It’s a day like any other. You are driving along your regular route on a clear day with good visibility.
Suddenly, things take a terrifying turn. A driver who isn’t paying attention to the road drives into your lane and is coming at you head on. There are only three directions you can go:
- Into a brick wall, which will put your life at severe risk;
- Into a group of school children walking on a field trip;
- Into a bus shelter that has one adult sitting in it.
What do you do?
Now consider self-driving vehicles in this scenario. Can a machine be expected to understand morals and make ethical decisions?
It’s possible for engineers to program vehicular software to, for instance, not swerve for squirrels, or ensure a car will prioritize avoiding humans over parked cars. However, when it comes to those split second, life-altering decisions, questions have been rightfully raised surrounding the decision-making of driverless cars.
In a University of Osnabruck study, researchers had human subjects use virtual reality to monitor their behaviour in simulated traffic scenarios. Taking the data collected, they created algorithms tailored to teach self-driving vehicles how to handle these potentially disastrous scenarios.
The outcomes of the study, that had participants go for a foggy “drive” in the suburbs and faced them with a sudden collision, were fascinating.
Researchers learned that drivers saved:
- Children over adults
- Dogs over other animals, such as deer, goat and boar.
However, researchers identified a key difference in the way that humans versus cars respond to these situations ¬– probability. For driverless cars, all decisions are made based on probability, therefore, instead of asking “should I hit the dog or the human”, a car might try to determine the risk factors, such as “If I hit the human, it will only experience a minor injury, but if I hit the dog it will die.”
At the end of the day, computers are not human. Nuance slips past them in favour of coded data, and before driverless cars hit the roads en masse, engineers must be willing to continue to conduct research into understanding just how to program these vehicles. There must be a way to consider these scenarios which may be more complex than current technology permits.
The article in this client update provides general information and should not be relied on as legal advice or opinion.