With the introduction and growth of Driverless Cars (or Autonomous Vehicles “AVs“) a pressing ethical question has arisen over how to program these cars in situations that will clearly result in an unavoidable collision. Faced with the choice of saving the “driver” (or the occupants of the vehicle) by plowing into a group of pedestrians, or sacrificing the occupants for the greater good, should the AV be programmed to save as many lives as possible?
A study by Azim Shariff of the Culture and Morality Lab at the University of Oregon with the help of additional researchers from France and MIT, tested the public’s attitudes toward these kinds of decisions. That study asked respondents to choose what they thought should be programmed to happen where they were the occupant of the AV in question as well as from the perspective of the pedestrians (or some other innocent party such as a school bus or another vehicle containing children). Reponses were mixed with a certain percentage at least saying they would choose the option that favoured the greater good. But the real answer becomes clear if we look at the issue from a broader perspective – set all AVs to protect the owner/driver. In an unavoidable collision between two AVs, both will respond to protect themselves.
With human drivers we can assume they would naturally default to protecting themselves, and their motor reactions in a split-second decision would reflect that – swerve to avoid accident; protect the driver. Once the majority of vehicles on the road are AVs, the basic programming default in all of them should be to protect the driver. That way even the vehicle about to be hit will respond, if it can, to protect its driver. Superseding that baseline programming with lines of code that instruct an AV to make proactive choices about who lives and dies introduces far too many thorny and complicated decisions. Moreover, what makes us so certain the computer will have enough data to make the right proactive decision to “kill or injure the fewest (or oldest) people” in any scenario. What if that school bus potentially carrying 20 children is empty?
The only reasonable solution is to program every AV to protect its occupants so that (eventually) all the cars on the road have that at their core.
And as we march toward more AVs on our roads one thing is becoming clear, “…driverless cars will be far better at avoiding collisions than humans.” So says a report from the Conference Board of Canada. That report predicts an 80% reduction in traffic fatalities once we achieve an era when driverless cars become the majority of vehicles on the road.
Source: CBC News – Edmonton (http://www.cbc.ca/news/canada/edmonton/computers-could-decide-who-“lives-and-dies-in-a-driverless-car-crash-1.3297177)
To get to that point, however, we need ever more data to train the computers that will eventually control driverless cars completely. Where Google's Waymo division was the foremost source of data in this respect, Tesla has taken over the lead simply by collecting more data from more data-points. Those data-points are every Tesla Model S, X and now Model 3 on the road. That broad spread of data-points that are collecting information about real world driving increases by 10's of thousands (and soon 100's of thousands) each year as Tesla puts more cars on the road. And as the cars proliferate around the globe, Tesla's real world data from driverless cars also grows in a wider variety of places and circumstances. Google's approach has been to collect data from their own cars, typically in controlled circumstances, but solely from the limited number of cars they have on the road. Tesla on the other hand, is building a data set from an increasing number of cars because each car they sell becomes a stream of data for analysis.
The solution to the Ethics of Driverless Cars, ultimately, is data to support the right answer not humans debating how they should be programmed. And Tesla has the most valuable data set that continues to become more valuable with every car that rolls off the assembly line.
No comments:
Post a Comment