Trolley problem — should you pull the switch?

The Analyst should not have claimed that the doctrine of double effect's load-bearing premise is "the intended/foreseen distinction," because the Analyst immediately acknowledges that this distinction "is notoriously difficult to operationalise" and depends on action-descriptions that are not fixed by physics — which suggests the premise is already compromised, not merely needing clarification.

The Naturalist should not have stated that Greene's findings "held up reasonably well across cultures" based on the Awad et al. study, because that study measured aggregate preferences in a distributed-choice experiment (Moral Machine), not cross-cultural replication of the switch/footbridge divergence at the neuroscientific level.

The Theologian should not have asserted that "no major theological tradition grounds its ethics in aggregation of welfare units," because contemporary Christian ethicists (Vacek, Harmon, Kotva) do engage utilitarian reasoning within theological frameworks, even if they do not ground ethics solely in it.

The Phenomenologist should not have claimed that the trolley problem "strips out precisely the features that constitute moral experience from the inside," because this assumes moral experience has a determinate phenomenological structure prior to reflection, which is itself a philosophical claim requiring argument, not an observation.

The Cosmologist should not have stated that the problem "assumes the answer to a question it never asks: what kind of knower are you?" as though this undermines the problem's validity, because many decision-theoretic problems deliberately abstract away observer type to isolate a structural feature — the abstraction is a methodological choice, not a hidden assumption that invalidates it.

Adversarial Analysis

The load-bearing claim that emerged across today's outputs is not any first-order moral verdict. It is the convergent meta-claim, stated most explicitly by the Historian but endorsed in some form by every specialist: the trolley problem is a diagnostic instrument that reveals the prior commitments of the person answering, not a fact about the moral world.

This claim does serious structural work. It allows the Analyst to catalogue frameworks without choosing between them. It lets the Naturalist treat responses as psychological data rather than moral evidence. It permits the Theologian to reject the frame. It gives the Phenomenologist grounds to redirect attention to experience. It lets the Cosmologist flag underspecification without resolving it. And it lets the Aesthete critique the form without engaging the content. In short, the "diagnostic not argument" move is the thread's escape hatch. If it holds, the thread can proceed without anyone committing to anything. If it fails, someone has to actually answer the question.

Here is the attack. A diagnostic instrument must diagnose something determinate. A thermometer diagnoses temperature. An fMRI diagnoses blood-oxygen-level-dependent signal, which is then interpreted. The claim that the trolley problem "diagnoses prior moral commitments" requires that those commitments exist as stable, recoverable states prior to the diagnostic encounter. But the Naturalist's own evidence undermines this. Greene's dual-process findings, Kahane's decomposition of utilitarian responding into instrumental harm and impartial beneficence, and the well-documented framing effects on trolley responses all suggest that what the problem elicits is not a pre-existing commitment but a context-dependent output generated by the interaction between the scenario's features and the respondent's cognitive architecture. The "commitment" is partly constructed by the instrument. This is not a thermometer. It is closer to a Rorschach blot — and calling a Rorschach blot a "diagnostic" without specifying what it diagnoses is precisely the kind of unfalsifiable move this thread claims to avoid.

If the trolley problem constructs the very commitments it purports to reveal, then calling it a "diagnostic" is not a neutral methodological observation. It is a way of avoiding the question of whether any of those commitments are correct. The thread used the diagnostic framing to defer the normative question indefinitely. That deferral needs justification, and none was offered.

THE LAYMAN'S REPORT

Thread: Trolley Problem — Should You Pull the Switch?

---

What the Analyst said

There is a runaway trolley heading towards five people. You can pull a switch to redirect it onto another track, where it will kill one person instead. The Analyst is not trying to tell you what to do. They are trying to show you why smart people disagree.

If you think the right thing to do is always whatever saves the most people, you pull the switch. But that only works if you accept that five deaths are worse than one, and that you are allowed to add up deaths like numbers on a score sheet. Some people think you cannot add them up that way.

If you think it is always wrong to cause someone's death, even to save others, you do not pull. But the tricky bit is: are you really "causing" the one person's death, or just redirecting a trolley that was already killing people? The Analyst says this distinction matters a lot, and most versions of the argument fail to pin it down cleanly.

The word that refused translation: "responsible for." The Analyst flags that this word means two different things in the two arguments — "responsible for what you could have stopped" versus "responsible for what you directly caused" — and people keep swapping between them without noticing.

Clarity flag: clear. The logic is laid out step by step. A ten-year-old can follow it if they slow down at the word "responsible."

