No, Really, How Do We Keep Robots From Destroying Humans?
In the demilitarized zone dividing North and South Korea, SGR-1 robots are on patrol, equipped with cameras and radar to detect intruders as well as speakers to warn them off. If that fails, they also carry machine guns and grenade launchers.
In the U.S., the Home Exploring Robotic Butler can retrieve a book from a shelf, a meal from a microwave or a drink from the kitchen. It can even separate an Oreo cookie.
In Japan, a seal-like robot called Paro provides companionship for seniors -- and seems to ease the effects of dementia.
Over the next few decades, robots will become part of everyday life. But as they grow more sophisticated and autonomous, they’ll confront situations of cultural and moral ambiguity that won’t be easily resolved -- situations that people, over the millennia, have learned to navigate but that resist codification that machines can easily understand. This means robots, from the battlefield to the nursing home, will require more advanced ethical-decision-making abilities. And humans will need to think through what should happen when they cause harm.
Three challenges in particular need to be explored.
The first and most immediate is in warfare. Some 40 countries are at work on weapons and military equipment that have some degree of autonomy -- from drones to the Legged Squad Support System -- as are plenty of private companies. The appeal seems obvious. Unmanned weapons don’t need health insurance or food or hazard pay. They never lash out in anger, disobey an order or suffer from post-traumatic stress. And, at least in theory, they could save the lives of many human soldiers.
At the same time, fully autonomous weapons -- those that are capable of making their own decisions about whether to attack or kill, without a human “in the loop” -- make us deeply uneasy. Only 26 percent of respondents to a survey by the University of Massachusetts Amherst favored their use. Human Rights Watch has asserted that they violate international humanitarian law and should be banned.
Yet bans on specific weapons systems -- such as military airplanes or submarines -- have almost never been effective in the past. Instead, legal prohibitions and ethical norms have arisen that effectively limit their use. So a more promising approach might be to adapt existing international law to govern autonomous technology -- for instance, by requiring that such weapons, like all others, can’t be used indiscriminately or cause unnecessary suffering. It may turn out that robotic weapons are actually better at meeting those requirements than humans are.
Outside of warfare, robots will confront situations with no obvious moral resolution. Suppose one is assigned to make sure your grandmother takes her pills. Then one day she refuses. A host of quandaries -- from medical ethics to privacy rights to cultural mores -- arise that would be hard enough for a person to resolve.
Situations like this will demand that engineers cooperate closely with ethicists in designing software, and they will require much more sophisticated rules than the “Three Laws of Robotics” made famous by Isaac Asimov. Ronald Arkin, of the Georgia Institute of Technology, has done pioneering work on creating “ethical governors” for robots. But we’re a long way from a satisfactory simulation of morality. Technology companies would be wise to boost their investment in such research for the sake of both profits and liability.
Which leads to our third concern. When a robot with some degree of autonomy unexpectedly harms someone or something -- as will surely happen -- who’s liable? The manufacturer? The software designer? The owner? To some extent, the existing tort system can be adapted to help sort things out. One insightful study suggests a hybrid liability system in which robots would be treated as domesticated animals in cases where their owners or victims acted negligently, and as commercial products in cases where the machines were defective.
As robots grow more sophisticated, and people more reliant on them, another model to consider is the National Vaccine Injury Compensation Program. The government could establish a fund, paid for by a tax on autonomous machines, to compensate accident victims, thus ensuring that manufacturers won’t fear rare but very costly lawsuits -- and won’t be discouraged from inventing new robots -- provided they follow best practices in designing and marketing them.
In warfare as in ordinary life, when robots cause harm, it will be critical that the lines of accountability are clear. Scenarios in which intelligent machines grow self-aware enough to enslave humanity -- evoked so vividly in movies such as “The Terminator” -- aren’t plausible. Yet they express an important human intuition: There is a danger in ceding too much control to technology.
This intuition can help guide us into the new robotic era. But we shouldn’t let it unduly impede a promising new field. With the right rules in place, the rise of the machines should be nothing to fear.
To contact the Bloomberg View editorial board: firstname.lastname@example.org.