Measure Responsibility

Ethics in a branching universe

A switch sits in front of you. Flipping it prevents a great harm — with 99.9% probability. In the remaining 0.1%, the flip itself causes a smaller harm that would not otherwise have occurred. In a single-world picture this looks like an easy call: flip the switch, and almost certainly no one gets hurt. On an Everettian description, the modeled outcomes occupy branches with different weights rather than waiting for one branch to be selected by collapse. Flip the switch and some descendant histories contain the smaller harm; refuse and much more of the modeled weight contains the great one.

At the moment of decision, the single-world question — which outcome will occur? — is no longer the only useful framing. Different descendant histories can contain different answers. This does not mean every imaginable outcome occurs, or that every policy necessarily produces both good and harm. It means that policy assessment cannot rely only on the one outcome a later observer remembers.

It is not over. It is clarified. But the clarification forces a re-architecture, and this chapter builds it.

The picture doing the work is the Quantum Branching Universe (QBU) — the Everettian reading of quantum mechanics that Volume 2 put to epistemic work: no fundamental collapse, decohering outcomes represented within the quantum state, and squared amplitude supplying Measure. Measure and Credence is the primer; the formal construction of branches and timelines belongs to the physics volume, with the details in The Quantum Branching Universe. This chapter takes that contested interpretation as a condition and asks what follows for responsibility, obligation, and the self. Nothing here makes QBU necessary to the rest of the volume.

From Binary to Distributional

Retrospective ethics often runs on binary questions. Did the harm occur, or not? Did you keep your promise, or break it? In the QBU, those questions still have local answers within a history, but no single answer describes the whole branching model. The patient lived in some modeled descendants and died in others; a conditional promise may come due in some and not in others.

One question that survives is: how much weight does the model assign to each relevant outcome under each available policy? Branch counting is not a well-defined substitute for quantum Measure. But Measure is not a complete moral currency either: it does not supply the value of an outcome, identify whose claims may be traded against whose, or override consent and protection. Responsibility can become distribution-sensitive without becoming aggregate consequentialism. The distribution is evidence evaluated under the agent’s stated commitments and the constraints developed elsewhere in this volume.

If an outcome is unaffected by any available policy, responsibility for that outcome cannot rest on selecting among those policies. Where policies correlate with different weighted distributions, however, deliberation still has work to do. The relevant question is not whether some bad branch exists but which material, foreseeable, and attributable changes the policy makes to the modeled distribution. Measure informs that comparison; it does not manufacture a duty by itself.

The previous chapter made a qualified version of this move for harm: material imposed risk can be a prospective setback when a policy foreseeably and attributably worsens another agent’s position relative to an appropriate baseline. This chapter applies the same distribution-sensitive lens to the acting agent. Physical weights describe part of the shift; classifying it as harm and assigning responsibility still require normative premises.

Navigation Without Steering

The obvious objection: the QBU is deterministic. The universal wavefunction evolves unitarily; no embedded act adds to or subtracts from its total normalized weight. What is there to navigate?

The answer lies in where you are standing. From the God’s-eye view, everything is fixed — but you have never occupied the God’s-eye view and never will. You are an embedded subsystem of the wavefunction, an agent whose deliberation is itself part of the physics, and the question that matters is indexed to your Vantage: the conditioning record for the present decision situation in the branching model. You cannot command counterparts in other branches. What deliberation can represent are alternative policies available at this Vantage and the descendant distributions associated with them. This is the ordinary counterfactual posture of decision theory, not control exercised from outside the physics.

Think of it as deterministic chess. The game is fixed once its state, rules, and players’ policies are fixed, yet deliberation is one of the processes fixing a move. Likewise in the QBU, the agent’s policy is part of the physical model, not an intervention from outside it. “Navigating Measure” is therefore counterfactual decision language: compare the descendant distributions associated with policies the agent can represent and enact. It is not a technical mechanism and not a claim that a local will reaches outside the wavefunction and edits its weights.

Where QBU supplies a physical branch-weight model, Credence should answer to that model while also representing ordinary uncertainty about states, causal structure, and the model itself. Ethics does not inherit a complete calculus directly. A Measure-weighted comparison can reveal that a policy predictably exposes more descendant weight to outcomes the agent disvalues. Whether the policy is permissible still depends on what those outcomes mean for affected agents, what was authorized, and which protections may not be aggregated away.

Obligations at a Vantage

If there are vastly many branches, do you owe a separate duty to each? Does every promise multiply into a trillion promises? No — and seeing why sharpens what obligations attach to.

Obligations do not attach to branch labels. They attach to agents, promises, relationships, and causal histories represented across the model. The multiplication of modeled outcomes need not manufacture new kinds of duty, though Measure can change the exposure at issue. You are evaluating how the futures represented by the model treat the people you owe and the commitments you made, without assuming that every claim can be collapsed into one total.

And the binding is local. This volume’s central device, agent-binding, binds a claim to the agent who holds it; evaluating it requires an explicit normative standard, agent or class, act model, scope, and conflict rules, and the QBU can add a Vantage index to the factual model. When alternative promise records have nonzero amplitude, responsibility still concerns the agent and causal history represented at the specified Vantage rather than one global kept-or-broken label over the multiverse. Measure does not supply the obligation. The promise or other normative premise does, while Vantage and the evidence help determine how it applies.