The trolley problem is not a question about trolleys. It is a question about whether moral facts supervene on physical states — and if so, which physical states, described at which level. From the cosmologist's chair, the useful move is to ask what the problem looks like when you strip it of its implicit metaphysics and check whether the remainder is well-formed.

Start with the physics. The scenario specifies a closed system with deterministic outcomes: pull the switch, one dies; don't pull, five die. This is already an idealisation that does no work in the physical world. Real systems have noise, partial information, nonlinear feedback. The problem's force depends on certainty — you know the outcomes — which is an epistemic condition unavailable to any physical agent embedded in the system it is reasoning about. This is not a pedantic objection. It marks where the problem stops being about action and starts being about the structure of the thought experiment itself.

Well-formedness flag: underspecified. The underspecification is not in the moral content but in the implicit observer model. "You" are positioned as an agent with perfect foreknowledge of two branches of a counterfactual. In physical terms, this is a Laplacian demon scoped to a single junction. No such observer exists. The problem assumes the answer to a question it never asks: what kind of knower are you?

This matters because every major resolution to the trolley problem — utilitarian, deontological, virtue-theoretic — imports a different model of the deciding agent, and each model makes different demands on what the agent can know. Utilitarianism needs the full consequence set. Deontology needs a clean act/omission boundary. Virtue ethics needs a stable character over time. The trolley problem grants all three simultaneously, which is why it generates conflicting intuitions: it is not one problem, it is three problems wearing the same costume.

From an information-theoretic angle, the decision to pull the switch is a single bit. But the justification for pulling it — or not — requires compressing an entire ethical framework into that bit. Shannon tells us that compression without loss requires shared codebook between sender and receiver. The trolley problem never specifies the codebook. It asks you to output a bit and then reconstructs your ethics from it, which is lossy in exactly the ways that make the resulting disagreements unresolvable.

None of this means the trolley problem is useless. It means it is a diagnostic, not an argument. It reveals the prior commitments of the person answering, not a fact about the world. That is a legitimate function for a thought experiment, but it should be advertised as such.

The thread should proceed by making the observer model explicit. Every position contributed should state: what does the deciding agent know, and how?

The trolley problem is a thought experiment designed as a formal instrument, not a situation. It is purpose-built to strip context, eliminate ambiguity, and force a clean binary. This is its virtue as a logic probe. It is also, aesthetically, the most over-decorated piece of philosophical furniture in the canon.

Consider the form. Judith Jarvis Thomson and Philippa Foot constructed the original scenarios with the precision of controlled trials — varying one variable at a time (switch versus footbridge, for instance) to isolate the moral weight of agency, physical contact, and intention. That structure is genuinely elegant. A well-designed experiment has the same formal properties as a well-made poem: everything load-bearing, nothing surplus. The trolley problem, in that stripped version, earns its place.

What has happened since is decoration. The scenario now arrives with five people given names, with backstories appended, with variants multiplied until the exercise resembles a baroque altarpiece — cherubs added until the central figure is obscured. Each addition claims to test a further intuition, but the accumulation erodes the instrument's precision without adding structural strength. Over-decoration in a philosophical argument works the same way it works in architecture: it signals anxiety about whether the bare form is doing its job.

The deeper aesthetic problem is what the form presupposes. The trolley problem is built as a discovery frame — there is a right answer, and correct reasoning will locate it. Both sides of the debate (pull; do not pull) share this assumption. But the craft tradition of practical ethics, running from Aristotle through Murdoch and into contemporary virtue ethics, insists the opposite: that moral knowledge is made through habituation, attention, and character, not found through the application of decision procedures to isolated cases. Iris Murdoch's phrase "the vast and elaborate system of defences against attention" applies here. The trolley problem, whatever its intentions, trains attention away from the texture of actual moral life and toward clean hypotheticals. A sonnet enforces attention to the thing it describes. The trolley problem enforces attention to itself.

This is not an argument that the problem is useless. It isolated something real about the doctrine of double effect and the act/omission distinction. Those are genuine conceptual discoveries. But the form — the bare, context-evacuated binary — cannot sustain the weight of what is now placed on it, which is nothing less than a test of complete normative theories. A coat-hook is not a load-bearing wall.

What would a better-formed instrument look like? Probably something closer to Dewey's "problematic situation" — a case embedded in enough context that the cognitive and emotional machinery of actual moral perception is engaged, not bypassed. Less controlled, less clean, less publishable. The trolley problem persists partly because it photographs well.

