The Trolley Problem: What Your Answer Reveals About Your Moral Psychology

Pull the lever or push the man? The arithmetic is identical; most people answer differently. Why — and what does it mean for AI ethics?

The Setup

A runaway trolley is heading toward five people tied to the tracks. You can pull a lever to divert it to a side track, where one person is tied. Pull the lever: one dies, five live. Don't pull it: five die, one lives. What do you do?

Most people say pull the lever. The utilitarian calculus is clear: five lives outweigh one. Now consider a variant. You're on a bridge above the tracks. The trolley is heading toward five people. The only way to stop it is to push a large man off the bridge and onto the tracks — his body will stop the trolley. He will die; the five will live. Do you push him?

Most people say no — even though the arithmetic is identical. This is the footbridge variant, and the difference in responses is what makes the trolley problem philosophically interesting.

Judith Jarvis Thomson and the Philosophy

The trolley problem was formalized by Judith Jarvis Thomson in a 1985 paper building on earlier work by Philippa Foot. Thomson was interested in distinguishing between actions that cause harm as a means versus harm as a side effect — the doctrine of double effect. Pulling the lever kills one as a side effect of saving five. Pushing the man kills him as the means to save five. Most people feel this distinction matters morally, even when the outcomes are numerically identical.

The utilitarian says: outcomes are what matter; both actions save four net lives; both are equally permissible or both equally required. The deontologist says: there are constraints on what you may do to a person even for good consequences — using someone's body as an instrument violates their dignity in a way that diverting a trolley does not.

Haidt's Challenge: Emotions Come First

Social psychologist Jonathan Haidt's research on moral psychology presents a challenge to both frameworks. In his 2001 paper "The Emotional Dog and Its Rational Tail," Haidt argued — based on experimental studies of moral dumbfounding — that moral judgments are primarily driven by fast emotional intuitions, with rational reasoning playing a post-hoc justification role rather than a causal one.

In moral dumbfounding experiments, participants make moral judgments they cannot justify — they feel that something is wrong but cannot articulate a coherent reason why. When their justifications are systematically dismantled, they don't change their judgment; they just say "I know it seems weird, but it's still wrong." The reasoning follows the feeling; it doesn't produce it.

What fMRI Studies Show

Joshua Greene at Harvard conducted neuroimaging studies of trolley problem responses. Personal moral dilemmas (like the footbridge variant, involving direct physical contact) activated emotional brain regions — the medial prefrontal cortex, posterior cingulate cortex, and amygdala — more strongly than impersonal dilemmas (like the lever variant). Utilitarian judgments in personal dilemmas correlated with activation in cognitive control regions (dorsolateral prefrontal cortex) and longer response times, suggesting the person was actively overriding an emotional response.

Greene interprets this as evidence that utilitarian judgment requires effortful suppression of emotional responses. Critics argue this doesn't tell us which judgment is correct — it just reveals the mechanism. Feeling-based responses might be tracking morally relevant features (like the distinction between using someone and allowing harm) that a purely utilitarian calculus misses.

Why This Matters for AI Ethics

As autonomous vehicles, surgical robots, and AI systems in medical triage make life-and-death decisions algorithmically, the trolley problem is no longer merely philosophical. MIT's Moral Machine project collected 40 million moral decisions from people in 233 countries making trolley-problem-style choices for autonomous vehicles — who should the car sacrifice in unavoidable accident scenarios? The data revealed strong cross-cultural patterns as well as significant cultural variations in factors like age, gender, and social status of potential victims. There is no universal answer, and embedding any particular answer in deployed AI systems amounts to an unelected ethical choice.

FAQ

Q: Is the trolley problem a realistic scenario, or is it too artificial to yield useful moral insights?

A: The trolley problem is artificial by design — it strips away confounding factors to isolate specific moral principles. This is its value and its limitation. Real moral situations involve uncertainty, relationships, contextual history, and options not present in the stylized scenario. But the intuitions the trolley problem reveals are genuine, and they drive real behavior: medical triage decisions, military targeting rules, public health resource allocation, and AI ethics all grapple with structurally identical tradeoffs.

Q: What does psychological research show about who chooses utilitarian responses to the trolley problem?

A: Studies by Bartels and Pizarro (2011) found that individuals who gave utilitarian responses to personal moral dilemmas scored higher on measures of psychopathy and Machiavellianism. A 2020 meta-analysis by Kahane and colleagues found that utilitarian judgment on these problems is associated with reduced empathy and increased deliberate reasoning — not necessarily with greater moral sensitivity. This complicates the assumption that utilitarian responses represent clearer moral thinking.

Curious minds belong together. Find people who ask the same questions on Good People Map.