The Ethics Of A Self-Driving Car

Patrick Lin’s WIRED article, “Here’s a Terrible Idea: Robot Cars with Adjustable Ethics Settings” raises a number of interesting points about the problems inherent in programming ethics into a computer system. A self-driving car has to make decisions. In the infamous trolley problem, inaction itself becomes a decision. There’s no way around decision-making; yet, short of endowing each and every car on the road with some kind of human-like self-awareness and consciousness (thereby defeating the purpose of having self-driving cars in the first place), the cars have no ability to make decisions outside of their programming. So, in a very real sense, the decisions a self-driving car makes in a difficult ethical scenario are the decisions of its creators, just deferred in space and time.

In normal automobile operation, the number of incidents where the driver has to choose between two options that both involve killing innocents is practically zero. So, while manufacturers may find clever solutions to these more extreme ethical dilemmas, and while lawyers and lawmakers may find a way to limit the carmakers’ liability, there are a number of ethical problems that self-driving cars may face that neither the manufacturers, the programmers, or the lawyers will consider. That is—while programmers may find ways to encode their explicit, idealized ethical rulesets into the cars, and even if these rulesets are (somehow) universally correct, and everyone agrees that its decisions are perfect—all humans have implicit biases, prejudices, and heuristics. These are unconscious, yet reflected in all of our actions. Troublingly, because they are unconscious, they are often also unacknowledged. Thus, in programming a self-driving vehicle, the implicit ethics of the designer may become explicit, encoded parts of the car’s operating