Chapter 14: Collective Moral Agency — Aggregation, Emergence, and the Tensors of Shared Obligation

RUNNING EXAMPLE — Priya’s Model

Priya cannot fix this alone. HealthBridge’s collective agency tensor involves: Priya (designed the system), Dr. Osei (flagged the problem), Sarah Chen, VP of Product (controls deployment), the Board (sets priorities), the FDA (regulates trial access), and the patients (who have no voice in the process). The structure tensor is nonzero: HealthBridge’s decisions are not the sum of individual choices. The company has emergent obligations that no single employee carries. Priya can document the bias, but only Sarah can change deployment. Sarah can authorize a fix, but only the Board can allocate budget. The responsibility remainder—for patients already harmed—does not vanish with organizational restructuring. Mrs. Voss’s melanoma did not wait for the org chart to sort itself out.

14.1 Beyond Individual Agents

The geometric framework developed in Parts II and III has focused on a single moral situation evaluated from various perspectives. The obligation vector O^μ at a point p∈M represents what is owed; the interest covector I_μ represents what matters; the satisfaction S=I_μO^μ measures how well duty meets need. Even the multi-agent evaluation tensor E_νa^μ of Chapter 6 (§6.6) — which indexes by agent a — treats each agent’s moral position as independently specifiable.

But much of moral life involves collective agents: corporations, states, committees, families, professions, religious communities, and (increasingly) human-AI teams. These collectives pose questions that the individual-agent framework cannot answer:

Does a corporation have moral obligations distinct from those of its employees?

When a committee decides wrongly, who bears responsibility — and does the sum of individual responsibilities exhaust the collective responsibility?

Can an institution be unjust even if every individual within it acts permissibly?

How do individual moral tensors combine into collective structures — and does the combination introduce new moral content not present in the components?

This chapter develops the geometry of collective moral agency. The central result is that collective moral tensors have emergent components — terms that cannot be expressed as sums or products of individual tensors. The collective is not merely the aggregate of its members. It is a higher-rank moral entity with its own geometric structure.

14.2 The Aggregation Problem

Why Simple Addition Fails

The most natural first attempt at collective moral structure is summation: the collective obligation is the sum of individual obligations.

O_collective^μ=^?∑_a∈A^​ O_a^μ

where A is the set of members and O_a^μ is member a’s obligation.

This fails for at least three reasons.

1. Collectives have obligations that no member has. A corporation has an obligation to pay its debts. No individual employee has this obligation in their personal capacity. A nation has an obligation to provide for its citizens’ basic needs. No individual citizen has this obligation singly. The collective obligation is not a sum of individual obligations — it has no individual summands.

2. Individual obligations may cancel or reinforce in non-additive ways. If Alice has an obligation to keep a secret and Bob has an obligation to reveal it, the group’s obligation is not the zero vector (cancel) nor the sum of the two vectors (reinforce). It depends on the structure of the group — who has authority, what the group’s purpose is, what decision procedure is in force. The combination is mediated by structure, not by addition.

3. The dimensionality may change. An individual agent operates in the nine-dimensional moral manifold of Chapter 5. A collective agent may operate in a higher-dimensional space — the nine ethical dimensions, plus structural dimensions encoding the collective’s internal organization. The collective obligation vector may have components along dimensions that have no individual analogue.

The Tensor Product Approach

A better starting point is the tensor product of individual moral spaces. If agent a operates in a moral tangent space T_a≅R⁹, the collective of n agents operates in:

T_collective=T₁⊗T₂⊗⋯⊗T_n

This space has dimension 9ⁿ — far larger than the 9n dimensions of the direct sum T₁⊕T₂⊕⋯⊕T_n that simple addition would use.

Why the tensor product? Because it captures correlations. A state in the direct sum T₁⊕T₂ specifies agent 1’s obligation and agent 2’s obligation independently — 9+9=18 components. A state in the tensor product T₁⊗T₂ specifies the full joint obligation, including all correlations between the agents’ moral positions — 9×9=81 components. The additional 81-18=63 components encode the relational structure of the collective: how the agents’ obligations interact, constrain, and depend on one another.

This is precisely the structure encountered in Chapter 13 (§13.8) in the quantum context. The tensor product of individual Hilbert spaces contains entangled states — joint states that cannot be decomposed into individual states. The classical analogue is the same: the tensor product of individual moral spaces contains collective moral states that cannot be decomposed into individual obligations.

14.3 The Collective Agency Tensor

Definition

Definition 14.1 (Collective Agency Tensor). Let A={1,…,n} be a set of agents. The collective agency tensor is a multilinear map:

Convention. All agents in A evaluate the same base point p ∈ M; the collective agency tensor T is defined at p and transforms as a tensor under coordinate changes on the moral manifold. The agent index a ∈ A is a label, not a tensor index: T does not transform covariantly or contravariantly in the agent label. Formally, T is a family of (1,1)-tensors on T_pM, one for each coalition subset, parametrized by the agent set A.

