The trolley problem has been a fixture of undergraduate philosophy courses for more than fifty years. In its basic form, a runaway trolley is heading toward five people on the tracks. You can pull a lever to divert it to a side track, where it will kill one person instead. Do you pull the lever? The exercise was never really about trains. It was designed to isolate a moral question: whether it can ever be ethical to cause harm to one person as a deliberate means of preventing greater harm to others.
For most of its history the dilemma stayed theoretical. You don’t actually pull levers to divert trolleys. Then engineers began building cars that might, in a fraction of a second, face the functional equivalent of that choice. And the question stopped being hypothetical.
The old problem in a new machine
In 2014, researchers at MIT built a platform called Moral Machine to gather public responses to exactly these kinds of dilemmas. It presented scenarios in which an autonomous vehicle faces an unavoidable accident and must choose between two outcomes — which people to spare, and which to put at risk. The variables shifted: the number of people, their ages, whether they were pedestrians following the rules or jaywalking, whether they were passengers or bystanders.
By 2018, when the results were published in Nature, the platform had collected nearly 40 million individual decisions from respondents in 233 countries and territories. The scale put it among the largest moral surveys ever conducted. Lead author Edmond Awad, a postdoctoral researcher at the MIT Media Lab, described the goal directly: “The study is basically trying to understand the kinds of moral decisions that driverless cars might have to resort to. We don’t know yet how they should do that.”
That admission — “we don’t know yet” — is worth sitting with. The study was not an attempt to prescribe the right answer. It was an attempt to map what human beings, in aggregate and across cultures, actually think about it.
What forty million people agreed on
Despite the scale of cultural variation in the data, some preferences crossed regional lines with surprising consistency. Three stood out.
Most respondents, regardless of origin, preferred outcomes that spared human lives over animal lives. They preferred saving more people rather than fewer. And they showed a general tendency to prioritize younger lives over older ones. These preferences showed up in the data across the three major cultural clusters the researchers identified, though not always with equal intensity.
Awad summarized the global picture: “The main preferences were to some degree universally agreed upon. But the degree to which they agree with this or not varies among different groups or countries.” The existence of some shared moral intuitions is, depending on how you look at it, either reassuring or modest. It does not resolve the harder question of what to program.
Where the agreement stops
The researchers identified three broad clusters of countries whose moral preferences differed in measurable ways: a Western cluster (including North America, Europe, and several other predominantly Western nations), an Eastern cluster (including many Asian countries), and a Southern cluster (covering much of Latin America and parts of Africa and the Middle East).
The differences were not arbitrary. They correlated with well-documented cultural variables: individualism vs. collectivism, levels of economic inequality, and broader measures of cultural values. Countries with higher economic inequality showed a stronger tendency to prefer sparing higher-status individuals over lower-status ones. The Eastern cluster showed a less pronounced tendency to favor younger people over older ones compared to other groups — a finding consistent with cultural norms around respect for age.
These are not abstract philosophical differences. They are measurable, cross-nationally consistent patterns that reflect the societies in which moral intuitions develop. And they create a direct problem for anyone trying to write a single rule that will govern how self-driving cars behave on roads across different countries.
The question the data leaves open
Knowing what people prefer does not settle what should be encoded in a machine. These are different questions. The study is descriptive: it maps what human moral preferences look like at scale. It does not, and by design cannot, determine which of those preferences ought to become policy.
That distinction matters because the scenarios the Moral Machine presents, while instructive, are also stylized. Real accident situations are rarely as clean as the dilemmas in the survey. The variable being tested is moral preference in a controlled scenario, not moral behavior in a chaotic real-world event. The relationship between the two is real but imperfect.
Iyad Rahwan, who led the MIT Media Lab group behind the research, described the dual purpose of the platform: “On the one hand, we wanted to provide a simple way for the public to engage in an important societal discussion. On the other hand, we wanted to collect data to identify which factors people think are important for autonomous cars to use in resolving ethical tradeoffs.”
The study achieved both. What it also demonstrated is something harder to resolve: that there is no single moral answer that is universally shared, and therefore no single rule that will feel acceptable to everyone. The trolley problem was always designed to show that moral reasoning is complicated. Forty million responses later, the complication has not gone away. It has acquired a deadline.
