Ethicists (and, to the chagrin of many, increasingly carmakers) use the so-called trolley problem to gauge people’s moral calculus: If a trolley threatens to run over people on one track, would you pull a lever to switch it to a track where it would only kill one person? Some people would prioritize the number of lives they could save. Others would prioritize inaction over interference. Still others would ask which people had permission to be on the tracks in the first place and flatten any that were breaking the rules.

In a survey of 2.3 million people living in 233 countries and territories, researchers at the Massachusetts Institute of Technology asked respondents to answer 13 variations on the trolley problem, involving self-driving cars and combinations of victims young and old, rich and poor, and rule followers or breakers, among others. The survey emerged from a previous study that discovered a paradox: respondents in that case wanted self-driving cars to favor pedestrians over passengers—but wouldn’t want to buy such a car.

Respondents to the new survey fell into three broad clusters of cultural preferences, the researchers report in Nature. The differences correlate with both modern institutions and deep cultural traits, the authors write.

For example, respondents from countries with strong government institutions and trust, such as Finland and Japan, were likelier to opt for hitting wayward pedestrians than to sacrifice rule-following passengers.

Those from countries with great income inequality, like Colombia, were likelier to favor hitting a homeless person than a business executive.

Carmakers are global, so they will need to use the findings at the very least to adapt how they sell their increasingly autonomous cars, if not how the cars actually operate.

“We need to come up with a social consensus about which risks we are willing to take,” Audi autonomous vehicle ethicist Barbara Wege told Nature.