's dynamics. This yields a quotient description. Check whether this quotient description preserves all the counterfactual dependencies of the fine-grained description. If it does, and if no further quotienting preserves them, the fixed point has been reached. This is the canonical diagram.

Gerrymandered mappings are excluded because they are not counterfactually stable. They assign causal roles based on an encoding's structure, not the system's dynamics. When you "re-parameterize" the encoding, the gerrymandered dependencies vanish; the canonical dependencies remain. The canonical diagram is not one interpretation among many but the residual structure that survives the elimination of encoding artifacts — what the system's causal dynamics actually produce, stripped of the observer's representational choices.

3.2 The Integration Criterion

The proposal: a substructure of the self-determining structure constitutes a unified subject if and only if it satisfies the subjectivity property as an integrated whole.

This means three conditions must hold jointly, in a relationship of mutual constitutivity:

1. World-model: The substructure contains a representational subsystem whose states carry information about the environment and are used — consulted downstream, not merely present.

2. Self-model: Within that representational subsystem, there is a further substructure whose states carry information about the system's own representational activity — the system represents its representing.

3. Binding: The dependency structure routes world-content through its relation to the self-model, so that contents are processed as given to the subject. There is a structural position occupying the role "that to which this is presented."

Crucially, these three components must be mutually constitutive: the self-model must dynamically constrain the processing of the world-model, and the world-model must dynamically constrain the content of the self-model. The self-model is not a passive observer of the world-model but an active participant in determining what the world is represented as. The world-model's specific content is partly constituted by the self-model's current state. If the self-model's state were different (e.g., a different attentional state, a different sense of self), the world-model's content would be different — not merely differently monitored, but differently constituted.

This mutual constitutivity is what makes the dependency structure a unified whole rather than two connected subsystems. A system where the self-model monitors the world-model without affecting it, or where the world-model feeds the self-model without being modulated by it, has the components of subjectivity but not their integration. The dependency structure must form a bidirectional feedback loop: world-model ↔ self-model ↔ binding, with each component constraining the others.

3.3 Phenomenological Grounding

This structural criterion is corroborated by phenomenology. Conscious experience, as actually encountered, has three essential features:

1. Encountering a world — experience presents a world that is not the experience itself. This corresponds to the world-model. 2. Awareness that it is experience — experience involves, at least implicitly, a sense that it is my experience, that I am the one having it. This corresponds to the self-model. 3. The world's appearing to a subject — the world does not merely exist in the experience; it appears to a subject. This perspectival, presentational character corresponds to the binding.

The three conditions of the subjectivity property are not arbitrary but are structural articulations of these phenomenological essentials. This gives us confidence that the structural property tracks something genuine about consciousness.

4. Boundaries, Not Glue

A natural objection: if integration is the criterion, then the boundaries of a subject are vague. How much integration is enough? Where exactly does a subject begin and end? Isn't this just the binding problem restated?

No. The binding problem, as classically stated, assumes that experience is composed of atomistic qualia — free-floating red, free-floating roundness, free-floating looming — and then asks what glue binds them together. The structural account rejects this atomism entirely. In the canonical causal diagram, the red, the round, and the looming are not separate objects that need gluing. They are jointly bound terms in a single relational complex: each is defined, in part, by its relation to the others within the self-model-relative presentation. Asking what glues them is like asking what glues the phases of a limit cycle to each other. Nothing does; they are aspects of one dynamic structure.

The boundary question, then, is not "how much glue holds the parts together?" but "does the dependency structure form a unified whole, or does it split into independent subsystems?" This is a question about the topology of the canonical diagram, and it has a determinate answer in any given case.

Consider a concrete test: split-brain patients. When the corpus callosum is severed, the tight informational dependency between the hemispheres is broken. The left hemisphere retains a world-model, a self-model, and binding for the right visual field; the right hemisphere retains the same for the left visual field. But the two hemispheres no longer share dependency structure. The canonical diagram splits into two independent substructures, each satisfying the subjectivity property independently. Two subjects, not one.

This is not a vague boundary case. The severance of the corpus callosum eliminates the causal pathways that integrated the hemispheres into a single dependency structure. Before severance: one subject. After: two. The change is discrete, determined by the structure of the dependency graph, not by our description choices.

