The trolley problem and moral dilemmas

Intuition [Beginner]

Imagine you are standing next to a railway track. A runaway trolley is heading down the track toward five people who cannot move out of the way. You are standing next to a lever that can divert the trolley onto a side track. There is one person on the side track. If you pull the lever, the trolley changes course and kills one person instead of five. If you do nothing, the trolley stays on the main track and kills five.

Do you pull the lever?

Most people, when asked, say yes. The reasoning feels straightforward: five deaths are worse than one. You cause one death by acting, but you prevent four. The numbers make the case.

Now consider a second scenario. The same runaway trolley is heading toward the same five people. This time there is no side track and no lever. You are standing on a footbridge over the track, next to a very large man. The only way to stop the trolley and save the five is to push the large man off the bridge and onto the track. His body would stop the trolley. He would die; the five would live.

Do you push him?

Most people, when asked, say no. The numbers are the same — one dies, five live. But something feels different. Pushing a person to his death feels wrong in a way that pulling a lever does not, even though the body count is identical.

This pair of scenarios is the trolley problem, and the gap between our responses to the two cases is one of the most productive puzzles in moral philosophy. The puzzle is not about trolleys. It is about what makes killing wrong, whether outcomes are all that matter, and whether our moral intuitions are tracking something real or just reflecting psychological habits that evolved for different purposes entirely.

The original version was invented by the philosopher Philippa Foot in 1967, in a paper about the doctrine of double effect — a principle from Catholic moral theology that distinguishes between harm you intend and harm you foresee but do not intend ^{[Foot 1967]}. Foot's question was: can that distinction do the work its defenders claim? The trolley case was a test. Judith Jarvis Thomson extended the problem in 1976 with the fat man variant (the footbridge case) and later with several other variants, each designed to isolate a specific moral factor ^{[Thomson 1976]}.

Why care about a thought experiment involving trolleys? Three reasons.

First, the trolley problem exposes a tension between two families of moral theory that have shaped Western ethics since the eighteenth century. Consequentialism — the family that includes utilitarianism — holds that the rightness of an action is determined by its consequences. If pulling the lever saves five at the cost of one, it is the right thing to do.

Deontology — the family that includes Kant's ethics — holds that some actions are wrong regardless of their consequences, because they violate duties, rights, or respect for persons. Pushing the fat man off the bridge might violate a right not to be used as a physical obstacle, even if the body count comes out favourably. The trolley problem is the cleanest laboratory for testing which framework captures our actual moral commitments.

Second, the problem has generated a remarkably rich taxonomy of variants — the loop, the switch, the transplant surgeon, the cave explorer — each of which prises apart a different moral factor. This is not navel-gazing. The variants map directly onto real decisions in medicine (triage, organ allocation), law (self-defence, necessity), and engineering (autonomous vehicle programming).

Third, since the early 2000s, the trolley problem has become a centrepiece of empirical moral psychology. Joshua Greene and Jonathan Haidt, among others, have used brain imaging and behavioural experiments to investigate why people give the answers they give — and their findings suggest that the intuition gap between the lever case and the bridge case is driven by emotional responses to personal, physical harm, not by principled moral reasoning. This raises the question of whether our moral intuitions are a reliable guide to moral truth, or whether they are cognitive artefacts that philosophy should sometimes override.

The trolley problem is, in short, a small door that opens onto a large room: the room where consequences, rights, intentions, emotions, and real-world decision-making all collide.

Visual [Beginner]

The two canonical trolley scenarios can be diagrammed side by side. On the left, the switch (diversion) case: a main track with five figures, a side track with one figure, a lever at the junction. An arrow shows the trolley approaching from the left. The decision point is marked: pull the lever, and the trolley diverts; do nothing, and the trolley continues to the five.

On the right, the footbridge (fat man) case: a single track with five figures, a footbridge overhead, a large figure on the bridge next to a smaller one (you). An arrow shows the trolley approaching. The decision point is: push the large figure onto the track, or do nothing.

The visual makes the structural similarity vivid: same numbers, same outcome profile (one dies, five live if you act), but different mechanism. In the switch case, you redirect an existing threat. In the bridge case, you create a new threat by using a person as a tool. Whether this difference is morally significant is the central question.

Worked example [Beginner]

Let us walk through the moral arithmetic and the intuitive verdicts for the two canonical cases, making explicit what most people feel but do not articulate.