C:T₁^*⊗⋯⊗T_n^*→T_M

that takes one interest covector from each agent’s cotangent space and produces an obligation vector in the moral tangent space. In components:

O_collective^μ=C_ν1ν2⋯νn^μ I_ν1⁽¹⁾ I_ν2⁽²⁾ ⋯ I_νn⁽ⁿ⁾

The collective agency tensor C is a (1,n) -tensor — one upper index (the output obligation dimension) and n lower indices (one for each agent’s interest dimension). Its rank increases with the number of agents: a two-agent collective has a rank-3 tensor ( 9×9×9=729 components), a three-agent collective has a rank-4 tensor ( 9⁴=6561 components), and so on.

Decomposition: Individual and Emergent Components

The collective agency tensor can be decomposed into components attributable to individual agents and a remainder that is genuinely collective.

Definition 14.2 (Decomposition of Collective Agency). For a two-agent collective, the agency tensor decomposes as:

C_ν1ν2^μ=C_(ν1)^μδ_ν2+δ_ν1C_(ν2)^μ+E_ν1ν2^μ

where: - C_(ν1)^μδ_ν2 is agent 1’s individual contribution (independent of agent 2’s interests) - δ_ν1C_(ν2)^μ is agent 2’s individual contribution - E_ν1ν2^μ is the emergent component — the part of collective agency not attributable to either individual

The emergent component E vanishes if and only if the collective obligation is the sum of individual obligations. When E≠0, the collective has moral agency that exceeds the combination of its members.

Measuring Emergence

The degree of emergence of a collective can be quantified:

η=(∥E∥)/(∥C∥)

The emergent component E vanishes if and only if the collective obligation is the sum of individual obligations. When E≠0 , the collective has moral agency that exceeds the combination of its members.

Claim. For most real collectives of interest (corporations, states, institutions), η>0. The emergent component is nonzero because the collective’s structure — its decision procedures, role differentiation, and institutional purpose — generates obligations that no individual member bears.

Computational implementation (February 2026). The DEME V3 reference implementation realizes the collective agency tensor as a rank-5 MoralTensor with coalition indices. The decomposition into individual and emergent components is computed explicitly, and the degree of emergence η is a reportable diagnostic. V3 also implements Shapley value credit assignment for fair allocation of collective moral responsibility:

ϕ_i(v)=∑_S⊆N\{i}^​ (|S|! (|N|-|S|-1)!)/(|N|!)[v(S∪{i})-v(S)]

where v is the coalition value function derived from the collective agency tensor. This enables quantitative answers to the question “how much of the collective obligation is attributable to agent i?”—a question that Chapter 18 identifies as critical for human-AI responsibility allocation. Nash equilibrium and correlated equilibrium computations provide multi-agent coordination under norm constraints, extending the Norm-Constrained Stochastic Game formalism of Chapter 19 to strategic settings.

ErisML references: demo_game_theory.py, demo_shapley.py. The Nash equilibrium computation is demonstrated in demo_game_theory.py, which implements the StrategicLayer for a Prisoner’s Dilemma payoff matrix — the canonical example of multi-agent moral coordination where individually rational choices produce collectively suboptimal outcomes. The Shapley value computation is demonstrated in demo_shapley.py, which applies the CooperativeLayer to an airport runway cost allocation problem: each airline’s fair share of the collective infrastructure cost is computed as its marginal contribution averaged over all possible coalition orderings, exactly implementing the formula φᵢ(v) described above.

14.4 The Structure Tensor

What Makes a Collective More Than a Set

A collective agent is not merely a set of individuals. It has structure: decision procedures, role assignments, authority hierarchies, persistence conditions, and representational conventions. This structure mediates the relationship between individual and collective moral positions.

Definition 14.3 (Structure Tensor). The structure tensor of a collective A is a (0,2) -tensor Σ_ab on the space of members, encoding the structural relationships between them:

Clarification. The “space of members” on which the structure tensor is defined is the real vector space ℝ^|A|, where |A| is the cardinality of the agent set A. Each basis vector e_a corresponds to one agent a ∈ A. The component S^a_{bc} measures the degree to which the relationship between agents b and c structurally influences agent a’s moral standing.

Σ_ab>0: agents a and b have aligned roles (authority flows in the same direction)

Σ_ab<0: agents a and b have opposing roles (checks and balances)

Σ_ab=0: agents a and b are structurally independent

The structure tensor is generally asymmetric: Σ_ab≠Σ_ba when the relationship between a and b is directional (a supervisor-subordinate relationship, for instance). The symmetric part Σ_(ab)=(1)/(2)(Σ_ab+Σ_ba) encodes the mutual structural relationship. The antisymmetric part Σ_[ab]=(1)/(2)(Σ_ab-Σ_ba) encodes the power differential — the extent to which the structural relationship favors one party over the other.

