Measuring the Difference We Make

← Back to Philosophy Essays

Consequentialism asks us to evaluate actions by their consequences. But what exactly should we measure when we ask what an action causes? Everything that happens afterward cannot count as its contribution, since much of that would have happened anyway. Nor can we simply count the consequences we can predict, since an action may have important effects beyond our knowledge. I propose that moral evaluation should begin with a comparison: how does choosing this action rather than an available alternative change outcomes? We should then ask what evidence supports that comparison and how uncertainty could reverse it. My central claim is that an established harm should not be justified by an alleged compensating benefit unless there is evidence connecting the action to that benefit. This is a proposal about justified choice, not a promise that we can calculate the best possible future.

James Lenman’s objection makes the problem urgent. In “Consequentialism and Cluelessness,” he argues that actions can have enormous, inscrutable effects on the distant future. Changes in who survives and reproduces alter who later exists, generating further consequences that we cannot trace. The consequences we can anticipate may be insignificant compared with those we cannot. His challenge does not require indeterminism: a future can be determined while remaining inaccessible to us (Lenman, pp. 344–350).

Consider Lenman’s Richard, a bandit who spares Angie, a pregnant woman who is Hitler’s many-times-great-grandmother. Her survival belongs to the causal history through which Hitler eventually exists. Yet Lenman also considers the possibility that killing her would lead to an even worse future. Preventing Hitler’s existence does not settle the comparison between the alternatives (Lenman, pp. 344–345).

Should Richard kill Angie? I argue that he should spare her. But the explanation cannot simply be that saving a life always produces the best overall consequences. We do not know that. Nor should the explanation be that Richard has no causal connection to later events. If Angie’s survival affects who eventually exists, that connection can be real. The question is what the connection establishes about the value of Richard’s choice.

We must distinguish changing the identity of future people from changing the future for the better. Killing Angie might prevent Hitler’s existence, but it would also alter many other lives and events. Identifying one terrible consequence of sparing her does not establish the comparative value of killing her. The relevant question is not merely whether Hitler would exist without Angie. It is whether the future resulting from her death would be better than the future resulting from her survival. The first question concerns dependence; the second concerns comparative value. An answer to the first does not supply an answer to the second.

A better measure of causal contribution therefore begins with explicit alternatives and outcomes. We should ask what choosing to spare Angie changes relative to killing her, rather than assigning Richard the value of everything that follows her survival. Where outcomes are uncertain, the comparison concerns changes in their probabilities. This approach builds on established work on causal interventions: what matters is what changing an action would do, rather than what is merely associated with it (Pearl, 1995). My proposal concerns how such comparisons should support moral judgment, not the invention of probabilistic causal measurement itself.

For example, suppose an action raises someone’s probability of surviving an immediate danger from twenty percent to ninety percent. Its contribution with respect to that outcome is a seventy-percentage-point increase in survival probability. That contribution remains relevant whether the person ultimately survives or dies. Indeterminism would not eliminate the difference between the probabilities. It would mean that an action can change the chances without guaranteeing an outcome. This measurement is limited: it concerns survival in a specified situation, not the value of the entire future. Making that limitation explicit is part of measuring accurately.

My proposed standard has two parts. First, identify the comparative effects supported by evidence, including plausible ranges rather than false precision. Second, ask what additional effects would have to occur to reverse the judgment those effects support, and what evidence we have for those possibilities. A supported reduction in harm is not automatically decisive, but neither should it lose all significance merely because we can imagine an unfavorable continuation of history.

Richard has a supported reason to spare Angie: doing so preserves her life. He has no corresponding evidence that killing her prevents a greater harm. This asymmetry should govern his choice. He cannot justify killing her merely by pointing out that her descendants might do terrible things. The possibility that killing her produces terrible consequences is also available. Neither imagined history, without evidence favoring it, establishes a comparative advantage. The evidential situation therefore contains a reason to spare her and no adequate consequential justification for killing her.

This does not assume that unknown consequences cancel. They might be enormous and radically unequal. My claim concerns what can justify Richard’s decision, not what the eventual balance must be. An unknown benefit is not a benefit Richard is entitled to invoke as compensation for killing someone. Likewise, an unknown harm does not erase his supported reason to preserve her life. Uncertainty limits what he can claim about the whole future without necessarily eliminating every reason he has to act.

