Self-driving cars may soon make troubling human choices. Their software could have a programmed reaction to the question: If an accident is inevitable, who or what is more valuable?

A human or a pet? Passengers versus pedestrians? The young or the elderly?

Researchers from the Massachusetts Institute of Technology built an online game called the Moral Machine to test how people around the world would answer those questions.

Players were shown unavoidable accidents with two possible outcomes depending on whether the car swerved or stayed on course, and then asked to choose the outcome they preferred. The research gathered 40 million responses from people in 233 countries.

The results, published today in the journal Nature, are just one step towards finding a social consensus around how we expect driverless cars to act, given it will be humans who write the code.

While there were intriguing trends from country to country, globally Moral Machine players showed a preference for sparing babies, little girls, little boys and pregnant women.

Of course, humans don’t always make clear, thought-out decisions when faced with a road accident, and it’s not clear if driverless cars will do any better

What should I read next?

• You decide: Would you let a car determine who dies?
• Driverless cars and the 5 ethical questions on risk, safety and trust we still need to answer
• Driverless cars generate ‘vast amounts of personal information’, committee warned