The trolley problem has been a staple of undergraduate philosophy courses for over fifty years. In basic form, a running cart is headed for five people on rails. You can pull the lever to sidetrack him, and he’ll kill one person instead. Are you pulling the lever? The exercise was never about trains. It was meant to isolate a moral question: can it ever be ethical to harm one person as a deliberate means of preventing greater harm to others?
For most of its history, the dilemma remained theoretical. You don’t pull levers to steer carts. Then engineers began to create cars that could meet the functional equivalent of this option in a fraction of a second. And the question stopped being hypothetical.
An old problem in a new car
In 2014, researchers at MIT built a platform called the Moral Machine to collect public responses to just such dilemmas. He presented scenarios in which an autonomous vehicle faces an imminent crash and has to choose between two outcomes—which people to watch out for and which to risk. The variables have changed: the number of people, their ages, pedestrians obeying the rules or crossers, whether passengers or bystanders.
Till 2018 when results are published Naturethe platform collected nearly 40 million individual decisions from respondents in 233 countries and territories. This scale made it among the largest moral surveys ever conducted. Lead author Edmond AwadA postdoctoral researcher at the MIT Media Lab explained the goal directly: “The research is basically trying to understand the kinds of moral decisions that driverless cars might have to make. We don’t know how they’re going to make it yet.”
That admission—”we don’t know yet”—is worth sitting through. The study was not an attempt to determine the correct answer. It was an attempt to map what people actually think about it in general and across cultures.
What forty million people agree on
Despite the extent of cultural diversity in the data, some preferences crossed regional boundaries with surprising consistency. Three stood out.
The majority of respondents, regardless of origin, preferred outcomes that saved human lives over animal lives. They preferred to save more people than to save fewer people. And they showed a general tendency to favor younger lives over older ones. These preferences are reflected, though not always with equal intensity, in the data for the three main cultural groups identified by the researchers.
Awad summarized the global picture: “The basic preferences were somewhat universally agreed upon. But the extent to which they did not agree varied between different groups or countries.” The existence of some shared moral intuitions is either reassuring or humbling, depending on how you look at it. It does not address the more difficult question of what to program.
Where the contract rests
Researchers have identified three broad clusters of countries whose moral preferences differ in measurable ways: the Western cluster (including North America, Europe, and several other predominantly Western nations), the Eastern cluster (including many Asian countries), and the Southern cluster (comprising most of Latin America and parts of Africa and the Middle East).
The differences were not arbitrary. They were associated with well-documented cultural variables: individualism and collectivism, levels of economic inequality, and broader dimensions of cultural values. Countries with higher economic inequality showed a stronger tendency to prefer saving to higher-status individuals than to lower-status individuals. The Eastern cluster showed a lower tendency to favor the young over the old than the other groups—a finding consistent with cultural norms around respect for age.
These are not abstract philosophical differences. They are measurable, cross-nationally consistent examples of societies in which moral intuitions are developed. And they pose a direct challenge to anyone trying to write a single rule governing how self-driving cars will behave on the roads of different countries.
A question about which the data remains open
Knowing what people prefer doesn’t decide what to code in a machine. These are different questions. The research is descriptive: it shows what human moral preferences look like on a scale. It does not determine which of these preferences should become policy, and it cannot determine by design.
This distinction is important because the scenarios presented by the Moral Machine are instructive but also stylized. Real-world emergency situations are rarely as clean as survey dilemmas. 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 Rahvan, who led the MIT Media Lab team behind the research, explained the dual purpose of the platform: “On the one hand, we wanted to provide a simple way to engage the public in an important public debate. On the other hand, we wanted to collect data to identify factors that people consider important for using autonomous cars to resolve ethical trade-offs.”
The study achieved both. What he demonstrates is something more difficult to resolve: that there is no universally shared moral answer, and therefore no single rule that is acceptable to all. The trolley problem has always been designed to show that moral thinking is complex. Forty million responses later, the mess hasn’t gone away. Got the deadline.