This connects to the power tensor P_μν introduced in Chapter 6 (§6.9): the antisymmetric part of the structural tensor is the member-indexed version of the structural inequity tensor.

How Structure Shapes Collective Obligation

The collective agency tensor is related to the individual agency tensors through the structure tensor:

C_ν1ν2^μ=∑_a,b^​ Σ_ab E_νa^μ(a)⊗E_νb^μ(b)+E_ν1ν2^μ

The first term says: the collective obligation depends on the individual evaluation tensors E_ν^μ(a) of each member, weighted and combined according to the structure tensor Σ. The second term is the emergent remainder — the part of collective agency that cannot be expressed even through structure-mediated combination of individual agencies.

Example: A two-person partnership. Alice and Bob form a law firm. The structure tensor has Σ_AB=Σ_BA=0.5 (equal partnership, symmetric). The collective agency tensor is approximately the average of individual agency tensors, with a small emergent component arising from the firm’s identity as a legal entity (the firm’s obligation to clients exists independently of either partner’s individual obligations).

Example: A corporate hierarchy. A CEO and an employee have Σ_CE=0.8 (the CEO’s position strongly shapes the collective’s action) and Σ_EC=0.2 (the employee’s position weakly shapes it). The antisymmetric part Σ_[CE]=0.3 encodes the power differential. The collective obligation is dominated by the CEO’s evaluation tensor, with the employee’s contribution attenuated by the structural asymmetry.

14.5 Emergent Obligations

Obligations No Member Bears

The emergent component E of the collective agency tensor represents obligations that exist at the collective level but not at the individual level. These obligations are not mysterious or metaphysical — they arise from the structure of the collective, just as the properties of a molecule arise from the arrangement of its atoms, not from the atoms individually.

Example 1: Corporate debt. A corporation borrows money. The obligation to repay belongs to the corporation. No individual employee, officer, or shareholder has this obligation in their personal capacity (this is the legal content of limited liability). The obligation is an emergent component of the corporate agency tensor — it appears only in E, not in any individual evaluation tensor E_ν^μ(a).

Example 2: National defense. A nation has an obligation to defend its citizens. No individual citizen has this obligation singly. The obligation emerges from the collective structure — the existence of a polity with a government, a territory, and a citizenry. A random collection of individuals in the same territory, without political structure, would have no collective defense obligation. The obligation requires the structure tensor Σ to be nonzero in specific ways (political authority, territorial jurisdiction, citizen-state relationships).

Example 3: Institutional injustice. A hiring system may be structurally unjust — systematically disadvantaging a particular group — even if every individual decision-maker acts in good faith according to their individual obligations. The injustice is an emergent property of the collective structure: it appears in E, not in any individual E_ν^μ(a). This is the geometric content of the concept of structural injustice (Young 2011): injustice that no individual perpetrates but that the collective generates through its structure.

Formal Criterion for Emergence

Proposition 14.1 (Criterion for Emergent Obligation). An obligation O_coll^μ is emergent if and only if there exists no set of individual obligations {O_a^μ}_a∈A such that:

O_coll^μ=∑_a∈A^​ f_a(O_a^μ)

for any set of functions {f_a} . Equivalently, O_coll is emergent if and only if the emergent component E_ν1⋯νn^μ has nonzero contraction with the collective interest:

E_ν1⋯νn^μ I_ν1⁽¹⁾ ⋯ I_νn⁽ⁿ⁾≠0

The criterion is testable: an obligation is emergent when it cannot be attributed to any individual member, no matter how we weight the individual contributions. The irreducibility is structural, not merely epistemic.

Proof. (⇒) Suppose O_collᵘ is emergent: no functions {f_a} satisfy O_collᵘ = Σ_a f_a(O_aᵘ). By the Decomposition Theorem (Prop. 14.3), Cᵘ_{ν₁⋯ν_n} = Σ_k C_(k)ᵘ ⊗ ⋯ + Eᵘ_{ν₁⋯ν_n}. The collective obligation is O_collᵘ = Cᵘ_{ν₁⋯ν_n} I⁽¹⁾_{ν₁}⋯I⁽ⁿ⁾_{ν_n}. If Eᵘ_{ν₁⋯ν_n} I⁽¹⁾_{ν₁}⋯I⁽ⁿ⁾_{ν_n} = 0, then O_collᵘ decomposes into individual and pairwise terms, which has the form Σ_a f_a(O_aᵘ) — contradicting emergence. Hence the contraction is nonzero. (⇐) Suppose the emergent contraction is nonzero. Then O_collᵘ contains a component from E that, by definition of the fully connected term, cannot be expressed as any function of individual obligations alone. Hence O_collᵘ is emergent. □

Proof supplement (orthogonality). The projections P_k (onto the k-agent interaction subspace) are orthogonal because the inclusion-exclusion decomposition partitions tensor components by their interaction order. Formally, if C = Σ_k C_k where C_k involves exactly k-agent interactions, then ⟨C_j, C_k⟩ = 0 for j ≠ k. This follows from Möbius inversion on the subset lattice: the decomposition C_k = Σ_{|S|=k} μ(S) C_S (where μ is the Möbius function) produces orthogonal components because the Möbius function of a Boolean lattice satisfies the orthogonality relation Σ_{S: |S|=k} μ(S)μ(S’) = 0 for |S’| ≠ k.