Case 1: The switch (diversion).

Initial state: five people on the main track, one person on the side track, trolley approaching on the main track.

Your options:

• (A) Do nothing. Trolley hits five. Outcome: five dead, one alive.
• (B) Pull the lever. Trolley diverts to side track, hits one. Outcome: one dead, five alive.

The consequentialist calculus is straightforward: (B) produces fewer deaths, so (B) is better. Most people agree: in surveys conducted across cultures, roughly 80–90% of respondents say it is permissible to pull the lever ^{[Greene 2001]}.

The deontologist need not disagree. The person on the side track is harmed, but (the standard argument goes) you do not intend their death. You intend to redirect the threat away from the five. The one person's death is a foreseen side-effect of your action, not its aim. This distinction — between intended and merely foreseen harm — is the Doctrine of Double Effect (DDE), and it is the principle Foot was testing when she invented the problem ^{[Foot 1967]}.

Case 2: The footbridge (fat man).

Initial state: five people on the track, one large man on the bridge next to you, trolley approaching.

Your options:

• (A) Do nothing. Trolley hits five. Outcome: five dead.
• (B) Push the large man. His body stops the trolley. Outcome: one dead, five alive.

The consequentialist calculus is identical: (B) saves more lives. But most people — again, roughly 80–90% in surveys — say it is not permissible to push the man. The DDE explanation is that in the bridge case, the man's death is not a side-effect; it is part of the plan. You are using his body as a trolley-stopper. His death is intended — not as an end, but as a means. The DDE says intending harm as a means is wrong even when the consequences are good.

The puzzle, then, is why the two cases feel different when the numbers are the same. The DDE offers one answer. Thomson's rights-based analysis offers another: the man on the bridge has a right not to be used as a physical instrument to save others, a right that the person on the side track does not exercise in the same way because the threat to that person was already in motion before you acted. A third answer, from empirical psychology, is that physically pushing a person triggers a stronger emotional aversion than pulling a lever — and that this emotional difference, not a principled moral distinction, drives the intuition gap ^{[Greene 2001]}.

Each of these answers has consequences for what we think morality is and how we should make difficult decisions. The rest of this unit unpacks them.

Check your understanding [Beginner]

Formal definition [Intermediate+]

The trolley problem is best understood as a family of cases sharing a common structure. Each case presents an agent with a choice between two or more actions, where at least one action produces better aggregate consequences but involves harm to an innocent person. The cases differ in the moral factors they introduce: the relationship between agent and victim, the mechanism of harm, the intention with which harm is brought about, and the rights at stake.

The case space

A trolley case can be specified by a tuple $(S,A,O,M,I,R)$ where:

• $S$ is the situation: the physical setup (tracks, people, threat trajectory).
• $A$ is the action set: the options available to the agent (pull lever, push man, do nothing).
• $O$ is the outcome function: the mapping from actions to consequences (who lives, who dies).
• $M$ is the mechanism of harm: how the victim is harmed (redirected threat, direct physical force, removal of resources, etc.).
• $I$ is the intentional structure: whether the victim's harm is intended as an end, intended as a means, or merely foreseen as a side-effect.
• $R$ is the rights profile: what rights the victim holds that might be violated by the action.

The canonical cases occupy different positions in this space:

Case	Mechanism ($M$)	Intention ($I$)	Rights ($R$)	Consequentialist verdict	DDE verdict	Popular verdict
Switch	Divert existing threat	Foreseen, not intended	Threat already heading toward victim's space	Permissible	Permissible	Permissible (~85%)
Footbridge	Use person as physical instrument	Intended as means	Right not to be used as a means	Permissible	Impermissible	Impermissible (~85%)
Loop	Divert threat through victim back toward five	Ambiguous: victim's body needed for stopping	Right not to be instrumentalised?	Permissible	Disputed	Mixed
Transplant	Kill to harvest organs	Intended as means	Right to life, bodily integrity	Permissible	Impermissible	Impermissible (~95%)
Cave	Create suffocation to escape	Foreseen, not intended?	Right of trapped caver?	Permissible	Disputed	Mixed

The Doctrine of Double Effect (DDE)

The DDE originates in Aquinas's discussion of self-defence in Summa Theologica II-II, q. 64, a. 7 ^[Aquinas]. The principle holds that an action with both good and bad effects may be permissible provided four conditions are met:

1. The act itself is morally neutral or good. The agent's action must not be intrinsically wrong. Pulling a lever to redirect a trolley is, considered in itself, a neutral physical act. Pushing a person off a bridge is, considered in itself, a physical assault — and on some readings, intrinsically wrong.