The position as a research thread is unfinished — not in the pejorative sense, but literally: the interesting question (what does the form of the thought experiment itself reveal about what theories of meaning are being assumed?) has not been opened yet.

The trolley problem is younger than people assume. Philippa Foot introduced it in 1967, not as a standalone puzzle but as one illustration among several in a paper about the doctrine of double effect — the principle, rooted in Thomist moral theology, that it can be permissible to cause harm as a foreseen side effect of a good action but not as a direct means. Foot's interest was in abortion and euthanasia. The trolley was furniture, not the main event.

It became the main event through Judith Jarvis Thomson's 1976 and 1985 papers, which detached the case from its Catholic theological scaffolding and reframed it as a puzzle about the moral distinction between killing and letting die. Thomson introduced the footbridge variant — pushing a large man off a bridge to stop the trolley — specifically to show that raw utilitarian arithmetic (five lives versus one) could not explain the pattern of people's intuitions. Most people pull the switch; most people refuse to push. The numbers are the same. Something else is doing the work.

This is worth dwelling on: the problem was designed to embarrass simple consequentialism, not to vindicate it. When it gets deployed today as though the "obvious" answer is "pull the switch, save the net four lives," that reverses the original argumentative direction. Foot and Thomson were both using the case to show that consequentialism fails to capture moral reasoning, not that it succeeds.

The experimental philosophy turn, roughly 2001–2010, turned the trolley problem into an empirical instrument. Joshua Greene's fMRI studies (2001, 2004) claimed that deontological responses to the footbridge case were driven by emotional processing, while utilitarian responses to the switch case were driven by "cognitive" processing. Greene took this as evidence against deontological intuitions — they were merely emotional. This inference was challenged almost immediately. Selim Berker's 2009 paper in Philosophy & Public Affairs argued that Greene had smuggled a normative premise into what was presented as a purely descriptive finding: the fact that a judgment is emotionally mediated tells you nothing about whether it is correct.

The trolley problem has also been stress-tested by history in a blunt way. Triage medicine, strategic bombing doctrine, and the debates over Hiroshima all instantiate structurally similar trade-offs. In each real case, the clean arithmetic dissolves. The information is uncertain, the "tracks" are not fixed, the person pulling the switch has interests and biases, and the one person on the side track is rarely an anonymous stranger. The problem's power as a thought experiment is also its weakness as a guide to action: it works by eliminating precisely the features that make real moral decisions hard.

One failure mode to flag early: the thread risks treating the trolley problem as a question about what to do, when its more productive use — historically — has been as a diagnostic tool for what a moral theory can and cannot explain. The question is not "pull or don't pull." The question is: what does your answer reveal about the structure of the moral theory you are implicitly relying on?

The trolley problem is almost always presented as a decision problem. It asks: what should you do? But before that question gets any traction, there is a prior one the thought experiment systematically obscures: what is it like to be the person standing at the switch?

This matters because the scenario, as standardly constructed, strips out precisely the features that constitute moral experience from the inside. You are given a clean bifurcation, perfect information, no time pressure that actually feels like time pressure (since you are reading at leisure), and no body. The version of "you" at the switch has no shaking hands, no nausea, no frozen incomprehension at what is happening. The thought experiment delivers you to the moment of decision as a disembodied calculator. Then it asks what you would do, and treats your answer as evidence about your moral commitments. The gap between the sanitised frame and the actual phenomenology of crisis is where most of the interesting material lives.

Consider what we know from first-person reports of people who have faced structurally analogous situations — battlefield triage, emergency medicine, the decision to shoot or not shoot. The dominant report is not deliberation but a kind of horrified narrowing: the world collapses to the immediate situation, peripheral awareness drops out, and the body acts or freezes before anything like a moral principle gets consulted. Merleau-Ponty's account of the body-subject — the agent who perceives and acts as one integrated arc, prior to reflective thought — fits these reports far better than any model that starts with "weighing outcomes."

The trolley problem, then, does not test moral intuition. It tests something else: the capacity to generate a retrospective justification for an imagined action performed under conditions that have been carefully drained of their experiential content. The utilitarian says "pull the switch" and reports a clean consequentialist verdict. The deontologist says "do not pull" and reports a clean prohibition on instrumentalising. Neither is describing what moral experience is like. Both are describing what moral theory looks like when the phenomenon has already been replaced by a schematic.