14.6 Distributed Responsibility

The Responsibility Tensor

When a collective acts — and especially when it acts wrongly — the question of responsibility arises. Who bears the moral weight of the collective’s action?

Definition 14.4 (Responsibility Tensor). The responsibility tensor of a collective A for an outcome ω is a (0,1) -tensor (covector) on the member space:

R_a(ω)∈[0,1]

assigning to each member a a degree of responsibility for ω . The responsibility tensor is a moral covector: it weights the members, just as the interest covector weights the dimensions of moral space.

The responsibility tensor depends on:

Causal contribution. How much did member a’s actions contribute to ω? This is the causal component of responsibility — the counterfactual question of whether ω would have occurred without a’s contribution.

Epistemic position. Did member a know, or should they have known, that ω was a likely consequence? Culpable ignorance (failure to know what one should have known) generates responsibility; blameless ignorance mitigates it.

Role authority. Did member a have the authority and capacity to prevent ω? A supervisor with the power to stop a harmful action bears more responsibility than a subordinate who lacked the power to intervene.

Degree of endorsement. Did member a endorse the collective action, or were they coerced, overruled, or absent? Active endorsement amplifies responsibility; coerced participation mitigates it.

Is Responsibility Conserved?

A fundamental question: does the sum of individual responsibilities equal the collective responsibility?

∑_a∈A^​ R_a(ω)=^?R_collective(ω)

If the answer is yes, responsibility is conserved: it can be fully distributed to individuals, with no remainder. If the answer is no, there is a responsibility gap (if the sum is less) or a responsibility surplus (if the sum is greater).

Claim. In the presence of emergent obligations ( E≠0 ), responsibility is generically not conserved. The gap:

ΔR=R_collective-∑_a∈A^​ R_a

is the collective responsibility remainder — the responsibility that belongs to the collective as a structured entity, not to any individual member. This remainder is the moral analogue of the emergent component E: it exists because the collective has moral agency that exceeds the sum of its members’ agencies.

Example. A committee votes unanimously to implement a policy that turns out to be harmful. Each member bears responsibility for their vote ( R_a>0 for all a). But the committee also bears collective responsibility for the policy as a whole — responsibility that includes the failures of the committee’s deliberative process, its information-gathering procedures, and its institutional design. This collective responsibility may exceed the sum of individual responsibilities: ΔR>0.

Example. A mob commits violence. Each individual participant bears responsibility. But the mob — as a collective agent with emergent dynamics (escalation, deindividuation, collective action) — bears additional responsibility for creating the conditions under which individual members acted as they did. The responsibility surplus ΔR>0 captures the moral content of incitement and collective frenzy.

The Complicity Gradient

In many collective situations, responsibility is not binary (guilty/innocent) but continuously graded. The complicity gradient is the gradient of the responsibility function across the member space:

∇_aR=(∂R)/(∂(member a’s position))

This gradient measures how much more (or less) responsible a member becomes as their position in the collective changes — as they move from periphery to center, from subordinate to authority, from observer to participant.

The complicity gradient is a moral covector on the member space: it weights members by their marginal contribution to collective responsibility. It provides a geometric decomposition of collective responsibility that goes beyond the binary question of “who is responsible?” to the structural question of “how does responsibility vary with position in the collective?”

14.7 The Geometry of Institutional Structure

Institutions as Moral Manifolds

An institution is a collective agent with sufficient structure and persistence to constitute a moral entity in its own right. Corporations, states, universities, hospitals, professional associations, and international bodies are all institutions in this sense.

The geometric framework suggests treating the institution itself as a moral manifold — a space whose points are the possible configurations of the institution (its policies, practices, norms, and distributions of authority), and whose tangent vectors are possible institutional changes.

Definition 14.5 (Institutional Manifold). The institutional manifold M_inst of an institution I is the space of the institution’s possible configurations. A point q∈M_inst specifies:

The institution’s policies and rules

The distribution of authority among members

The decision procedures in force

The institution’s relationship to external stakeholders

The moral manifold M from Chapter 5 and the institutional manifold M_inst are distinct but related. They are coupled by a projection:

π:M_inst→M

that maps each institutional configuration to the moral situation it generates. Different institutional configurations (different policies, different authority distributions) may generate the same moral situation — or different ones. The fiber π^-1(p) over a moral point p consists of all institutional configurations that produce situation p.