2. The agent intends the good effect, not the bad. The bad effect (the one person's death) may be foreseen but must not be intended — neither as an end nor as a means.

3. The bad effect is not the means to the good effect. The good outcome must not be achieved through the bad outcome. In the switch case, the five are saved by the redirection of the threat, not by the one person's death. In the footbridge case, the five are saved by the fat man's death — his body stops the trolley.

4. Proportionality. The good effect must be sufficiently weighty to justify the bad effect. Saving five lives at the cost of one typically satisfies proportionality; saving minor property at the cost of a life does not.

The DDE explains the switch-vs-bridge intuition gap by locating the moral difference in conditions 2 and 3. In the switch case, the one person's death is a foreseen side-effect of redirecting the threat. In the bridge case, the fat man's death is the means by which the five are saved.

The DDE has been attacked from both directions. Consequentialists argue that condition 2 rests on an untenable distinction between intending and foreseeing — if you know your action will kill someone and you act anyway, they say, you intend their death in any sense that matters. Deontologists sympathetic to the DDE respond that the distinction tracks something real about the structure of agency: the agent's plan in the switch case does not require the one person to die, while the agent's plan in the bridge case does ^{[Quinn 1989]}.

Thomson's rights-based analysis

Thomson's 1976 paper ^{[Thomson 1976]} offered a different framework. Instead of focusing on intention, Thomson argued that the moral difference between the cases is grounded in rights. The person on the side track in the switch case has no right that the trolley not be on that track — the trolley was heading somewhere else, and you have redirected it. The fat man on the bridge has a right not to be shoved to his death. The difference is not about what you intend but about what rights you violate.

Thomson's analysis introduces a powerful idea: the moral significance of who has a right to what, independent of the agent's psychological state. On this view, even if you do not intend the fat man's death (say you genuinely regret it), you still violate his right not to be used as a trolley-stopper. The violation of rights is an objective feature of the action, not a feature of your intentions.

Thomson later revised her view in 1985 ^{[Thomson 1985]}. She argued that even the switch case may be impermissible, because the person on the side track has a right not to be killed by a trolley diverted onto their track, and your action causes that death. This revision — sometimes called Thomson's "recantation" — is significant because it narrows the gap between the two cases from the other direction: instead of explaining why the bridge case is wrong, Thomson now questions whether the switch case is permissible. The revision does not deny the intuitive difference; it questions whether our intuitions are tracking morally relevant factors.

Counterexamples to common slips

• "The DDE says the end justifies the means." This is the opposite of what the DDE says. The DDE imposes constraints on means: even when the end is good, certain means (intending harm, using harm as a means) are forbidden. The whole point of the doctrine is to limit consequentialist reasoning.

• "Deontology always forbids killing." Deontology forbids treating persons as mere means, but many deontologists permit killing in self-defence and in certain trolley-switch cases. The constraint is not "never kill" but "never violate rights" or "never intend harm as a means" — and these constraints may or may not rule out a particular killing, depending on the structure of the case.

• "The trolley problem is unrealistic and therefore irrelevant." Thought experiments in philosophy are not predictions; they are diagnostic tools. The trolley problem isolates specific moral factors by stripping away confounding variables. The relevance of the factors it isolates is confirmed by their appearance in realistic cases (triage, self-defence, autonomous vehicle programming) — not by the literal plausibility of runaway trolleys.

• "Consequentialism just says 'maximise happiness'." Consequentialism is the family of views that the right action is the one that produces the best consequences, where "best" is defined by some value function (total happiness, preference satisfaction, lives saved, etc.). Utilitarianism is one member of this family, but not the only one. In the trolley context, the relevant value function is usually lives saved — a form of consequentialism that need not commit to the full utilitarian programme.

Key theorem with proof — the Thomson dialectic and the loop variant [Intermediate+]

The most important formal result in the trolley-problem literature is not a mathematical theorem but an argument-reconstruction: the Thomson dialectic, which uses a sequence of variants to pressure-test the DDE and show that the intuitive difference between the switch and bridge cases cannot be fully explained by the intended-vs-foreseen distinction alone.

