A runaway trolley is barreling down the tracks toward five people who cannot move. 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. Pull, and one dies because of you.
That is the whole setup. No tricks, no third option. And yet this little scenario has kept philosophers arguing for six decades, spawned a field called "trolleyology," become a meme format, and forced real engineers at real car companies to ask: what should the car do?
Here is where it came from, the variants that make it genuinely uncomfortable, what millions of real answers revealed, and the part most explainers skip: why the people building self-driving cars threw it out.
One Runaway Trolley: The Original 1967 Setup
The trolley problem is older than most people guess, but not as old as the meme makes it feel. According to [Britannica](https://www.britannica.com/topic/trolley-problem), the British philosopher Philippa Foot introduced the scenario in a 1967 essay about the doctrine of double effect. In her version it was the trolley driver facing the choice: stay the course and kill five workers, or steer onto a side track and kill one.
Foot was not trying to invent a party question. She wanted to probe a puzzle: why does diverting the trolley feel permissible, while other ways of trading one life for five feel obviously monstrous? Her answer leaned on the difference between negative duties (do not harm) and positive duties (help), and why the first tends to outrank the second.
Then in 1976, the American philosopher Judith Jarvis Thomson picked up the scenario, gave it the name "trolley problem," and moved you from the driver's seat to the trackside as a bystander with a lever. That small change matters. The driver is already entangled; you could just walk away. Choosing to act makes the one death feel like your doing.
Would You Pull the Lever? The Numbers Say Yes, Your Gut Says Maybe
Ask a room full of strangers and most hands go up for pulling the lever. Five lives against one looks like simple arithmetic, and utilitarians (people who judge actions by outcomes) say exactly that: minimize total harm, pull, done.
But notice what your gut does while your hand is going up. There is a flicker of resistance, and that flicker is the interesting part. Pulling the lever means a specific person dies because of a decision you made. Doing nothing keeps your hands technically clean while five people die in front of you. Deontologists (people who judge actions by rules and duties) take that flicker seriously: maybe there is a real moral line between killing and letting die, and the lever crosses it.
The trolley problem works precisely because both answers are defensible and neither feels fully clean. If you answer instantly and feel zero discomfort, you have probably not understood the question yet.
The Famous Variants, Played as Dilemmas
Thomson and the philosophers who followed her did something clever: they kept the body count identical and changed only the mechanism. If your answer flips between versions, the raw math of five versus one was never the whole story.
The Footbridge (the "Fat Man" Variant)
Same trolley, same five people. But now you are on a footbridge above the tracks, standing next to a large man. The only way to stop the trolley is to push him off; his body will stop it, he will die, the five will live. Arithmetic unchanged, yet most people who happily pulled the lever refuse to push. Using someone's body as a brake feels different from redirecting a threat. Would you push?
The Loop
Now the side track loops back onto the main line. The one person on the loop is the only thing that will stop the trolley before it rejoins and hits the five. Pull the lever here and you are not merely diverting the threat; you need that person to be hit for your plan to work. It is the footbridge in disguise, wearing the lever's clothes.
The Transplant Surgeon
Thomson's sharpest variant leaves the tracks entirely. You are a surgeon with five patients, each dying for want of a different organ. A healthy traveler walks in for a checkup. Harvest his organs and all five live. Almost nobody says yes, and almost everybody struggles to explain why this differs from the lever, since it is once again one life traded for five. If you want to know what you believe about ends and means, this is the variant that tells you.
These hit differently as live dilemmas. Reading that most people refuse to push is trivia; locking in your own answer, then watching it contradict your lever vote from two minutes earlier, is self-knowledge.
Would YOU pull the lever? [Rule on real dilemmas and see how the world voted →](https://bejudge.com)
What 40 Million Real Answers Revealed
For decades the trolley problem lived in seminar rooms. Then researchers at MIT turned it into a global experiment called the Moral Machine: an online game presenting self-driving car dilemmas where someone has to be put at risk, and asking millions of people who it should be.
According to [MIT News](https://news.mit.edu/2018/how-autonomous-vehicles-programmed-1024), the project gathered nearly 40 million decisions from more than 2 million participants across 233 countries and territories. A few preferences showed up almost everywhere: spare humans over animals, spare more lives over fewer, spare the young over the old.
The regional differences are where it gets uncomfortable. Eastern countries showed a much weaker preference for sparing the young over the elderly than southern countries did. Even if you wanted to program a car with "what humanity prefers," humanity does not have one answer. That turns the trolley problem from a quirky puzzle into a governance question.