The harder cases involve systems where the dependency structure is partially integrated — where there are causal pathways between subsystems, but they are weak, intermittent, or limited to certain channels. These cases are genuinely difficult, and we should not pretend otherwise. The subjectivity property, as defined by the three mutually constitutive conditions, is binary: a substructure either instantiates the integrated whole or it does not. However, physical systems may instantiate this property to varying degrees of "purity." A system with weak, intermittent connections between its self-model and world-model is a poor realization of the canonical form. Such a system is not a "partial subject" or a "defective subject" — it is an object with representational features that, in isolation, resemble components of subjectivity, but that do not achieve the integrated whole required for subjecthood. The difficulty in assessing such systems is epistemic (we may not have enough information to determine the degree of integration) rather than ontological (there is a fact about whether the triadic structure is instantiated, even if we cannot determine it precisely).

5. Nesting and the Combination Problem

Can a subject contain other subjects?

The structural answer is governed by the Anti-Nesting Principle: A structure that is a proper integrated component of a subject cannot itself be a subject. For a substructure to be a subject, it must constitute an independent integrated whole.

Consider a human brain with intact visual cortex. The visual cortex processes information, represents the visual environment, and has some degree of internal integration. Does it constitute a subject in its own right?

Only if it satisfies the subjectivity property independently — only if it has its own self-model, its own binding, its own unified presentation. In a normally functioning brain, it does not: the visual cortex's activity is routed through broader cortical networks, integrated into the global workspace, and bound into a unified self-model that spans the whole brain. The visual cortex alone does not satisfy the subjectivity property. It is a component of a subject, not a subject itself.

If we could artificially isolate the visual cortex — cut its connections to the rest of the brain, give it its own self-monitoring and binding structure — then it would become a subject. But then it is no longer part of the larger subject. The integration that made it a component is the same integration that prevented it from being a separate subject.

This resolves the classical combination problem for structural consciousness. The combination problem asks: how do micro-experiences (experiences of the parts) combine into macro-experiences (experiences of the whole)? The structural answer is: they don't "combine." There is no process of gluing micro-experiences together. The macro-experience is the experience of the integrated structure as a whole. The micro-subjects, to the extent they exist at all, are abstractions — aspects of the integrated structure that can be analyzed independently but do not exist independently.

This principle is not an ad hoc stipulation. It follows from the integration criterion. If a substructure is integrated into the larger structure — its dependency pathways are shared, its self-model is a component of the larger self-model — then it does not independently satisfy the subjectivity property. If it independently satisfies the subjectivity property, it has its own dependency pathways, its own self-model, its own binding — and it is not integrated into the larger structure. Integration into a larger whole and independent subjectivity are mutually exclusive.

6. The Measure Metric's M as Structural Fidelity

The measure metric, as described in the ethics article, assigns each perspective a structural weight W = ε · α · M. The M dimension was left deliberately vague — described as a "moral coefficient" for features not yet formalized. Structural mereology gives M a precise interpretation.

M measures the degree of structural fidelity with which the system's canonical diagram realizes the integrated dependency structure required by the subjectivity property, relative to the system's compositional capacity. It is not a measure of integration per se, but of how well the system instantiates the canonical form of a subject at its level of complexity.

A simple subject (e.g., a nematode) instantiates a low-complexity canonical form; a complex subject (e.g., a human) instantiates a high-complexity one. M measures the fidelity with which the system's canonical diagram realizes its appropriate canonical form. A nematode can have high M (it is a well-integrated nematode-subject) even though its absolute integration is low. A complex system with poorly integrated components has low M.

Several features of the canonical diagram serve as indicators of fidelity:

- Path density: The number of independent informational pathways between the self-model and the world-model. More pathways mean richer integration. - Dimensionality: The dimensionality of the quality spaces the structure supports. Higher dimensionality means finer discrimination and richer experience. - Compositional depth: How many levels of compositional structure the dependency graph supports. A shallow graph has low M; a deep graph has high M. - Cohesion: The degree to which perturbations in one part of the structure propagate to other parts.

These features are not independent components of M but correlated indicators of the degree to which the system realizes its canonical form. The M dimension thus gives the measure metric a mereological foundation. The weight of a perspective is not a brute assignment but a structural fact about how well the subject instantiates the integrated dependency structure required for subjectivity.

