Think outside the dilemma’s box

Ethical dilemmas, by definition, force us to choose between two or more bad options. But often, the choice appears impossible only because we have failed to consider all possible options.

Consider the dilemma of whether an autonomous car should swerve to kill one pedestrian in order to save five. As I have argued earlier, we can benefit from suspending disbelief and forcing ourselves to consider the cartoonish simplification of the scenario. But we can also benefit from considering radically different ways to subvert the scenario altogether.

For example, we might consider new technologies that might enhance safety, akin to seatbelts and the airbags, which radically enhance passenger safety. Then, crashing the car into a tree no longer sacrifices the passenger. Boom! No more dilemma.

We may also think of ways to re-engineer the urban environment, by giving cars their own dedicated and fully enclosed tracks, so that humans can walk around without ever being hit by a car.

This ability to think outside of the dilemma’s box can be helpful not just for autonomous vehicles, but more broadly. Consider an AI system for predicting academic success for the purpose of deciding on college admission. Suppose the algorithm designers face a dilemma: They can make the algorithm as accurate as possible, but then the algorithm will miss some rare but important cases, including the next Einstein. Alternatively, the algorithm designers may reduce accuracy, allowing the next Einstein to get in, but at the expense of admitting many weak students who wouldn’t graduate. While this looks like a dilemma, we can ask questions like: Is there a way to improve the educational experience so that those weak students also succeed? Why are they weak in the first place? Is there a problem in the highschool system that we can fix?

Whenever AI presents us with a dilemma, we should see it as an invitation to re-imagine the entire situation, for the possibility of eliminating the dilemma altogether.