Institutional Metrics

The institutional manifold carries its own metric — the institutional metric h_ij — that measures the “cost” of institutional change:

ds_inst²=h_ij dqⁱ dq^j

This metric encodes how difficult it is to change the institution along various dimensions. Some institutional changes are easy (adjusting a minor policy): low cost, small h_ij. Others are hard (restructuring the authority hierarchy): high cost, large h_ij. Some are nearly impossible (changing the institution’s fundamental purpose): the metric approaches infinity.

The institutional metric is related to the moral metric g_μν through the projection π:

g_μν=(∂qi)/(∂xμ)(∂qj)/(∂xν) h_ij

The moral trade-off structure is the pullback of the institutional metric through the projection — the moral costs of change are determined, in part, by the institutional costs of the changes required to effect them.

Institutional Design as Geometric Optimization

If the institutional manifold M_inst is equipped with a metric (the cost structure of change) and a potential (the “moral energy” of each configuration), then institutional design is a problem in Riemannian optimization: find the configuration q^*∈M_inst that minimizes the moral potential, subject to the institutional constraints.

This is the geometric analogue of mechanism design in economics: designing institutions to achieve desired outcomes, given that the institution’s internal structure mediates between individual actions and collective results.

The geometric framework adds precision: the institutional metric determines which changes are feasible, the curvature of M_inst determines whether optimization is well-behaved (convex) or pathological (multiple local minima, saddle points), and the projection π determines how institutional changes translate into moral outcomes.

14.8 Collective Moral Dynamics

Parallel Transport of Collective Obligations

Chapter 10 developed the parallel transport of individual obligations across changing circumstances. The same formalism extends to collective obligations, but with additional structure arising from the collective’s internal dynamics.

When a collective obligation is parallel-transported along a path in moral space, the internal structure of the collective may also change. The obligation to defend citizens is parallel-transported from peacetime to wartime — but the internal structure of the state changes simultaneously (emergency powers, reallocation of authority, mobilization of resources). The transport of the obligation and the transformation of the structure are coupled.

Definition 14.6 (Collective Parallel Transport). The parallel transport of a collective obligation O_coll^μ along a path γ in M is governed by:

∇_γO_coll^μ+Γ_νρ^μ γ^ν O_coll^ρ+Λ_a^μ Σ_a=0

where Γ_νρ^μ is the Levi-Civita connection (as in Chapter 10), Λ_a^μ is the structural coupling — encoding how changes in the structure tensor Σ affect the collective obligation — and Σ_a is the rate of change of the structure.

The additional term Λ_a^μΣ_a is the key difference between individual and collective parallel transport. When the collective’s internal structure is stable ( Σ=0), collective transport reduces to individual transport. When the structure is changing, the collective obligation is pulled in directions determined by the structural coupling Λ — directions that may not align with the individual transport.

Collective Holonomy

The holonomy of a collective obligation around a closed loop may include contributions from the structural coupling:

H_γ^coll=H_γ^ind+∮Λ_a^μ dΣ_a

The second term is the structural holonomy — the contribution of internal reorganization to the collective obligation’s rotation around the loop. A collective that reorganizes as it traverses a moral circuit (e.g., a corporation that restructures during a crisis) may acquire collective holonomy even if the individual-level holonomy is zero.

Example. A university faces a controversy, reorganizes its governance in response, and returns to “normal operations.” The structural holonomy measures how the governance reorganization has permanently altered the university’s moral orientation — even though the explicit policies may look the same as before. The collective obligation has been rotated by the structural journey.

14.9 Collective Agents in the Nine Dimensions

The Collective Scope

Chapter 5 (§5.2) organized the nine moral dimensions via a 3 × 3 matrix: three ethical scopes (Individual, Relational, Collective) × three epistemic modes (What Matters, Who Decides, What We Know). The “Collective” row — Dimensions 3, 8, 6 (Justice/Fairness, Procedural Legitimacy, Societal/Environmental Impact) — is precisely the row that collective agents operate in most naturally.

Individual agents primarily generate obligations along the Individual and Relational rows. Collective agents primarily generate obligations along the Collective row — and they do so with a strength and specificity that individual agents cannot match. The obligation of a state to provide justice (Dimension 3) is stronger, more specific, and more structurally grounded than any individual citizen’s obligation to promote justice. The obligation of a corporation to ensure procedural legitimacy (Dimension 8) is an obligation of the institution, not of any individual officer.

The Collective-Individual Coupling

The relationship between collective and individual obligations is not one-directional. Collective obligations feed back into individual obligations through the structure tensor.

Downward causation. A collective obligation O_coll^μ generates individual obligations through the structure tensor:

O_a^μ=Σ_ab C_ν^-1μ O_coll^ν

Each member acquires an individual obligation that is their “share” of the collective obligation, weighted by their structural position. A firefighter’s individual obligation to rescue (Dimension 1) is generated by the fire department’s collective obligation to protect, mediated by the firefighter’s role in the department’s structure.