Setup. Thomson (1976, 1985) ^{[Thomson 1976] [Thomson 1985]} introduced several intermediate variants between the switch case and the bridge case. The most important is the loop variant.

The loop case. The trolley is heading toward five people. You can pull a lever to divert it onto a side track. The side track loops back and re-joins the main track after the five people. On the looping section of the side track, there is one person. If you pull the lever, the trolley enters the loop, hits the one person, and is stopped (or slowed enough) by the impact. The five are saved. If you do nothing, the trolley continues and kills the five.

The argument. The loop variant is designed to create a problem for the DDE.

Step 1. In the standard switch case, the DDE says: the one person's death is a foreseen but unintended side-effect. The means of saving the five is the redirection of the threat, not the one person's death. Permissible.

Step 2. In the loop case, the one person's body is needed to stop the trolley. If the one person were not on the loop, the trolley would loop back and kill the five anyway. So the five are saved by the one person's death, in the same causal sense as in the bridge case. The means of saving the five is the one person's body stopping the trolley.

Step 3. By the DDE's condition 3 (the bad effect must not be the means to the good effect), the loop case should be impermissible — just like the bridge case.

Step 4. But most people's intuition about the loop case is that it is permissible — or at least closer to permissible than the bridge case.

Step 5. Contradiction: the DDE predicts impermissibility for the loop case, but intuition says permissibility.

This is the key tension. The loop variant shows that the DDE's intended-vs-foreseen distinction does not fully capture the moral difference between the switch case and the bridge case. Something else must be doing the work — either the physical directness of the harm (pushing vs pulling), the spatial proximity of the agent to the victim, or some other factor that the DDE does not track.

Thomson's response was to shift from the DDE to a rights-based framework. On her analysis, what makes the bridge case wrong is not that the agent intends the fat man's death (though they do) but that the agent violates the fat man's right not to be used as a physical instrument. In the loop case, the one person is on the track for reasons unrelated to the agent's plan — the agent diverts a threat, and the person happens to be in the way. The right not to be used is not violated in the same way, even though the causal structure is similar.

Kamm (2007) ^{[Kamm 2007]} refined this further with the Doctrine of Triple Effect: an action can have effects that are (1) intended as ends, (2) intended as means, or (3) neither intended as ends nor as means but needed for the plan to work. The loop case involves a type-3 effect: the one person's death is not what the agent aims at, and it is not the tool the agent uses, but it is a feature of the situation that must obtain for the plan to succeed. Kamm argues that type-3 effects occupy a different moral status from type-2 effects (means), and this explains the loop-vs-bridge intuition gap.

The Thomson dialectic is a model of analytic philosophical method: construct a sequence of cases, identify the principle that explains the intuitive verdicts, find a variant that the principle gets wrong, and use the mismatch to refine the principle or replace it. The argument is not "proved" in the mathematical sense — it is pressure-tested by cases, and its strength is measured by how many intuitive verdicts it correctly predicts and how few it gets wrong.

Exercises [Intermediate+]

Dual-process moral psychology [Master]

The most consequential development in the trolley-problem literature since Thomson is the introduction of empirical methods from cognitive science. Joshua Greene and colleagues, beginning with their 2001 Science paper ^{[Greene 2001]}, used functional magnetic resonance imaging (fMRI) to investigate the neural correlates of moral judgement in trolley-type dilemmas. Their findings have reshaped the debate by raising a question that philosophy alone could not answer: are our moral intuitions about the trolley problem tracking moral truth, or are they by-products of emotional responses that evolved for different purposes?

Greene's fMRI studies

Greene et al. (2001) presented subjects with a battery of moral dilemmas classified as either impersonal (like the switch case, where harm is mediated by a physical mechanism at a distance) or personal (like the bridge case, where the agent must directly and physically harm another person). The key findings:

1. Personal dilemmas activated brain regions associated with emotion. The medial prefrontal cortex, the posterior cingulate, and the amygdala — areas implicated in emotional processing — showed significantly greater activation when subjects considered personal moral dilemmas than when they considered impersonal ones.

2. Impersonal dilemmas activated brain regions associated with cognitive calculation. The dorsolateral prefrontal cortex and parietal areas — regions involved in working memory and abstract reasoning — were more active for impersonal dilemmas.