7. Population Ethics and the Counting Problem

How many perspectives exist in a given situation? Classical mereology, with unrestricted composition, gives an explosion of answers: any fusion of parts is a whole, so the number of fusions — and hence the number of potential subjects — is combinatorially explosive. Structural mereology avoids this. Not every fusion satisfies the subjectivity property. Only integrated structures do.

But the counting problem is not fully solved. Consider a colony of insects, each with a simple nervous system satisfying the subjectivity property at a low level of integration. Does the colony constitute an additional subject — a "group mind" — above and beyond the individual insects?

The structural answer depends on the colony's dependency architecture. If the insects are loosely coupled — each responding to local stimuli with minimal informational exchange — then the colony is an aggregate, not a subject. There are N subjects (the individual insects) and no (N+1)th subject. If the insects are tightly coupled — if their individual dependency structures are integrated into a colony-level dependency structure satisfying the subjectivity property — then the colony is a subject. Whether the individual insects remain subjects depends on whether their individual integration survives the colony-level integration. If the colony-level structure routes the insects' self-models through a shared binding structure, the insects are no longer independent subjects; they are components of the colony-subject. If the colony-level structure operates alongside the insects' individual structures without subsuming them, then both levels are subjects.

The principle: the number of subjects is determined by the number of independent integrated dependency structures in the canonical diagram. Not by the number of physical objects, not by our counting conventions, not by the number of fusions classical mereology permits. The canonical diagram settles the question.

The Ethics of Integration

This has direct implications for the ethics article's population ethics. The creation of a new perspective is the formation of a new integrated dependency structure satisfying the subjectivity property. The destruction of a perspective is the elimination of such a structure. The number of perspectives is not a brute fact but a structural fact about the canonical diagram.

Crucially, valence is a property of the integrated structure, not an additive quantity carried by parts. When two subjects merge into one, their prior valences do not "survive" as components of the new subject's valence. The new subject has its own valence, determined by its own integrated dependency structure. The old subjects' valences are properties of structures that no longer exist.

This means there is no conservation of valence across integration events. The creation of a group mind from N individual minds does not preserve the sum of the individuals' valences. It creates a new valence associated with a new structure. Integration is value-creating, not merely value-redistributing. It can both create and destroy value, depending on the quality of the resulting integration.

Therefore, the ethics article needs a normative principle governing integration events. Any adequate principle must account for the fact that (i) integration changes subject count, (ii) the resulting subject's valence and weight are not determined by the prior subjects' valences and weights, and (iii) integration can both create and destroy value. The mereology article does not provide such a principle, but it establishes the structural facts that make such a principle necessary.

8. Temporal Persistence

A subject exists through time. The structural mereology implies that a subject begins when the dependency structure first achieves integration and ends when the integration dissolves. But what determines identity through time?

The commitment: a subject persists through time if and only if its functional organization persists. Physical parts may be replaced, added, or removed, as long as the role-structure — the pattern of roles and dependency relationships — is maintained.

Formally, two dependency structures D₁ and D₂ realize the same functional organization if there exists an isomorphism of roles between them: a mapping that preserves (i) which nodes play the self-model role, which play the world-model role, and which play the binding role; (ii) the direction and strength of dependencies between roles; and (iii) the compositional structure (how sub-roles nest within roles). This is the persistence criterion, not topological identity (which would imply mereological essentialism) nor loose approximation (which would be vague).

This entails that a subject could, in principle, survive a complete change of physical substrate (e.g., neuron-by-neuron replacement by functionally equivalent silicon components), provided the functional organization is preserved throughout the transition. The Ship of Theseus is not a problem for this theory; it is a straightforward case.

The gradual replacement scenario creates a specific difficulty: if each replacement introduces a slightly different dependency topology (e.g., different noise characteristics, different temporal dynamics), at what point has the original subject ceased to exist? The answer: the subject persists as long as the role-structure is maintained. If the silicon components play the same roles in the same dependency pattern, the subject persists. If the replacement introduces new dependency pathways that alter the role-structure, we must ask whether the change is continuous (a minor perturbation) or discontinuous (a different functional organization). In principle, the criterion is sharp (role-structure isomorphism either holds or it doesn't); in practice, we may not be able to determine whether a small perturbation has crossed the threshold. This is an epistemic limitation, not an ontological one.