The standard applies to benefits as well as harms. Someone proposing a rescue should count supported risks to other people, rather than treating the intention to save a life as sufficient. Someone proposing to withhold help must likewise support any alleged advantage of doing so. Acting and refraining are both alternatives whose consequences require examination. The principle does not favor inaction simply because it is inaction, and it does not make every preservation of life decisive regardless of the evidence. It requires comparable evidential standards for the competing claims.

Suppose Richard somehow learns that Angie is Hitler’s many-times-great-grandmother. Even this information would not be enough. It would strengthen a claim about the dependence of a particular person’s existence on her survival, but would not establish the net advantage of killing her. Now suppose instead that Richard possesses credible evidence that sparing Angie would initiate a particular catastrophic process, and that an available intervention would prevent it without comparable harms. The deliberative situation would change. He would face competing supported consequences, rather than a known killing set against an imagined benefit. Whether killing would be permissible would require further argument, including consideration of less harmful alternatives. Evidence about comparative consequences, rather than ancestry alone, must do the justificatory work.

One objection is that I have only changed the subject. Consequentialism asks which action produces the best consequences, while I have described which action Richard can justify choosing. That objection identifies a real limit. If actual rightness depends entirely on the total future, my method cannot establish that sparing Angie is objectively right. Richard might choose justifiably while producing a worse history. I accept that possibility. The proposal is a revision to the standard governing choice: an agent should act on adequately supported comparative reasons rather than claim to know which action maximizes the value of all subsequent events. It does not refute a theory that deliberately separates objective rightness from justified deliberation.

Why accept this revision? Because a demand to justify imposing harm should require more than an unsupported possibility of compensation. Otherwise, nearly any harm could be defended by inventing a sufficiently favorable distant outcome. This would not make our decisions more sensitive to consequences. It would make them sensitive to whichever story we found convenient. Requiring evidence constrains the reasons we may offer to the people affected by our decisions. In Richard’s case, Angie’s survival provides a concrete reason, while an unspecified improvement in the distant future provides no adequate justification for killing her.

Another objection concerns ignorance. Could Richard make his choice easier by refusing to investigate possible harms? He could not. The relevant evidence includes what he could reasonably obtain and assess, not merely what he happens to know. When investigation is feasible and might reveal a serious comparative disadvantage, he has reason to investigate. The demands of inquiry also depend on the situation: an urgent rescue does not allow the same investigation as a policy planned over several years. The standard must remain open to correction without requiring an impossible survey of all future history.

Nor does the proposal privilege immediate consequences simply because they are immediate. A delayed effect can be better understood than an immediate one. Evidence that an action initiates a harmful process remains relevant even if the harm appears decades later. Conversely, an event tomorrow may depend on conditions we cannot assess. Temporal distance is not itself a measure of causal importance or evidential strength. What matters is the support for a comparative claim and the extent to which remaining uncertainty threatens it.

Sometimes serious uncertainty will defeat a conclusion. If credible causal explanations disagree about whether an action’s harms outweigh its benefits, we should acknowledge that the evidence does not establish a clear ranking. We may need further inquiry, a more reversible option, or an additional principle for choosing under uncertainty. This is different from allowing any imaginable consequence to defeat a judgment. The distinction is between uncertainty supported by an identifiable causal concern and an unrestricted invitation to speculate. Neither should be misrepresented as certainty, but they need not have the same deliberative force.

Richard should spare Angie. That conclusion does not rest on calculating the goodness of all her descendants or denying her place in Hitler’s ancestry. It rests on a supported reason to preserve her life and the absence of an adequate consequential justification for killing her. Causal dependence alone does not establish a moral debit. To measure the difference an action makes, we must specify the alternative, identify the relevant changes, and state the evidence and uncertainty attached to them. To turn that measurement into justified choice, we need a further commitment: alleged benefits cannot justify established harms merely because the future leaves room for them. This preserves a role for consequences in moral reasoning while refusing to pretend that we know the whole history our actions will help produce.

References

Lenman, James. “Consequentialism and Cluelessness.” Philosophy & Public Affairs 29, no. 4 (2000): 342–370.

Pearl, Judea. “A Causal Calculus for Statistical Research.” Proceedings of the Fifth International Workshop on Artificial Intelligence and Statistics (1995): 430–449.