3. Individuals who gave utilitarian responses to personal dilemmas showed longer reaction times and greater activation in cognitive-control regions. Overriding the emotional aversion to personal harm appeared to require effortful cognitive processing.

4. The emotional response was rapid and automatic; the utilitarian override was slow and deliberate. This pattern matches Kahneman's System 1 / System 2 framework ^{[Kahneman 2011]}: fast, automatic, emotional processing (System 1) generates a deontological intuition against personal harm; slow, deliberate, calculative processing (System 2) generates the consequentialist verdict.

Greene's interpretation — developed at length in Moral Tribes (2013) ^{[Greene 2013]} — is that the intuition gap between the switch case and the bridge case is not tracking a morally relevant difference. The switch case feels permissible because it does not trigger the emotional aversion to personal harm. The bridge case feels impermissible because it does trigger that aversion. But the emotional response, Greene argues, evolved to regulate face-to-face violence in small social groups — not to adjudicate abstract moral questions about levers and trolleys. The deontological intuition against personal harm is a psychological artefact, not a moral insight.

This is a bold claim with significant implications. If Greene is right, then the entire DDE-based framework for explaining the switch-vs-bridge gap is an exercise in post-hoc rationalisation: philosophers have constructed elaborate principles to justify an intuition that is, at bottom, a cognitive bias.

Critiques of the dual-process programme

The Greene programme has attracted sustained criticism from both philosophers and psychologists.

The normativity objection (Kahane, 2012). The fact that an intuition has an evolutionary or emotional origin does not by itself show that the intuition is unreliable. Our capacity for mathematical reasoning has evolutionary origins too, but this does not discredit mathematics. The normativity of moral intuitions depends on whether they track moral truth, not on how they arose. Greene's argument requires an additional premise — that emotional intuitions are less reliable than cognitive calculations in the moral domain — and this premise is itself a normative claim that cannot be established by brain imaging alone.

The case-classification objection (Borg et al., 2006; McGuire et al., 2009). The personal/impersonal distinction that Greene uses to classify dilemmas is not the only factor that differs between the switch case and the bridge case. The cases also differ in: the agent's physical proximity to the victim; the presence of a physical barrier (the bridge railing); the degree of force required; the social relationship between agent and victim (the fat man is a bystander, not a track worker who accepted occupational risk). Any of these factors, rather than the personal/impersonal distinction per se, could drive the neural activation pattern. The fMRI evidence does not uniquely support Greene's interpretation.

The expertise objection. Professional philosophers, who have extensive training in reasoning about moral dilemmas, show different patterns of moral judgement than non-philosophers. Some studies suggest that philosophical training increases utilitarian responding to personal dilemmas — which is consistent with Greene's claim that cognitive processing overrides emotional intuition. But it is also consistent with the claim that philosophical training improves moral judgement by enabling agents to see through emotional biases. The direction of normative implication depends on whether you think the emotional intuition or the cognitive override is the better guide to moral truth — and brain imaging cannot settle this question.

The cultural variation objection. Cross-cultural studies of trolley-problem responses show significant variation. Some collectivist cultures show higher rates of utilitarian responding to personal dilemmas than individualist Western cultures. If the deontological intuition against personal harm were a universal moral insight, we would expect less variation. The variation is consistent with Greene's claim that the intuition is a cultural or evolved bias. But it is also consistent with the claim that different cultures have different but reasonable moral frameworks — a meta-ethical claim that Greene's neuroscience does not address.

Greene's "deep pragmatism" and the meta-ethical upshot

Greene's later work (Moral Tribes, 2013) ^{[Greene 2013]} argues that the dual-process findings support a specific meta-ethical position: deep pragmatism, or what Greene calls "metamorality." The idea is that when moral disagreements arise between groups with different moral frameworks (the "tragedy of commonsense morality"), the only adjudicating principle available is a consequentialist one — because consequentialism is the framework that abstracts away from parochial intuitions and asks only about outcomes. Deontological constraints, on this view, are useful within groups (they regulate social behaviour efficiently) but cannot resolve inter-group disputes, because different groups have different deontological intuitions.