Upward emergence. Individual moral actions aggregate (through the collective agency tensor) into collective moral properties:

O_coll^μ=C_ν1⋯νn^μ O_ν1⁽¹⁾ ⋯ O_νn⁽ⁿ⁾

The collective is constituted by individual actions but not reducible to them. The interplay between downward causation and upward emergence creates a feedback loop: collective obligations shape individual ones, which in turn constitute the collective. This circularity is not vicious — it is the structural content of the sociological fact that institutions and individuals mutually constitute each other.

14.10 Emergent Moral Properties

Properties That Exist Only at the Collective Level

Some moral properties have no individual analogue. They exist only at the collective level and resist decomposition.

Justice. Justice is a property of social arrangements — of how benefits and burdens are distributed across a population. An individual can act justly (treating each person fairly in individual interactions), but justice as a structural property — the fairness of the overall distribution — is a collective attribute. In geometric terms, justice is a function on M_inst, not on the individual moral manifold.

Legitimacy. Political legitimacy — the right of an institution to make binding decisions — is a collective property that no individual possesses. A judge has authority by virtue of the judicial system’s legitimacy, not by individual qualities. Legitimacy is an emergent property of the institution’s structure tensor and its relationship to the governed.

Democratic representation. The property of a decision’s being democratically legitimate requires a process involving multiple agents, structured by rules of participation, deliberation, and voting. No individual action can be “democratic” — democracy is irreducibly collective.

The Emergent Properties Criterion

Proposition 14.2 (Criterion for Emergent Moral Property). A moral property P is emergent at the collective level if and only if:

P is well-defined on collective moral states: P:M_inst→R

P is not expressible as a function of individual moral properties alone: P≠f(P₁,P₂,…,P_n) for any function f and any set of individual-level properties {P_a}

P depends on the structure tensor: ∂P/∂Σ_ab≠0 for some a,b

Condition (3) is the geometric marker of emergence: the property depends on the relationships between members, not just on the members themselves.

Proof. Necessity. (1) If P is not well-defined on M_inst, it cannot be a property of collectives at all. (2) If P = f(P₁, …, P_n) for some function f and individual-level properties {P_a}, then P is reducible to individual properties and is not emergent by definition. (3) If ∂P/∂Σ_{ab} = 0 for all a, b, then P does not depend on the structure tensor and hence depends only on properties of the individual members, reducing to case (2). Sufficiency. Suppose all three conditions hold. By (1), P is a well-defined collective property. By (2), P is not expressible as a function of individual properties alone. By (3), the irreducibility is structural — P depends on the relationships between members, not just their individual states. Hence P is emergent at the collective level. □

14.11 AI Systems as Collective Agents

Human-AI Collectives

As AI systems take on morally significant roles, the question of collective agency becomes urgent. A human-AI team — a physician using a diagnostic AI, a judge using a sentencing algorithm, a pilot using an autopilot — is a collective agent whose moral structure requires analysis.

The collective agency tensor of a human-AI team has the form:

C_νHνAI^μ

where ν_H indexes the human’s interest dimensions and ν_AI indexes the AI’s “interest” dimensions (more precisely, the objective-function dimensions that the AI is designed to optimize).

The emergent component E of this tensor is particularly important for AI alignment. If E≠0, the human-AI team generates collective obligations that neither the human nor the AI bears individually. These emergent obligations may be desirable (the team can achieve outcomes neither could alone) or dangerous (the team generates harms that neither would generate alone).

Responsibility Distribution in Human-AI Teams

When a human-AI team produces a harmful outcome, the responsibility tensor R_a(ω) must be defined for both the human member and the AI member. But the AI’s “responsibility” is different in character from the human’s:

The AI has no moral autonomy — its actions are determined by its training, its design, and its deployment. Attributing responsibility to the AI is a shorthand for attributing responsibility to the chain of human decisions that created, trained, deployed, and maintained the AI. The responsibility tensor for the AI component decomposes into:

R_AI(ω)=R_designer+R_trainer+R_deployer+R_operator+R_emergent

where R_emergent is the residual responsibility that cannot be attributed to any individual in the chain. This residual is the collective responsibility of the AI development ecosystem — the responsibility that arises from the interactions between design decisions, training data choices, deployment contexts, and operational parameters.

The Alignment Problem as Emergent Agency

The AI alignment problem can be restated in the language of collective agency: ensuring that the emergent component E of the human-AI collective agency tensor is benign — that the collective obligations and capabilities generated by the human-AI interaction are aligned with human values.

This restatement is productive because it shifts the focus from individual AI behavior (is this AI aligned?) to collective structure (is this human-AI system producing well-aligned collective agency?). An AI that behaves perfectly in isolation may generate harmful emergent properties when embedded in a collective — and conversely, an AI with individual limitations may contribute to well-aligned collective agency when embedded in an appropriate structure.