The Switch, the Promise, and the Rescue

Three cases, worked through, show the machinery running.

The switch. Return to the opening case: the model assigns 99.9% of the relevant Measure to great-harm-averted after the flip and 0.1% to a smaller harm caused by it. If the harms are comparable in kind, the estimates are reliable, no protected claim is being traded away without justification, and no safer policy is available, flipping is warranted. Those conditions matter: the numbers alone cannot authorize sacrificing one person for others. A counterpart who witnesses the smaller harm may still have enacted a reasonable policy while owing recognition or repair to the person harmed. Assessment attaches both to the policy chosen with the information available and to the obligations that survive its outcome; moral luck is constrained, not abolished.

The promise. You promise to donate if event \(E\) occurs. In the QBU, \(E\) and \(\neg E\) both occur. Is the promise kept? The global question has no answer worth wanting, and the vantage-indexed question is easy: in branches where \(E\) happened here, you owe the donation. The version of you on a \(\neg E\) branch owes nothing — the promise’s condition never fired on his history. And the version of you on an \(E\) branch cannot point across the multiverse and say “somewhere I already kept it” or “somewhere it never came due.” Those branches are not his. Promises are context-indexed, not globally duplicated, and no agent gets to settle a local debt with a distant counterpart’s circumstances.

The rescue. You can attempt a rescue that will very probably succeed, with a small modeled weight in which you fail and die. The comparison should include the distribution under each policy, but no branch-weight arithmetic can decide for you how much risk you may accept, what you owe dependents, or whether you caused the danger. The rescue may be justified when the expected protection is large, the risk is informed and voluntary, and the action respects rather than spends other agents’ standing. Measure disciplines the factual comparison; it is not the only unit ethics uses.

You Are Your Choices

So far I have talked as if there is a self that stands at the vantage and steers. A more speculative model inverts this. Most philosophy treats the self as something that has choices — a subject standing apart from the world, acting on it. In a functional QBU account, the self can instead be modeled as a pattern whose decision history helps identify its continuations. This is an interpretation of identity, not a result forced by Everettian physics.

Here is the inversion, stated carefully. Total normalized Measure is invariant — the weights of mutually exclusive, exhaustive event sectors sum to one, and no agent can add to or subtract from that total. What a choice alters in this identity model is not the amount of Measure but the mapping between your ongoing computation and represented continuations. If \(\mathcal{M}\) is the modeled history space and \(S_t\) your cognitive state, the histories consistent with you are

\[Y_t = \{\, \omega \in \mathcal{M} \mid \omega \text{ encodes the decision history of } S_t \,\},\]

and each choice narrows the set:

\[Y_{t+1} = Y_t \cap \{\, \omega : C_{t+1}(\omega) \,\}.\]

On this model, a represented choice narrows \(Y_t\) rather than changing total Measure. Choice can be described as conditionalization within the identity model, while remaining an ordinary physical cause inside a history. The world does not bend to a will outside it; deliberation is part of the world. You are your choices is therefore a proposed account of functional continuity, not a theorem of quantum mechanics.

This offers identity a physical model. Personal continuity may be tracked through correlations among memories, dispositions, bodies, and cognitive processes rather than sameness of an immaterial substance. Coherence can matter to an agent who values integrity, but it is not identical to goodness: a coherent agent can pursue cruelty, and an internally conflicted agent can still act with care. Nor is physical decoherence a synonym for evil. The model can describe continuation without deriving moral valence from physics.

Measure responsibility also qualifies simple outcome-based desert. Responsibility concerns the policy an agent enacted, the foreseeable distribution associated with it, and the local obligations present after an outcome is observed. An agent cannot rewrite the total quantum state from outside it. It can deliberate within its causal position and make a difference to later records through action. Freedom, on this reading, is not unlimited possibility but reasons-responsive participation in the physical process.

The Fatalism Objections

Every version of this picture meets the same three objections. They deserve direct answers.

“But somewhere I still do wrong, no matter what I choose.” That follows only if the relevant alternative appears in the model, and it does not make low-Measure people ghostly or half-real. Yet the existence of a bad descendant history does not exhaust assessment. What matters prospectively is how the available policies differ in foreseeable exposure; what matters locally is what the agent in a history did and now owes. Measure is one morally relevant fact, not the entire moral fact, and an embedded agent’s control is limited by information and circumstance.

“Isn’t this just fatalism?” Not necessarily. Fatalism says deliberation makes no difference; this model treats deliberation as one of the physical processes correlated with different continuations. That does not settle the larger free-will debate. It only rejects the idea that agency must cause one world to exist and every alternative not to. Agency does not override the dynamics; it inhabits them.

“Then how do blame and praise work?” Locally, with attention to policy, knowledge, capacity, causation, and outcome. The reckless driver who arrives home without incident can be blameworthy for materially increasing others’ exposure; the careful rescuer whose attempt fails can remain praiseworthy. But proportionality is not read directly from Measure, and outcomes may create duties of aid or repair even when the antecedent choice was reasonable.

On the Everettian condition, then, morality is not erased. Measure adds a physical weighting to prospective comparisons, while Credence, value, consent, causation, and standing continue to do work that physics cannot. An agent’s task is not to maximize one aggregate called “good,” but to choose with foresight under its avowed commitments while respecting the agency constraints this volume defends. QBU changes the description of uncertainty. It does not relieve ethics of stating its premises.