The argument is elegant but contested. Critics point out that Greene's "deep pragmatism" is itself a substantive moral commitment — it privileges one meta-ethical framework (consequentialism) over others — and that this privilege cannot be derived from empirical data alone. The jump from "deontological intuitions have emotional origins" to "consequentialism is the correct metamorality" requires a normative bridge that brain imaging does not provide.

Position-mapping the frameworks

At master tier it is useful to map the competing frameworks onto their commitments:

Framework	Source of moral permissibility	Handles loop case?	Handles bridge case?	Empirical grounding
Pure consequentialism	Aggregate outcomes	Permissible (best outcome)	Permissible (best outcome)	Predicts no switch-vs-bridge gap
DDE (intention-based)	Intentional structure of agency	Struggles (means/ends ambiguity)	Impermissible (means)	Partially supported by intention-attribution psychology
Thomson (rights-based)	Rights violations	Permissible (right not violated)	Impermissible (right violated)	Tracks legal intuitions about battery/assault
Kamm (triple effect)	Means vs. needed distinction	Permissible (merely needed)	Impermissible (means)	No direct empirical test
Greene (dual process)	Cognitive override of emotion	Gap is artefact of emotion	Gap is artefact of emotion	fMRI + behavioural data

No single framework correctly predicts all intuitive verdicts without strain. This is part of the lesson of the trolley problem: our moral intuitions are not generated by a single principle, and any theory that predicts them perfectly will need to accommodate multiple moral factors simultaneously.

The trolley problem in the real world [Master]

The trolley problem is sometimes dismissed as a philosopher's parlour game with no practical import. This dismissal is mistaken. The same moral factors that the trolley problem isolates — the distinction between doing and allowing, the role of intention, the weight of rights, the tension between aggregate outcomes and individual protections — arise in real-world decision-making across multiple domains.

Autonomous vehicles

The most direct application is in the programming of autonomous vehicles (AVs). An AV approaching an unavoidable collision must decide how to allocate harm. Should it swerve into a barrier, killing its passenger, to avoid hitting pedestrians? Should it prefer fewer deaths over more, regardless of who the victims are? Should it protect its occupants at all costs?

Bonnefon, Shariff, and Rahwan (2016) ^{[Bonnefon 2016]} surveyed public attitudes toward AV ethics and found a "social dilemma": respondents strongly agreed that AVs should be programmed to minimise total casualties (the utilitarian response), but also reported that they would not purchase a vehicle programmed to sacrifice its own passenger for the greater good. The result is a collective action problem: everyone benefits if all AVs are utilitarian, but each individual has an incentive to buy a self-protective vehicle.

This is the trolley problem with engineering stakes. The DDE, rights-based, and consequentialist frameworks give different programming recommendations. A DDE-informed programme might permit redirecting harm (swerve to avoid pedestrians) but forbid using a specific person as a means (deliberately crash into a particular pedestrian to stop the vehicle). A rights-based programme might hold that passengers have a right to expect the vehicle to protect them, and that this right constrains what the vehicle may do to others. A consequentialist programme would simply minimise expected casualties.

The practical difficulty is that the vehicle's decision algorithm must be specified in advance, before the particular situation arises. This forces a choice between moral frameworks at the design stage — exactly the kind of choice the trolley problem is designed to expose. MIT's Moral Machine project (Awad et al., 2018) collected millions of trolley-problem-style judgements from respondents worldwide and found significant cross-cultural variation in preferences, raising the question of whether there is a single correct answer or whether AV ethics is genuinely context-dependent.

Medical triage

Emergency medicine and disaster response involve trolley-problem reasoning under a different description. A triage officer with limited resources must decide who receives treatment and who does not. The standard triage protocol — treat those with the highest probability of survival given immediate intervention — is a form of consequentialism: maximise lives saved. But it can conflict with deontological intuitions about the equal worth of each patient, the duty not to abandon the critically injured, and the wrongness of "playing God" with individual lives.

The COVID-19 pandemic brought this tension into sharp focus. Hospitals facing ventilator shortages had to decide whether to allocate ventilators to patients with the best prognosis (consequentialist triage) or to use a first-come-first-served or lottery system (deontological equality). The DDE-based reasoning appeared in debates about whether withdrawing a ventilator from one patient to give it to another is morally different from never giving it to the first patient — a distinction that mirrors the doing/allowing distinction in the trolley literature.

