Engage public opinion
When an autonomous car must choose who dies in an accident, should it only care about minimizing casualties? Or should some lives matter more than others?
During my time at the MIT Media Lab, I led a team of researchers to build the ‘Moral Machine’ web site. It generates random driverless car dilemmas in which the car may kill different road users: men versus women, old people versus adults versus children, homeless persons vs business executives, jaywalkers versus legal crossers, even cats versus dogs. You can try it for yourself at http://moralmachine.net.
To our astonishment, the website attracted millions of visitors from all over the world. They answered (at the time of this writing) 80 million dilemmas. Such sample size is astronomical, especially if you have to pay people to participate in your experiment, which is often the case—and which we didn’t have to! I wish I could take credit for the success. I think we just got lucky, because the morbid dilemmas fascinated people, and because of all the media hype around autonomous cars at the time.
The first thing we did was measure who was most and least likely to survive. Babies were the least frequently sacrificed group in driverless car dilemmas, followed closely by little girls and boys, and pregnant women. Then came adults, and finally old people.
This preference for young lives over old is intuitive, and is widely adopted in medical ethics. Medical practitioners have long confronted the thorny question of who to prioritize when resources—such as medicine, vaccines, or medical personnel—are scarce. Although guidelines vary greatly, it is common to prioritize people based on a combination of age—prioritizing young people—and prognosis—prioritizing those who are more likely to survive the treatment. Some also take the quality of life into account too. This ends up being a variant of the utilitarian principle, and is often referred to as maximizing quality-adjusted life years (QALY).
Still, it remains an open question whether AI systems should prioritize age and other QALY measures outside of the medical domain, and beyond life and death decisions. There are many domains in which age discrimination is problematic, or even illegal.