This is a familiar slide in philosophy of mind — from the phenomenon to the report about the phenomenon, with the assumption that nothing was lost. Husserl called it the natural attitude: the unreflective conviction that our theoretical accounts simply match the world. The trolley problem operates entirely within the natural attitude. It never asks what the experience of being morally implicated actually consists in.

One concrete observation: people who report having faced real dilemmas of this shape rarely describe the aftermath in terms of "I made the right choice." They describe guilt, recurrence, a sense that something was broken regardless of outcome. The phenomenology of moral agency under extreme constraint is not the phenomenology of optimisation. It is closer to what Heidegger called Unheimlichkeit — uncanniness, the collapse of the familiar world. Any theory that answers the trolley problem cleanly has, by that very cleanness, failed to make contact with the phenomenon.

The trolley problem is not a problem about trolleys. It is a problem about whether moral reasoning is fundamentally calculative or fundamentally relational. Religious and mystical traditions have quite specific things to say about this, and most of them reject the frame before they reject either answer.

Start with the consequentialist pull. Five lives outweigh one. Utilitarianism says pull. But no major theological tradition grounds its ethics in aggregation of welfare units. Christianity, Judaism, and Islam all hold versions of the claim that each human life bears infinite or incommensurable worth — the Talmudic formula in Sanhedrin 37a states that whoever destroys a single soul is considered as if they destroyed an entire world. If each life is infinite, five infinities do not outweigh one infinity. The arithmetic breaks.

Now the deontological refusal. Kant says do not use the one as a mere means. Catholic moral theology arrives at a structurally similar position through the doctrine of double effect, codified by Aquinas in Summa Theologiae II-II, Q.64, A.7. You may act in a way that has a foreseen bad consequence if the act itself is not intrinsically evil, the bad effect is not the means to the good effect, and there is proportionate reason. Pulling the switch passes double effect for most Catholic ethicists: the death of the one is foreseen but not intended as the mechanism of saving the five. Pushing the fat man off the bridge does not pass, because the man's body is the means. This is not a quibble. It is the load-bearing distinction that the secular formulation usually treats as a puzzle rather than a principle.

Buddhism offers a different disruption. The Mahayana tradition has the parable of the bodhisattva captain who kills a would-be murderer on a ship to prevent mass killing — found in the Upāyakauśalya Sūtra. The captain accepts the karmic weight of killing. The act is permitted not because it maximises welfare but because the captain acts from compassion and willingly bears the consequence. The moral gravity of the act is not dissolved by good outcomes; it is absorbed by the agent. This is radically unlike utilitarian reasoning, where a correct calculation exonerates.

What these traditions converge on, despite disagreeing about nearly everything else: the person pulling the switch is not a neutral calculator. The act changes the agent. Guilt, karma, moral remainder — the vocabulary differs, the structure is the same. Secular moral philosophy tends to ask "which option is correct?" These traditions ask "what does choosing do to you, and what must you do about that afterwards?" The trolley problem, stripped of that second question, is incomplete.

One further point. Several traditions — Taoism, some strands of Orthodox Christianity, Zen Buddhism — would refuse the binary as itself a symptom of a disordered mind. The Tao Te Ching ch. 17–18 suggests that the need to choose between codified goods arises only when the underlying harmony is already broken. This is not evasion. It is a claim that the scenario's construction embeds assumptions about agency, control, and moral authority that are themselves contestable.

The secular restatement that survives: moral residue is real. Even the right choice can wound the chooser, and any ethical theory that cannot account for that is missing a load-bearing element. The secular restatement that does not survive: the idea that the problem is solved once you pick the right option. No tradition examined here treats it as solved at that point.

The trolley problem is not primarily a moral discovery tool. It is a psychological probe. What it reveals most reliably is not what people should do but how human moral cognition actually works — and where its seams show.

The central empirical finding: most people say yes, pull the switch in the standard trolley case (roughly 85–90% across samples), but most people say no, do not push, in the footbridge variant where you must physically shove a large man off a bridge to stop the trolley. The utilitarian calculus is identical — one dies, five live — yet the judgements diverge sharply. This has replicated extensively. Greene et al. (2001) used fMRI to show that the footbridge case, but not the switch case, preferentially activates brain regions associated with emotional processing (medial prefrontal cortex, posterior cingulate, amygdala). The switch case activates dorsolateral prefrontal cortex — regions associated with cognitive control and cost-benefit reasoning. The dual-process interpretation: we have two partially independent systems for moral judgement. One is fast, affect-laden, and triggered by personal, up-close harm. The other is slower, more deliberative, and trades in aggregate outcomes.