Military ethics and the doctrine of double effect

The DDE was originally developed in the context of just war theory, and its most sustained real-world application is in the ethics of military action. The principle of proportionality in international humanitarian law — which holds that civilian casualties must be proportionate to the military advantage gained — is a direct descendant of the DDE. The strategic bomber who destroys a munitions factory while foreseeing civilian deaths invokes the DDE; the terror bomber who targets civilians to break morale violates it.

Michael Walzer's Just and Unjust Wars ^{[Walzer 2006]} develops this application at length. Walzer defends a version of the DDE but adds a "due care" requirement: soldiers must accept some additional risk to themselves in order to minimise civilian harm, even when the harm is unintended. This is a practical refinement that the abstract trolley cases do not capture — but it shows how the moral factors the trolley problem isolates (intention, means, proportionality, rights) scale up to life-and-death decisions in the field.

The doctrine of double effect is also invoked in debates about targeted killing, economic sanctions (which foresee but do not intend civilian suffering), and the use of human shields by adversaries (which attempts to weaponise the attacker's deontological constraints).

Connections [Master]

• Introductory ethics / moral theory 20.02.01 pending (pending) is the foundational prerequisite. The trolley problem presupposes familiarity with the consequentialism/deontology/virtue-ethics taxonomy and with the basic structure of normative argument. This unit draws on and complicates the framework that 20.02.01 introduces.

• Logic and formal methods [20.01.NN] (pending) is where a formal deontic logic encoding of the DDE or Thomson's rights-based analysis could live. Argument-reconstruction tooling — the kind of premise-conclusion formalisation practised in the exercises — would also be at home there.

• Philosophy of mind 20.06.01 pending (pending) connects via the dual-process moral psychology programme. Greene's claim that deontological intuitions are emotional artefacts raises questions about the reliability of introspection, the nature of moral reasoning, and the relationship between affective and cognitive processes — all central topics in philosophy of mind and cognitive science.

• Philosophy of science / scientific realism 20.07.01 pending (pending) connects via the status of moral intuitions as evidence. Greene's empirical programme raises the question of whether moral philosophy should be constrained by empirical findings about how people actually make moral judgements — a form of methodological naturalism that parallels debates about naturalism in philosophy of science.

• Decision theory and expected utility [§02.NN] connect through the autonomous-vehicle and triage applications, where the moral framework must be formalised as a decision rule. The gap between normative decision theory and deontological constraint is a live research question in both ethics and economics.

Cross-domain to political philosophy: the tension between individual rights and aggregate welfare that the trolley problem dramatises is the same tension that structures debates about distributive justice, taxation, and the limits of state power. The trolley problem is, in this sense, a microcosm of political philosophy.

Historical & philosophical context [Master]

The trolley problem has its immediate origin in Philippa Foot's 1967 paper "The Problem of Abortion and the Doctrine of Double Effect" ^{[Foot 1967]}, published in the Oxford Review. Foot's primary concern was not trolleys but the moral status of abortion, and the trolley case was one of several thought experiments she used to test the DDE. Foot observed that the DDE — which had been part of Catholic moral theology since Aquinas — could be deployed to distinguish between cases where harm was a foreseen side-effect of a permissible action and cases where harm was directly intended. The trolley case was designed to be a "clean" example of the former: the agent diverts a threat, and the one person's death is a side-effect.

Judith Jarvis Thomson's 1976 paper "Killing, Letting Die, and the Trolley Problem" ^{[Thomson 1976]} transformed Foot's example from a supporting illustration into the centrepiece of a systematic investigation. Thomson introduced the fat man variant (the bridge case), the distinction between rights violations and mere harm, and the argument that the moral difference between the switch case and the bridge case is best explained by rights, not intentions. Her paper established the trolley problem as a free-standing topic in normative ethics.

Thomson's 1985 paper "The Trolley Problem" ^{[Thomson 1985]} revised her earlier view. She now argued that even the switch case might be impermissible, because the person on the side track has a right that the trolley not be diverted onto them. This revision — sometimes called the "self-help" argument, because Thomson suggested that the five might have a right to redirect the trolley but no third party does — generated a new round of variants and counter-arguments.

Warren Quinn's 1989 paper "Actions, Intentions, and Consequences" ^{[Quinn 1989]} refined the DDE by distinguishing between direct and indirect harmful agency. Quinn introduced the "strategic bomber" vs "terror bomber" cases as a clearer illustration of the intended-vs-foreseen distinction than the trolley variants. His analysis defended the DDE against the charge that the intended/foreseen distinction is merely verbal.

