When Risk Is Harm

Prospective setbacks, baselines, and materiality

Lock a stranger in a room inside a burning hospital, then let the story go well. An outcome-only account records no injury, but many ethical and legal theories already recognize endangerment, lost control, and exposure to serious danger as present setbacks. The problem is therefore not that ethics lacks a word for risk. It is how to identify material, attributable risk increases without treating every interaction as harm.

An exclusively outcome-based framing fails under adversarial pressure in three ways. It can instrumentalize outcomes: two agents can make equally reckless choices and receive divergent verdicts based solely on luck. It can externalize risk: an agent may impose non-trivial danger without accountability if the bad outcome does not materialize. And it can turn ethics into a moral lottery by ignoring the quality of the antecedent decision. Outcome still matters, but ethics in disaster zones, adversarial conflicts, and coordination failures must also assess conduct while the dice are in the air.

The repair is a single move, and this chapter is that move:

Harm includes material setbacks to welfare, functional capacity, or viable options. A foreseeable and attributable worsening of exposure relative to an appropriate baseline can be a present prospective harm, even when the adverse outcome does not occur. Authorization and justification affect wrongfulness, not whether the exposure changed.

Harm at the Moment of Decision

This is the promised extension of the harm concept — the upgrade that What Counts as Harm announced and deferred. The definition given there — material degradation of an agent’s capacity or viable options — is not discarded. What changes is the moment of evaluation. A setback can be realized — you break an arm — or prospective — you wrongfully raise a person’s chance of death from two percent to eight. These are related but not identical impacts: one is an injury, the other an exposure. Both can degrade the agent’s position, and their different magnitudes and remedies must remain visible.

The prospective harm occurs when the act materially worsens the distribution relative to the appropriate baseline. Locking the door exposes the stranger to danger and removes control before any injury lands. Run the four endings: escape without injury, light injury, grave injury, death. The initial endangerment exists in every story, but exposure, injury, and death remain different harms with different magnitudes and remedies. Outcomes still matter to responsibility, repair, and grief.

This limits one form of moral luck without annihilating it. The quality of the decision can be assessed at the time of action using foreseeable risk, while actual consequences remain morally and legally relevant. Choice and outcome answer different questions; neither makes the other disappear.

This also pays a debt. What Counts as Coercion showed that a credible threat reshapes the target’s decision landscape before the threatened setback occurs. Material imposed risk can likewise be a present setback rather than a morally empty possibility. It is coercive only when it is used as a conditional threat to obtain compliance; negligent endangerment, predation, and coercion can overlap, but they are not synonyms.

Ambient and Imposed Risk

Taken absolutely, the definition would be unlivable. Every action in a physical universe carries non-zero risk for somebody: driving past a pedestrian, walking around with a cold you don’t know you have, building a house near other houses. If all non-zero risk counted as coercive harm, every agent would be a predator by default and the ethics would collapse into paralysis.

The account needs both a baseline and a materiality threshold. Ambient risk is a useful label for background hazards not fairly attributable to the act under review, but baselines are normative and institutional choices, not facts delivered by nature. Imposed risk is a foreseeable increase relative to an appropriate counterfactual — often due care, a legitimate status quo, or the trajectory absent the intervention. Causation alone is insufficient: the increase must be material in magnitude, severity, or concentration and not validly authorized or justified.

The demand cannot be zero net added risk; responsible driving still adds some risk. The defensible rule is that agents owe due care and may not impose material, unjustified risk increases beyond the baseline appropriate to the shared practice. Swerving toward a pedestrian and negligently running a power plant cross that line for reasons ordinary participation does not.

Consent as Topology

The distinction helps allocate responsibility but does not settle it mechanically. Risks can be jointly caused, institutionally distributed, unknown, or accepted under imperfect information. Consent authorizes some exposure; it does not make a resulting injury unreal or waive every duty of care. Agents are judged by risks they could reasonably infer from their vantage, subject to evidence about negligence, manipulation, and asymmetric knowledge.

