Here’s a scenario you’ve possibly encountered before. If not, consider it now.
A runaway trolley is about to hit five people tied to a track. You’re standing next to a lever. If you pull the lever, the trolley will divert onto a side track where, unfortunately, there is one person tied to the rail. The one will be killed; the five will be saved. Do you pull the lever?
Most people say yes. It seems, on balance, better for five people to live and one to die than for five people to die. The choice is uncomfortable, but defensible.
Now consider a second scenario. The same runaway trolley is about to hit the same five people. This time you’re on a footbridge above the track. Next to you is a very large man — large enough that his body, pushed off the bridge, would stop the trolley. You are not large enough to do this yourself; it would have to be him. Do you push him?
Most people say no. Something about this version feels deeply different, even though the numerical calculation is the same: five saved at the cost of one.
Philosophers and psychologists have spent half a century on the question of why these two cases produce such different intuitions, and what that difference tells us about how moral thinking actually works. The trolley problem — originally designed as a narrow ethical thought experiment — has become one of the most persistent sources of insight, and one of the most persistent subjects of criticism, in modern ethics.
The origins
The trolley problem was introduced in 1967 by the British philosopher Philippa Foot, who used it not to pose an unresolvable dilemma but to illustrate a specific philosophical point about the distinction between doing harm and allowing harm. Foot was interested in whether there was a real moral difference between actively killing one person to save five and failing to intervene in a way that would have saved the same five at the cost of one. The trolley case was constructed to isolate the variable — the same number of deaths either way — while varying whether the death was a deliberate act or a side effect.
In 1976, the American philosopher Judith Jarvis Thomson extended Foot’s thought experiment with the footbridge variant and several others. Thomson’s goal was to examine what made actions morally permissible or impermissible — whether it was the outcomes, the intentions, or something about the use of another person as a means to an end. Her work generated decades of philosophical argument about what specific feature of the footbridge case (but not the lever case) produces the widespread moral intuition that it’s wrong to push.
For a long time, this was a philosophical discussion — important in ethical theory, relevant to debates about utilitarianism and deontology, mostly confined to academic journals. Then in 2001, a young neuroscientist changed the conversation.
Greene’s neuroscience
The American neuroscientist and philosopher Joshua Greene, working at Princeton, used functional magnetic resonance imaging to scan people’s brains as they considered trolley problems. His 2001 paper, published in Science, produced findings that reframed the entire discussion.
What Greene found was that the brain handled the two kinds of trolley problems differently. The lever case — pulling a switch to divert a train — primarily engaged brain regions associated with abstract reasoning and calculation. The footbridge case — physically pushing a person — additionally engaged brain regions associated with emotional response, particularly regions involved in empathic distress and direct social engagement. The two scenarios, philosophically identical in their consequences, activated different brain systems because they were psychologically different kinds of events.
Greene’s interpretation, developed across subsequent papers and his 2013 book Moral Tribes, was that human moral cognition isn’t a single unified system. It’s at least two systems — a fast, emotional, intuitive system that produces strong immediate reactions, and a slower, more deliberative system that produces abstract calculations. These systems often agree, but in edge cases like the trolley problems, they disagree. The footbridge case triggers the emotional system strongly — pushing a person is viscerally wrong — while the deliberative system calculates that it would produce the same outcome as the lever case. The intuition wins in most people because the emotional system fires faster and more strongly.
This has important implications. It suggests that our ordinary moral intuitions are not primarily the output of careful ethical reasoning. They’re the output of evolved emotional responses that were shaped by life in small groups, where direct violence was always within arm’s reach and abstract decisions about populations at distance weren’t. These intuitions are reliable guides in many familiar situations and less reliable guides in unfamiliar ones. Modern ethical problems — global poverty, long-term environmental decisions, mass-scale policy choices — often require the deliberative system to over-ride intuitions that were calibrated for a different kind of life.
What the problem is really about
Stripped of the specific scenarios, the trolley problem is a contest between two major ethical traditions.
Utilitarianism, associated with the British philosophers Jeremy Bentham and John Stuart Mill, holds that the right action is the one that produces the greatest balance of good over harm, counted across all affected parties. By this standard, both trolley cases are straightforward: save five, even at the cost of one, either by pulling a lever or pushing a person. The mechanism doesn’t matter; the outcome does. Modern utilitarians, including the Australian philosopher Peter Singer, have often argued that the moral intuitions resisting this conclusion are mere emotional reactions that ought to be over-ridden by careful calculation.
Deontology, associated most famously with the German philosopher Immanuel Kant, holds that some actions are wrong in themselves, regardless of their consequences. Using a person as a means to an end — treating them as a tool, a resource, a object to be deployed — violates their moral status as a rational being. The footbridge case uses the large man’s body as a means of stopping the trolley; the lever case doesn’t. For Kant, this distinction is morally fundamental. The lever case is permissible because the one person’s death is a side effect of a necessary action; the footbridge case is impermissible because the one person’s death is the mechanism by which the five are saved.