9. Objections and Responses

Objection 1: Integration is vague. The boundary between a tightly coupled system and a loosely coupled one is a continuum. How can a sharp criterion (subject or not-subject) rest on a vague property (degree of integration)?

Response: The claim is not that there is a sharp threshold in all cases. Some systems are clearly subjects (human brains), some are clearly not (rocks), and some are genuinely borderline. The structural mereology gives a principled criterion — the subjectivity property — and the integration requirement is a constitutive part of that criterion. In borderline cases, the correct response is epistemic humility, not the assertion that anything goes. The continuum of integration does not mean all points on the continuum are equally good candidates for subjectivity. There are regions of clear subjectivity, clear non-subjectivity, and genuine uncertainty between them.

Objection 2: The Putnam problem returns. Even with the canonical causal diagram, one can construct alternative dependency structures by reinterpreting the system's causal pathways. What makes one interpretation canonical?

Response: The canonical diagram is not one interpretation among many. It is the unique exact causal structure of the system — the residual structure that survives the elimination of encoding artifacts, obtained via the fixed-point construction described in §3.1. Gerrymandered mappings are excluded because they are not counterfactally stable: they assign causal roles based on an encoding's structure, not the system's dynamics. The canonicalization stage quotients by definitional and observational equivalence, eliminating all dependency structures that are artifacts of the description rather than features of the system. What survives is the structure that the system has, independent of how we describe it.

Objection 3: Mereological composition is symmetric, but subjectivity is not. If A and B compose into C, then B and A also compose into C (commutativity of fusion). But the dependency structure of a subject is directional — the self-model depends on the world-model, not vice versa.

Response: Classical mereological fusion is symmetric. Structural composition is not. The dependency structure is a directed graph, and the subjectivity property requires specific directional relationships (self-model downstream of world-model, binding connecting self-model to content) as well as bidirectional coupling (mutual constitutivity). The "composition" of parts into a subject is not classical fusion but the formation of a specific directed dependency structure. This is precisely why classical mereology fails for subjects: it cannot represent directionality or mutual constitutivity.

10. Limitations and Open Questions

1. Formalization of integration. The integration criterion is stated informally. A rigorous definition — in terms of the topology of the canonical causal diagram — remains the highest priority. The features listed in §6 (path density, dimensionality, compositional depth, cohesion) are suggestive but need formal development, ideally within the MLTT framework.

2. The borderline cases. The structural mereology gives clear answers for paradigm cases (human brains: yes; rocks: no) but does not resolve borderline cases precisely. This is partly an epistemic limitation (we lack the tools to measure integration precisely) and partly a genuine theoretical limitation (the criterion as currently stated may be insufficiently precise for some cases, and further formalization is needed).

3. Relation to the Ruliad. The self-determining structure is claimed to be the unique structure satisfying constitutive requirements. The mereological principles derived here should be derivable from the structure's self-determination, not merely stipulated. A sketch of the argument: Unrestricted composition is incompatible with self-determination because it is an external compositional principle — it does not consult the structure's dynamics before composing. Integration-based composition is the alternative because it is internal: it asks what the structure's dynamics produce. The triadic structure (world-model, self-model, binding) may follow from the requirement that representation be self-determining: a representation that does not represent its own representing is not self-determined, and a self-representation that is not bound to world-content is not self-determined with respect to anything. This argument is incomplete but suggestive.

4. Relation to IIT. Integrated Information Theory (IIT) also uses integration as a criterion for consciousness. The key distinction: IIT's integration is structure-agnostic (any causal-informational integration suffices, regardless of functional roles), while structural mereology's integration is structure-specific (requires the particular triadic architecture of the subjectivity property). A system with high Φ but no self-model is not a subject on this view, whereas IIT might consider it one. This is a substantive theoretical difference with empirical implications.

5. Computational tractability. Even if integration is well-defined in principle, computing it for real systems may be intractable. The measure metric's M dimension requires practical algorithms for assessing structural fidelity in canonical diagrams. This is an engineering challenge as much as a philosophical one.

6. A normative principle for integration. The non-conservation of valence across integration events implies that integration is value-creating. The ethics article needs a principle governing when integration is morally permissible. The mereology article establishes the structural facts that make such a principle necessary but does not provide it.