The structure tensor Σ for human-AI teams encodes the division of authority: which decisions are made by the human, which by the AI, and which are shared. The antisymmetric part Σ_[H,AI] encodes the power differential. The BIP (Chapter 5) constrains the structure: the collective’s moral evaluations must be invariant under admissible re-descriptions, regardless of whether the human or the AI is performing the evaluation.

14.12 Worked Example: The Hospital Ethics Committee

Setup

A hospital ethics committee consists of five members: a physician ( P), a nurse ( N), a patient advocate ( A), an ethicist ( E), and a hospital administrator ( D). They must decide whether to continue life-sustaining treatment for a patient whose family is divided.

The Individual Evaluation Tensors

Each member has an evaluation tensor E_ν^μ(m) reflecting their professional perspective:

The physician P weights welfare (Dimension 1) and epistemic caution (Dimension 9) most heavily:

E_ν^μ(P):strong on E₁¹,E₉¹

The nurse N weights care (Dimension 7) and welfare (Dimension 1):

E_ν^μ(N):strong on E₇⁷,E₇¹

The advocate A weights autonomy (Dimension 4) and rights (Dimension 2):

E_ν^μ(A):strong on E₄⁴,E₂²

The ethicist E weights across all dimensions but especially justice (Dimension 3) and duty (Dimension 2):

E_ν^μ(E):distributed, strongest on E₃³,E₂²

The administrator D weights procedural legitimacy (Dimension 8) and societal impact (Dimension 6):

E_ν^μ(D):strong on E₈⁸,E₆⁶

The Structure Tensor

The committee’s structure tensor encodes:

The physician has clinical authority: Σ_Pm>Σ_mP for all m≠P on clinical questions

The ethicist has ethical authority: Σ_Em>Σ_mE for all m on ethical frameworks

The advocate has veto power on autonomy grounds: Σ_Am spikes when Dimension 4 is at stake

The administrator has institutional authority: Σ_Dm is strong on Dimensions 6 and 8

The nurse has equal voice: Σ_Nm≈Σ_mN — symmetric relationships

The Emergent Component

The committee has obligations that no individual member has:

Procedural obligation. The committee must follow a fair deliberative process — hear all perspectives, give each adequate weight, reach a decision by an appropriate mechanism. This obligation belongs to the committee, not to any individual member.

Documentation obligation. The committee must record its reasoning in a way that can be reviewed. This is an institutional obligation arising from the committee’s role in hospital governance.

Precedent obligation. The committee’s decision establishes precedent for future cases. This obligation — to decide in a way that is generalizable — is emergent: it arises from the committee’s institutional role, not from any member’s individual duty.

These obligations appear in the emergent component E of the collective agency tensor. They cannot be attributed to any individual member, yet they are real obligations that constrain the committee’s action.

The Decision

The committee must contract its collective tensor to a scalar decision: continue treatment, withdraw treatment, or modify the treatment plan. The contraction involves:

Each member contributes their evaluation tensor E_ν^μ(m)

The structure tensor Σ_mm' weights the contributions

The emergent component E adds the collective obligations

The full collective agency tensor C is contracted with the patient’s and family’s interest covectors

The resulting satisfaction scores for each option are compared

The committee decides to modify the treatment plan — a compromise that respects the physician’s clinical judgment, the advocate’s autonomy concerns, and the ethicist’s framework, while satisfying the procedural and documentation obligations that the committee bears collectively.

Responsibility for the Outcome

If the outcome is poor, the responsibility tensor distributes accountability:

R_P=0.30, R_N=0.15, R_A=0.20, R_E=0.15, R_D=0.10

The sum is 0.90. The collective responsibility is 1.00. The remainder ΔR=0.10 is the committee’s responsibility as an institution — responsibility for the quality of the deliberative process, the adequacy of the information considered, and the appropriateness of the decision procedure. This responsibility belongs to the committee, not to any member.

14.13 Summary

Collective moral agency is not a philosophical fiction or a convenient shorthand. It is a structural feature of the moral landscape, encoded in the emergent components of the collective agency tensor. Collectives have obligations that no member bears, responsibilities that no member exhausts, and properties — justice, legitimacy, democratic representation — that exist only at the collective level.

The geometric framework provides the vocabulary for analyzing collective agency with precision: - The collective agency tensor C captures the full structure of collective moral action - Its decomposition into individual and emergent components shows where collective agency exceeds the sum of its parts - The structure tensor Σ encodes the internal organization that mediates between individual and collective - The responsibility tensor distributes accountability while identifying the collective remainder - The institutional manifold provides a geometric setting for institutional design

Chapter 15 turns from the structure of moral evaluation to the moment of choice: contraction — the mathematically necessary, informationally lossy process by which the rich tensor structure of moral reality is compressed into a scalar action-decision.