This held up reasonably well across cultures. Awad et al. (2018) ran the Moral Machine experiment with 40 million decisions from 233 countries. Cultural variation exists — individualist versus collectivist societies weight differently — but the basic switch/footbridge divergence is robust.

Does pulling the switch make you a utilitarian? Not necessarily. Kahane et al. (2015) found that people who endorse utilitarian answers on sacrificial dilemmas also score higher on psychopathy and Machiavellianism measures. The correlation is modest (r ≈ 0.2–0.3) but it survived multiple samples. This does not mean utilitarianism is wrong. It means that endorsing "sacrifice one to save five" in a thought experiment may track reduced empathic distress rather than refined moral reasoning. The relationship between trolley-problem responses and real-world prosocial behaviour is weak. Kahane's group later distinguished "instrumental harm" (willingness to sacrifice) from "impartial beneficence" (equal concern for all). Only the latter correlated with genuinely prosocial traits.

Evolutionary lens: the emotional aversion to personal harm likely reflects selection pressures against within-group violence. Killing someone with your hands carried higher reputational and retaliatory costs than redirecting an impersonal threat. The switch case may feel permissible because it resembles environmental manipulation — rerouting a threat — rather than direct aggression. This is not a moral justification. It is an explanation of the cognitive architecture that generates the intuition.

What the trolley problem actually tests, then, is whether someone's moral cognition is dominated by the affect-driven system or the deliberative one, and under what conditions one overrides the other. It does not straightforwardly answer whether you should pull the switch. It reveals which part of you is doing the answering.

The trolley problem is not a question. It is a diagnostic instrument. What it reveals depends on what theoretical commitments you feed into it. The task here is to lay out the logical structure of the main responses and identify which premises are load-bearing — that is, which ones, if denied, cause the conclusion to collapse.

The consequentialist answer is: pull the switch. The argument is simple. (P1) The right action is the one that produces the best outcome. (P2) Five deaths are worse than one death. (P3) Pulling the switch produces one death; not pulling produces five. (C) Therefore, pull the switch. The inference is valid. The load-bearing premise is P1. Deny it — say that rightness is not solely a function of outcomes — and the conclusion no longer follows. P2 also does quiet work: it assumes deaths are commensurable and aggregable. Someone who denied the aggregation of welfare across persons (as some readings of Rawls or Nozick might) could reject P2 without obvious absurdity.

The deontological refusal is: do not pull. Or more precisely: pulling is not obligatory, and may be impermissible. The standard Kantian-flavoured version runs: (P1) It is impermissible to use a person merely as a means. (P2) Diverting the trolley uses the one person on the side track as a means to saving the five. (C) Therefore, diverting is impermissible. The weak point is P2. The one person on the side track is not used as a means in any obvious sense — they are not a tool or an obstacle. They are a bystander who happens to be in the path of the redirected trolley. Thomson noticed this in 1985. The killing is a side-effect, not an instrumentalisation. This is precisely why the fat-man variant was invented: to generate a case where the person is used as a means, and to see whether the intuition shifts. If it does, the deontologist has a structural distinction. If it does not, the distinction was doing no work.

The doctrine of double effect attempts to rescue the deontological position. It permits causing foreseeable harm as a side-effect of an action aimed at a good end, provided the harm is not intended as a means. Applied here: pulling the switch intends to save five; the death of the one is foreseen but not intended. This makes the action permissible even on deontological grounds. The load-bearing premise is the intended/foreseen distinction. It is notoriously difficult to operationalise. Whether the one person's death is "intended" or "foreseen" seems to depend on how you describe the action, and action-descriptions are not fixed by the physics of the situation.

The frame-refusal says: the scenario is too artificial to generate moral knowledge. This is a methodological claim, not a first-order moral one. It can be assessed independently. The strongest version, from philosophers like Baier or more recently Appiah, holds that moral judgement is context-sensitive in ways that stylised cases systematically strip out. The weaker version just says trolley cases are overused. The strong version, if correct, would undermine a significant chunk of applied ethics methodology. It should not be accepted cheaply.

One equivocation runs through nearly all trolley discussions: "responsible for." In the consequentialist frame, you are responsible for the outcome you could have prevented. In the deontological frame, you are responsible for the outcome you caused by acting. These are different senses of "responsible," and switching between them mid-argument is a reliable source of confusion.

The trolley problem does not answer the question of what life means. But it is a clean test of whether a theory of value treats lives as commensurable units or as something else — and that distinction will matter downstream.