The trolley problem keeps being argued about because these two ethical traditions keep being argued about. The thought experiment doesn’t resolve the dispute; it makes the dispute vivid. The utilitarian says five saved is better than one saved, obviously. The Kantian says you can’t use a person as a tool, obviously. Both feel obvious. They generate different verdicts. The trolley problem is where we most clearly see that our moral intuitions are producing competing verdicts and that we have no agreed mechanism for resolving the conflict.
The counter-thread worth hearing
Before accepting the trolley problem as a central feature of ethical thinking, a serious counter-critique is worth engaging with.
The American philosopher Barbara Fried, writing in 2012, argued that the trolley problem has distorted contemporary ethics by focusing attention on cases that almost never actually occur. Most real moral decisions aren’t about killing one to save five. They’re about what to do with resources, how to divide attention, how to balance competing ordinary commitments. The trolley problem, Fried argued, has misled a generation of ethicists into treating the extreme as typical, and has crowded out philosophical work on the much more common moral questions of everyday life.
A related concern has come from philosophers working on applied ethics, who argue that trolley problems are often used to test moral frameworks in ways that don’t transfer to real policy questions. The decision whether to push a large man onto a track bears little resemblance to actual decisions about, say, whether to deploy a medical treatment with uncertain side effects. The thought experiment isolates variables in ways that real situations don’t allow, and the intuitions it generates may not tell us much about what to do when actually deciding.
Peter Singer’s defence has been that thought experiments like the trolley problem, while artificial, usefully clarify the structure of ethical reasoning. They don’t tell us what to do in real situations; they help us notice what principles we’re actually operating on, and whether those principles are defensible when examined. The question isn’t whether the trolley problem predicts ordinary behaviour. It’s whether the intuitions it reveals can withstand scrutiny.
Both positions have merit. The trolley problem is probably both useful and overused — a useful tool for specific purposes, and a misleading tool when treated as the central question of ethics.
The real-world application
One domain where trolley-problem-like questions have become genuinely urgent is the programming of self-driving cars.
A self-driving car heading into a situation where an accident is unavoidable has to choose, in milliseconds, among outcomes. Should it swerve to avoid a pedestrian, risking harm to its passenger? Should it prioritise children over adults? Should it minimise total harm, or minimise the harm to the person who paid for the car? These questions sound like trolley problems, and in their technical structure, they are.
The designers of these systems have had to make decisions that ethicists had previously only thought about. What has become clear, from the experience of actually implementing autonomous-vehicle systems, is that the thought-experiment framing often doesn’t fit the real problem. In actual traffic, you rarely know with certainty what outcomes different actions will produce. The relevant question is almost never save one or five? — it’s what combination of actions produces the lowest expected harm given substantial uncertainty?. The trolley problem’s clean numerical trade-offs disappear under real-world epistemic conditions.
This doesn’t mean the trolley problem is useless for thinking about self-driving cars. It means the problem, in its canonical form, captures only a thin slice of what the real ethical challenges look like.
What to hold from all this
For a student beginning to engage with serious ethics, a few working principles seem defensible.
The trolley problem is worth knowing because it reveals something real about how moral cognition works — that we have multiple moral systems that often conflict, and that our intuitions aren’t a unified output of rational reasoning. This is an important discovery, and knowing it should change how you hold your own moral intuitions.
The specific question of what to do in a trolley case is less important than the general question of how much weight to give intuitions that feel strong. Your moral intuitions are information; they’re not conclusive evidence. Sometimes they reflect wisdom that abstract reasoning misses. Sometimes they reflect biases that abstract reasoning can correct. Distinguishing the two is one of the genuinely hard skills in ethical thinking.
And — importantly — don’t mistake thought experiments for the real ethics of ordinary life. The moral questions you’ll actually face probably won’t be trolley problems. They’ll be slower, messier, involving real people you know, without clean numerical trade-offs. The thought experiment is a useful tool for sharpening your thinking; it’s not a substitute for the harder work of thinking well about the situations you’re actually in.
The question that remains
The deepest thing the trolley problem has contributed to ethical thought is probably the recognition that moral certainty isn’t as well-grounded as we usually feel. The intuitions we bring to the lever and the footbridge cases are strong, consistent, and produced by mechanisms we don’t have direct access to. Those mechanisms evolved for situations different from the ones we now live in. Whether we should over-ride them through deliberate reasoning, or trust them against the reasoning, is genuinely unresolved — and the unresolved quality is itself part of what ethics is.
The question to carry, the next time a moral situation produces an immediate strong reaction in you:
Is this reaction telling you something real about what matters, or is it being generated by a system that’s responding to features of the case that have more to do with your evolutionary inheritance than with what’s actually at stake?
Key research referenced: Philippa Foot, “The Problem of Abortion and the Doctrine of the Double Effect” (1967); Judith Jarvis Thomson, “Killing, Letting Die, and the Trolley Problem” (1976); Joshua Greene’s fMRI research on moral cognition (beginning with “An fMRI Investigation of Emotional Engagement in Moral Judgment” in Science, 2001) and his 2013 book Moral Tribes; Peter Singer’s utilitarian ethics; Barbara Fried’s critique of the trolley-problem tradition.