Why Self-Driving Car Engineers Actually Rejected It
The people closest to autonomous vehicles mostly think the trolley framing is the wrong tool, and they are not shy about saying so.
According to [IEEE Spectrum](https://spectrum.ieee.org/av-trolley-problem), ethics researcher Veljko Dubljević puts it bluntly: "The trolley paradigm was useful to increase awareness... but it is a misleading framework to address the problem." His objection is structural. The trolley problem hands you two fixed outcomes with total certainty. A real vehicle never has that. Outcomes are probabilistic, sensors are noisy, and crucially, an AV's choices have to be programmed in advance rather than improvised in the moment.
According to [Brookings](https://www.brookings.edu/articles/the-folly-of-trolleys-ethical-challenges-and-autonomous-vehicles/), researcher Heather Roff goes further: AVs make sequential decisions under uncertainty, continuously updating probability estimates rather than facing one clean binary fork. She argues the real ethical questions are quieter and bigger: what tradeoffs engineers encode into the system's objectives, how AVs reshape cities, who gets surveilled inside shared vehicles. And with more than 30,000 traffic deaths in the US every year, the case for automation was always about reducing ordinary crashes, not adjudicating exotic ones.
There is also a darker downstream question. According to [The Conversation](https://theconversation.com/the-self-driving-trolley-problem-how-will-future-ai-systems-make-the-most-ethical-choices-for-all-of-us-170961), researchers Jumana Abu-Khalaf and Paul Haskell-Dowland warn that AI inherits the values and biases of whoever builds it; facial recognition could in principle let systems assign different worth to different lives. That is a design decision, not a thought experiment. More on that in [this piece on outsourcing decisions to AI](/blog/ethics-outsourcing-decisions-ai-2026).
So the engineers are right that no car will literally face Foot's fork in the tracks. But somebody still has to choose the values in the value function. The trolley moved from the road into the codebase.
What It Still Teaches You About Your Own Judgment
Strip the tracks away and the trolley problem is a diagnostic for how you judge. It exposes three things fast.
First, your moral intuitions are not one consistent system. Most people are utilitarians at the lever and deontologists on the footbridge, within the same five minutes. That is not hypocrisy; it is information about which features of a situation your conscience actually tracks.
Second, action and inaction do not feel morally equal even when the outcome sheet says they are. "I just didn't do anything" quietly becomes a decision.
Third, judgment degrades under load. Your answer on a fresh morning differs from your answer after a draining day, which is why [decision fatigue bends your moral compass](/blog/decision-fatigue-moral-compass-2026) more than you expect. The same blind spots surface in [digital ethics dilemmas](/blog/deepfakes-digital-ethics-2025) too.
Judge It Yourself
Reading is the easy mode. The trolley problem only does its real work when you commit to an answer with no takebacks, then see your verdict sitting next to everyone else's. That is what BeJudge is built for: real dilemmas, your ruling, and a live view of how the world voted. Some of your answers will surprise you. The contradictions will surprise you more.
Would YOU pull the lever? [Rule on real dilemmas and see how the world voted →](https://bejudge.com)
FAQ
Who invented the trolley problem?
The British philosopher Philippa Foot introduced the core scenario in a 1967 essay, with the trolley driver choosing between five workers and one. The American philosopher Judith Jarvis Thomson coined the name "trolley problem" in 1976 and added the most famous variants, including the footbridge and the transplant surgeon.
What is the "correct" answer to the trolley problem?
There is no settled correct answer, and that is the point. Utilitarian reasoning says pull the lever to save five at the cost of one, while duty-based reasoning warns that actively killing differs from failing to save. The scenario exists to expose the tension between those two ways of judging, not to crown a winner.
Do self-driving cars actually use the trolley problem?
Not in any literal way. Researchers who study autonomous vehicles argue the trolley problem assumes certainty and a single binary choice, while real cars make continuous probabilistic decisions with noisy sensor data. The ethical questions instead live in the objectives and tradeoffs engineers program in advance.
Why do people pull the lever but refuse to push the man off the footbridge?
The outcomes are identical, one death instead of five, but the mechanism differs. Pulling the lever redirects an existing threat; pushing uses a person's body as a tool, which most people's intuitions treat as a harder moral line. That gap is one of the most studied findings in moral psychology.
Is the trolley problem still relevant in 2026?
Yes, though its job has changed. It is no longer a blueprint for programming cars; it is a fast way to discover where your moral instincts contradict each other. With AI systems now making consequential calls, knowing how value judgments get embedded in machines matters more than ever.