A man is racing toward a concrete pillar, the brakes of his car have failed. A collision would mean his certain death. He could dodge, but then he would have to run over a child who is walking across the street. Who do you think should survive the crash?

So far, such considerations are hardly possible in practice. After all, the driver would only have seconds to decide. He probably would not realize who was walking across the street at all. And even if: would not he intuitively save his own life rather than a complete stranger?

Smart cars could be more neutral in such situations. You expect to have enough computing power to predict the consequences of a crash and to analyze the situation accurately. The cars would then decide who is allowed to live and who has to die. But according to what criteria?

Moral dilemma

This decision would continue to be made by people who program the cars. An international research team has now analyzed which criteria would most likely be accepted, as reported in the journal "Nature". The scientists have evaluated a huge, international survey.

The Moral Machine team

Scenario of the Moral Machine: If there is no correct answer

The data comes from the freely accessible online platform Moral Machine, where users can play through various scenarios in car accidents - including the example with the man and the child. Read more about the experiment here.

The test persons can decide how the autonomous driving car should behave. According to the researchers, millions of people from 233 countries have already participated in the experiment. Overall, more than 40 million decisions have been documented. Nevertheless, the experiment is not representative. For example, young men were disproportionately represented.

"We wondered if there are moral principles that apply worldwide and wanted to collect as much data as possible," says Iyad Rahwan from the Massachusetts Institute of Technology's Media Lab, MIT, the SPIEGEL. After all, it's about explosive questions. For example, what is the lesser evil: If three adults die or two children?

Manufacturers have avoided this moral dilemma so far. They are still too busy to bring a self-driving car to market. Another argument of the car industry: If the road is only in robotic hands, no more accidents should happen anyway.

How safe is autonomous driving? The fairy tale of the faultless self-steering car

"I'm sure autonomous cars can help reduce the number of accidents, but you never can rule them out," says Rahwan. The researchers therefore wanted to provoke an ethics debate with their study.

And they succeed too. Because in the individual scenarios, the test subjects not only have to choose whether a child or an old man should survive. Other information about those affected is also provided: some participants are fat, others are trained. Some are homeless, other doctors or thieves. Some ran across the street in red and were in danger only because of this and again others are not humans but dogs or cats.

According to the researchers, the data show several tendencies:

  • Most test persons are more likely to save as many lives as possible. Age, gender or social position play a minor role in these cases.
  • On average, more women are rescued than men and more children than the elderly. The fitness, however, has little effect on the decision. The social position against it already. Homeless people and thieves are more likely to die than doctors.
  • When it comes to choosing between human and animal survival, most people choose humans.
  • People who have run across the street at red will be saved less often than those who have behaved correctly.
  • Riders in the car are no longer protected as pedestrians. According to the researchers, this is an indication that the test persons actually see themselves as uninvolved observers and do not identify with the driver or the passenger.

"Although the test subjects often made similar decisions, we were surprised at how many regional differences there are," says Rahwan. In many Asian countries, for example, young people were spared less often than the elderly. Probably, because older people would be particularly respected in these countries, the researchers suspect.

Participants in Central and South America also decided much more often to intervene in the events, instead of renouncing the steering. This also applied to subjects from Germany. An overview of the individual countries can be found here.

According to the researchers, the study results can also be applied to other areas of artificial intelligence. For example, when robots are to decide who to rescue first after an earthquake. "We need to discuss such issues socially," Rahwan says.

Which life is most valuable?

Other researchers warn against the possible conclusions of the study, because behind every decision is the controversial question: which life is more valuable?

"Nothing can be learned about the ethical permissibility of norms either from games or surveys," says Armin Grunwald of the Karlsruhe Institute of Technology, who was not involved in the study. "Otherwise, after every serious crime, a poll could be made that would almost certainly lead to the introduction of the death penalty."

The German ethics committee sees it similarly. In its 2017 report, "Autonomous and Connected Driving", it states: "In the event of unavoidable accident situations, any qualification for personal characteristics (age, gender, physical or mental constitution) is strictly prohibited."