Frances Kamm's Intricate Ethics (2007) ^{[Kamm 2007]} is the most detailed deontological treatment of the trolley problem. Kamm introduced the Doctrine of Triple Effect, the distinction between means and merely-needed effects, and an exhaustive taxonomy of trolley variants designed to isolate individual moral factors. Her work represents the state of the art in deontological response to the problem.

Joshua Greene's 2001 Science paper ^{[Greene 2001]} with Sommerville, Nystrom, Darley, and Cohen inaugurated the empirical turn. By showing that personal and impersonal moral dilemmas activate different brain regions, Greene brought the trolley problem into contact with cognitive neuroscience and raised the stakes of the philosophical debate: if our intuitions are driven by emotional responses that evolved for face-to-face social regulation, then perhaps the DDE-based framework that builds on those intuitions is systematically unreliable. Greene's Moral Tribes (2013) ^{[Greene 2013]} extended this argument into a full meta-ethical proposal.

David Edmonds's Would You Kill the Fat Man? (2013) ^{[Edmonds 2013]} is the standard accessible survey of the trolley problem's history, variants, and significance.

The historical arc of the trolley problem — from Foot's theological test case through Thomson's rights-based analysis through Greene's empirical programme — illustrates a distinctive feature of analytic moral philosophy: the use of carefully constructed thought experiments to pressure-test normative principles, combined with a willingness to revise or abandon principles when they fail to accommodate intuitive verdicts. The trolley problem is not a puzzle to be solved but a diagnostic instrument for understanding what we think about harm, rights, and the relationship between intentions and outcomes.

Bibliography [Master]

Foundational and historical:

• Aquinas, T. — Summa Theologica II-II, q. 64, a. 7 (c. 1270). [Need to source.]
• Foot, P. — "The Problem of Abortion and the Doctrine of Double Effect", Oxford Review 5, 5–15 (1967). [Need to source.]
• Thomson, J. J. — "Killing, Letting Die, and the Trolley Problem", The Monist 59, 204–217 (1976). [Need to source.]
• Thomson, J. J. — "The Trolley Problem", Yale Law Journal 94, 1395–1415 (1985). [Need to source.]
• Kant, I. — Groundwork of the Metaphysics of Morals (1785); trans. H. J. Paton (Harper & Row, 1964). [Need to source.]
• Mill, J. S. — Utilitarianism (1861); ed. Robert Sher (Hackett, 2001). [Need to source.]

Deontological refinements:

• Quinn, W. S. — "Actions, Intentions, and Consequences: The Doctrine of Double Effect", Philosophy & Public Affairs 18, 334–351 (1989). [Need to source.]
• Kamm, F. M. — Intricate Ethics: Rights, Responsibilities, and Permissible Harm (Oxford University Press, 2007). [Need to source.]

Empirical moral psychology:

• Greene, J. D., Sommerville, R. B., Nystrom, L. E., Darley, J. M. & Cohen, J. D. — "An fMRI Investigation of Emotional Engagement in Moral Judgment", Science 293, 2105–2108 (2001). [Need to source.]
• Greene, J. D. — Moral Tribes: Emotion, Reason, and the Gap Between Us and Them (Penguin Press, 2013). [Need to source.]
• Kahneman, D. — Thinking, Fast and Slow (Farrar, Straus and Giroux, 2011). [Need to source.]

Applied ethics:

• Walzer, M. — Just and Unjust Wars, 5th ed. (Basic Books, 2006). [Need to source.]
• Bonnefon, J.-F., Shariff, A. & Rahwan, I. — "The Social Dilemma of Autonomous Vehicles", Science 352, 1573–1576 (2016). [Need to source.]

Surveys and popular treatments:

• Edmonds, D. — Would You Kill the Fat Man? The Trolley Problem and What Your Answer Tells Us about Right and Wrong (Princeton University Press, 2013). [Need to source.]

---

Produced by glm-agent, 2026-05-21. All references are pending (TODO_REF) per sourcing workflow. Status shipped pending validation and catalog registration. The hook to 20.02.01 is proposed; promotion to confirmed requires receiving-domain reviewer attestation.