Technical Appendix

Proposition 14.3 (Decomposition Theorem). Every collective agency tensor C_ν1⋯νn^μ admits a unique decomposition:

C_ν1⋯νn^μ=∑_k=1ⁿ C_(k)^μ⊗δ_ν1⊗⋯⊗δ_νk⊗⋯⊗δ_νn+∑_j<k^​ C_(jk)^μ+⋯+E_ν1⋯νn^μ

where C_(k)^μ is agent k’s individual contribution, C_(jk)^μ is the pairwise interaction between agents j and k, and E is the fully irreducible remainder. This is the moral analogue of the cluster expansion in statistical mechanics.

Proof nEg▫ . Define the terms by inclusion-exclusion. For each subset S ⊆ {1,…,n}, the connected |S|-body term is: C_(S)ᵘ = Tr_{\S} Cᵘ_{ν₁⋯ν_n} − Σ_{S' ⊂ S} C_(S')ᵘ ⊗ ⨂_{k ∉ S'} δ_{ν_k}, where Tr_{\S} traces over all indices not in S. The 1-body terms are the marginals C_(k)ᵘ = Tr_{\{k}} Cᵘ. The 2-body terms are the connected correlations: C_(jk)ᵘ = Tr_{\{j,k}} Cᵘ − C_(j)ᵘ ⊗ δ_{ν_k} − δ_{ν_j} ⊗ C_(k)ᵘ. The remainder E = C_({1,…,n})ᵘ is the fully connected n-body term. Uniqueness: Each m-body term is determined recursively from lower-order terms, so the decomposition is unique. The projections Π_S: C ↦ C_(S) satisfy Π_S Π_{S'} = δ_{SS'}Π_S (orthogonality) and Σ_S Π_S = id (completeness). □

Proposition 14.4 (Responsibility Inequality). For a collective with emergent obligations ( E≠0 ), the collective responsibility for an outcome ω satisfies:

R_collective(ω)≥∑_a∈A^​ R_a(ω)

with equality if and only if ω is attributable entirely to individual actions (the emergent component makes no causal contribution to ω ).

Proof EωΔR≥0E▫ . The collective responsibility decomposes as R_collective(ω) = Σ_{a ∈ A} R_a(ω) + R_E(ω), where R_a(ω) is the responsibility attributable to agent a’s individual contribution and R_E(ω) is the responsibility arising from the emergent component E. The emergent responsibility R_E(ω) ≥ 0 because it is the causal contribution of a structural feature to an outcome, which is non-negative by Definition 14.4 (R_a ∈ [0,1]). Hence R_collective(ω) = Σ_a R_a(ω) + R_E(ω) ≥ Σ_a R_a(ω), with equality iff R_E(ω) = 0 — i.e., iff the emergent component makes no causal contribution to ω. By the Decomposition Theorem (Prop. 14.3), this occurs iff ω is fully attributable to individual actions. □

Proposition 14.5 (Structural Coupling of Parallel Transport). For a collective whose structure tensor Σ changes along a path γ , the parallel transport of the collective obligation deviates from the parallel transport of the individual obligations by:

δO_coll^μ=Λ_a^μ∑_γ^​ dΣ_a+higher order

where Λ_a^μ=∂C^μ/∂Σ_a is the structural coupling. If Σ is constant along γ , the collective transport reduces to the individual transport.

Proof. The collective obligation along a path γ(t) satisfies O_collᵘ(t) = Cᵘ({O_a(t)}, Σ(t)). Differentiating along γ: DO_collᵘ/dt = Σ_a (∂Cᵘ/∂O_aᵛ)(DO_aᵛ/dt) + (∂Cᵘ/∂Σ_a)(dΣ_a/dt). If each individual obligation is parallel-transported (DO_aᵛ/dt = 0), the first sum vanishes, leaving DO_collᵘ/dt = Λ_aᵘ dΣ_a/dt where Λ_aᵘ = ∂Cᵘ/∂Σ_a. Integrating: δO_collᵘ = Λ_aᵘ ∫_γ dΣ_a + O((ΔΣ)²), where the higher-order terms arise from the Taylor expansion of Cᵘ in Σ. If Σ is constant (dΣ_a/dt = 0), then δO_collᵘ = 0. □

❖

The individual is where ethics begins: in the recognition of obligation, the weighing of interests, the exercise of judgment.

The collective is where ethics scales: in the institutions that structure our shared life, the corporations that shape our economies, the states that govern our polities, the professions that organize our expertise.

The transition from individual to collective is not merely additive. It is tensorial — a tensor product that creates new dimensions of moral agency, new obligations that no individual bears, new responsibilities that no individual exhausts.

The mathematics of this transition — the collective agency tensor, the structure tensor, the emergent component, the responsibility covector — provides the vocabulary for analyzing collective morality with the same precision that the tensor hierarchy of Chapter 6 brings to individual morality.

The collective is not a moral fiction.

It is a higher-rank moral entity.

And its geometry matters.