The boundary case that illuminates the distinction is emergency intervention. Pulling someone out of a fire may introduce a small risk of bruising while sharply reducing a grave danger. That can be justified rescue when consent is unavailable and the intervention reasonably protects the person’s agency. Pushing someone into greater danger to clear a path for your friends instead uses that person as an instrument. The rule is not simply “never worsen another’s probability”: ordinary shared practices and medical treatment can impose authorized risk, and emergencies can require uncertain tradeoffs. The questions are whether the change is material, attributable, authorized where possible, protective or instrumental, and consistent with due care.

The Optional QBU Description

Everything so far stands on its own as decision-level ethics. But the framework can say something stronger, because probability itself is not what most ethical theories assume it is.

On the Quantum Branching Universe (QBU) interpretation introduced in Measure and Credence, Measure is the Born weight of a specified event sector, while an agent’s uncertainty remains Credence. Even within that interpretation, “twenty percent chance of death” should not be redescribed as 0.20 literal deaths without specifying the event, state, decomposition, Vantage, and relation between decision weights and persons. The useful point is conditional: alternative policy interventions may be associated with different Measure-weighted exposure to histories in which an agent’s capacity is degraded. That comparison does not globally shift amplitude.

The branching picture can represent distributed consequences, but it does not by itself turn the moral classification “risk is harm” into physics. That classification still depends on the chosen baseline, protected interests, materiality threshold, and account of responsibility. Attempt, negligence, endangerment, and realized injury remain distinct even when a common distributional model represents them.

One clarification carries over and must survive the upgrade. A negative-outcome future is not any future that feels bad. Material welfare setbacks, severe suffering, loss of function, and contraction of viable options can qualify. Offense, disappointment, and heartbreak do not establish prospective harm merely by being unwelcome; the materiality, causal, and baseline tests still apply. Extreme or prolonged suffering need not wait for separately proved agency loss before counting as welfare harm, though its effects on optionality and long-term function often make the case stronger.

Measuring the Damage

Distributional change can be modeled, but no validated scalar measure of prospective harm is established here. One candidate begins with a baseline distribution \(P_1\) and a post-intervention distribution \(P_2\), then computes Kullback–Leibler divergence:

\[D_{\mathrm{KL}}(P_2 \,\|\, P_1) = \sum_{x} P_2(x) \log \frac{P_2(x)}{P_1(x)}\]

KL divergence measures a directional difference between distributions; it is unsigned with respect to welfare, can become infinite when the reference assigns zero probability, and depends on how outcomes are partitioned. It cannot supply the baseline, causal attribution, interpersonal valuation, or valence of the change. At most it is one component of a future metric. Rescue, negligence, and coercion require further classification by authorization, purpose, due care, and where the distribution moved.

The Firewall Against Probabilistic Exploitation

The structural consequence is narrower. A predator, in this framework, is an agent who knowingly and systematically imposes material setbacks on others for instrumental gain. Established predation can justify proportionate defensive coercion under The Boundaries of Force, but it does not erase the predator’s standing wholesale or authorize revenge. Evidence, attribution, necessity, due process where feasible, and continuing limits on force remain load-bearing. What stable multi-agent coexistence requires is the work of The Ethics of Viability; what an agent owes under a branching description is the subject of Measure Responsibility.

Here the narrower verdict suffices. An ethics built for uncertainty should treat material, non-consensual endangerment as a present setback rather than waiting for injury. It must also distinguish ordinary reciprocal risk, justified rescue, due-care practices, negligence, and predatory risk transfer. Agents inevitably affect one another’s futures; the ethical work lies in baselines, authorization, materiality, and justification.

Harm is not exhausted by what someone reports feeling, nor is felt suffering excluded. In this framework it concerns a material setback to welfare, capacity, or viable options, including an attributable worsening of prospective exposure relative to an appropriate explicit baseline. Physics can help model the change; authorization, materiality, causation, duty, and justification determine responsibility and wrongfulness.