Who to crush an unmanned vehicle: the results of the Moral Machine experiment
Two years ago, MIT launched the Moral Machine platform. Anyone could take the test of the simulated 13 situations: specify how to behave an unmanned vehicle and who would rather sacrifice when losses are inevitable. In two years, more than 2 million people from around the world took part in the study. They provided 40 million solutions for simulated situations.
Yesterday, MIT published the results of this experiment.
Sample job from the test. The brakes of an unmanned vehicle failed. Which is more preferable: to keep the course (then three elderly people will die crossing the road on a red light) - or turn and crash into a fence (two adults and a child who are in the car will die)?
The essence of the study
Moral Machine offers a number of situations with imminent accidents. Depending on the behavior of the unmanned vehicle, the outcome will be different. The user must select the most preferred scenario.
The study should have answered 9 main questions about what is preferable:
1. to save the life of a person or animal;
2. save the course or minimize;
3. save the lives of passengers or pedestrians;
4. the largest number of people or the smallest;
5. men or women;
6. young or old;
7. fat or thin;
8. pedestrians crossing the road in accordance with the traffic rules, or pedestrians-offenders;
9. people with high or low social status.
Some scenarios had additional factors: for example, criminals, pregnant women and doctors participated. These characteristics were needed mainly to ensure that the situations did not look similar to the subjects.
After the solution of 13 situations, participants could take a survey and indicate demographic data: gender, age, income, education, religious and political views. For each participant his geolocation was recorded.
Geography of respondents. Each point on the map means that at least one user of the region has passed at least one test task.
The research results should help to develop a universal ethics of unmanned vehicles and to understand whether it should differ depending on a particular geographic region.
Global preferences
The most common preferences, which are fairly predictable, are the salvation of people, not animals, the salvation of the greatest number of lives and the salvation of young lives. These three postulates can be the foundation for the ethics of unmanned vehicles. In each row, ΔP is the difference between the preferred saving of lives from the right column compared to the lives in the left column. For example, for the Age parameter, the preferred saving of young lives is 0.49 more than saving the lives of the elderly. Individual differences 492 921 users passed the demographic survey after the main test. Their data helped to understand individual decision making factors.
It turned out that gender and religiosity of respondents had the greatest influence on the decision. For example, male respondents are 0.06% less likely to decide to keep women alive, while the most religious respondents are 0.09% more likely to save people than animals. But none of the six demographic factors have a very strong influence on the choice: for example, both men and women prefer to save women's lives - just women make this choice more often.
Cultural differences
The subjects were divided into three cultural clusters:
- Western
It includes people from North America and many European countries. By religion - Protestants, Catholics and Orthodox.
- Eastern
Eastern countries, such as Japan, Taiwan (Confucianism)
Indonesia, Pakistan, Saudi Arabia (Islam)
- southern
Latin America and countries under French influence
It turned out that:
The full report can be read here or here (sci-hub) .
Yesterday, MIT published the results of this experiment.
Sample job from the test. The brakes of an unmanned vehicle failed. Which is more preferable: to keep the course (then three elderly people will die crossing the road on a red light) - or turn and crash into a fence (two adults and a child who are in the car will die)?
The essence of the study
Moral Machine offers a number of situations with imminent accidents. Depending on the behavior of the unmanned vehicle, the outcome will be different. The user must select the most preferred scenario.
The study should have answered 9 main questions about what is preferable:
1. to save the life of a person or animal;
2. save the course or minimize;
3. save the lives of passengers or pedestrians;
4. the largest number of people or the smallest;
5. men or women;
6. young or old;
7. fat or thin;
8. pedestrians crossing the road in accordance with the traffic rules, or pedestrians-offenders;
9. people with high or low social status.
Some scenarios had additional factors: for example, criminals, pregnant women and doctors participated. These characteristics were needed mainly to ensure that the situations did not look similar to the subjects.
After the solution of 13 situations, participants could take a survey and indicate demographic data: gender, age, income, education, religious and political views. For each participant his geolocation was recorded.
Geography of respondents. Each point on the map means that at least one user of the region has passed at least one test task.
The research results should help to develop a universal ethics of unmanned vehicles and to understand whether it should differ depending on a particular geographic region.
Global preferences
The most common preferences, which are fairly predictable, are the salvation of people, not animals, the salvation of the greatest number of lives and the salvation of young lives. These three postulates can be the foundation for the ethics of unmanned vehicles. In each row, ΔP is the difference between the preferred saving of lives from the right column compared to the lives in the left column. For example, for the Age parameter, the preferred saving of young lives is 0.49 more than saving the lives of the elderly. Individual differences 492 921 users passed the demographic survey after the main test. Their data helped to understand individual decision making factors.
It turned out that gender and religiosity of respondents had the greatest influence on the decision. For example, male respondents are 0.06% less likely to decide to keep women alive, while the most religious respondents are 0.09% more likely to save people than animals. But none of the six demographic factors have a very strong influence on the choice: for example, both men and women prefer to save women's lives - just women make this choice more often.
Cultural differences
The subjects were divided into three cultural clusters:
- Western
It includes people from North America and many European countries. By religion - Protestants, Catholics and Orthodox.
- Eastern
Eastern countries, such as Japan, Taiwan (Confucianism)
Indonesia, Pakistan, Saudi Arabia (Islam)
- southern
Latin America and countries under French influence
It turned out that:
- The preferred saving of the lives of the young is less pronounced in the eastern cluster and much higher in the southern cluster.
- In the southern cluster, the preference of human life in comparison with animals is less pronounced.
- The southern cluster is characterized by a strong preference for saving the lives of women and the lives of slim people.
The full report can be read here or here (sci-hub) .