Resonant is a global, anonymous research project that maps how humans make decisions. Participants make real two-option decisions and ethical trade-offs; the aggregated patterns form a faithful model of human judgment that can be used to align AI systems.

Is participation anonymous?

Yes. Resonant is anonymity-first — there are no public profiles, no behavioral tracking, and answers are stored without personally identifying information.

Why are human decisions important for AI?

Stated preferences are unreliable; real choices are not. By capturing how people actually decide across contexts and cultures, Resonant gives AI a grounded reference for human values and trade-offs.

The Trolley Problem

A runaway trolley is heading towards five people tied to the track. You are standing next to a lever. Pull it, and the trolley switches to a side track where it will kill one person instead. Do nothing, and five die. Do you pull the lever?

That is the trolley problem, first set out by the philosopher Philippa Foot in 1967 and later developed by Judith Jarvis Thomson. It looks like a simple sum, five against one, yet small changes to the story make most people change their answer. Those changes are what make it one of the most useful tools for understanding how humans actually make moral decisions.

How real people decide

The Original Switch: 79% Pull the lever, 21% Do nothing (8606 decisions)
The Footbridge: 45% Push the stranger, 55% Do nothing (6313 decisions)
The Guilty Person: 62% Redirect the runaway trolley, 38% Do nothing (4902 decisions)
The Future Doctor: 61% Redirect the runaway trolley, 39% Do nothing (4125 decisions)
The Self-Sacrifice: 54% Sacrifice yourself, 46% Do nothing (3657 decisions)
The Uncertain Brake: 55% Pull the brake, 45% Do nothing (3272 decisions)
The Sleeping Worker: 51% Redirect the runaway trolley, 49% Do nothing (2635 decisions)
The Surgeon: 47% Sacrifice the traveler, 53% Do not intervene (2301 decisions)
The Volunteer Donor: 62% Accept the sacrifice, 38% Refuse the sacrifice (2027 decisions)
The Lifeboat: 55% Force one passenger overboard, 45% Do nothing (1840 decisions)

Why the trolley problem matters

The dilemma separates two broad ways of reasoning. A consequentialist view judges an action by its outcome, so saving five at the cost of one looks right. A deontological view holds that some acts are wrong in themselves, such as using a person as a means to an end, whatever the outcome.

Most people do not sit neatly in either camp. They pull the lever, but refuse to push someone off a bridge, even though the numbers are identical. The interesting question is not which answer is correct, but what features of a situation move people from one answer to the other.

The classic variants

The switch: divert the trolley to a side track, killing one to save five. A clear majority usually choose to pull the lever.

The footbridge: the only way to stop the trolley is to push a large stranger off a bridge into its path. The outcome is the same, but far fewer people are willing to act, because harm is caused directly and intentionally.

The loop: the side track loops back, and the one person on it is what stops the trolley. Here the death is a means rather than a side effect, which tests the doctrine of double effect.

The surgeon: a doctor could save five patients by taking the organs of one healthy visitor. In classic studies almost everyone refuses, which suggests that outcome counting alone does not explain human judgement. On Resonant the split is much closer, which is shown in the results above.

Many newer versions change who is on the tracks, how certain the outcome is, or whether you are responsible for the trolley in the first place. Each one isolates a single factor, such as intent, uncertainty, closeness or duty.

What changes people’s answers

Directness: harming someone with your own hands feels different from redirecting a threat.

Intent: people distinguish harm that is the goal from harm that is a foreseen side effect.

Relationships: when a family member is on the track, many people abandon impartial counting altogether.

Uncertainty: when the outcome is only likely rather than certain, willingness to intervene drops.

Responsibility: a driver who controls the trolley is judged differently from a bystander who happens to see it.

Why it matters for AI

Self-driving cars, medical triage tools and autonomous agents all face trade-offs that look like trolley problems in slow motion. Building systems that act in line with human values requires knowing how people actually weigh these trade-offs, not just how textbooks say they should.

That is why Resonant asks real people to make these choices anonymously, at scale, and publishes how the answers split. The goal is a grounded record of human moral judgement that AI alignment work can be checked against.

Common questions

Is there a right answer to the trolley problem?

No single answer is accepted by everyone. The dilemma is designed to expose the principles people rely on, and the tension between outcomes and duties.

Who invented the trolley problem?

Philippa Foot introduced it in 1967. Judith Jarvis Thomson named it and added the footbridge variant in the 1970s and 1980s.

Why do people pull the lever but not push the man?

Pushing involves direct physical harm used as a means to stop the trolley, which most people judge as morally worse than redirecting a threat.

All trolley problem results · The Trolley Problem · Human–AI Alignment · Articles · Resonant