NEXT week, a in Geneva will discuss autonomous armed robots. Unlike existing military drones, which are controlled remotely, these new machines would identify and attack targets without human intervention. Groups including the hope the meeting will lead to an international ban.
But while fiction is littered with cautionary tales of what happens when you put guns in the cold, metallic hands of a machine, the situation may not be as simple as “human good, robots bad”.
To understand why, we should look at what people are saying about the ethics of driverless cars, which advocates see as a way of reducing accidents. If your life is safer in the hands of a robot car than a human driver, might the same apply for military hardware?
Advertisement
Clearly, replacing a human combatant with a robot one is safer for that individual, but armed robots could also reduce civilian casualties. For example, a squad that has to clear a building must make a split-second decision about whether the occupant of a room is an armed insurgent or an innocent civilian – any hesitation could get them killed. A robot can wait for confirmation, when the enemy starts firing.
The same principle applies to air strikes. An autonomous system can make several runs over a target to confirm it is really an enemy outpost, but a pilot can risk only one pass. In both cases the only downside is the loss of machines due to excessive caution, not casualties.
Human rights groups now see the use of precision-guided weapons as essential to avoiding civilian casualties; crude unguided weapons are considered reckless and irresponsible. Is there a similar moral justification for developing armed robots?
“We cannot simply accept the current status quo with respect to non-combatant deaths,” says , a roboticist at the Georgia Institute of Technology, who campaigns for regulating, not banning, armed robots. “We should aim to do better.”
Most weapon systems operate with a person in the loop, meaning the ultimate decision to kill is not made autonomously, but some weapons have more leeway, such as the used by the Royal Air Force. The operator can define a box on the map, and fire a salvo of missiles that search the area for tanks, striking when they spot one or detonating safely otherwise.
“Robots can’t understand the complexities of a situation or the nuances of human behaviour”
Human judgement in similar situations is all too fallible. In one lethal “friendly fire” incident in 2003, two US pilots , despite orange panels marking them as friendly and the fact that they were in a “no engagement” zone.
Can robots therefore make better decisions than humans? Arkin argues that machines would not suffer from the cognitive error of ““, which was blamed for a tragedy involving a civilian Iranian airliner in 1988. As the plane flew towards a US warship, radar operators reported that it was descending for an attack run and shot it down, . Later analysis of radar recordings showed the plane had been climbing. The radar operators had imagined that they were under attack and unconsciously made the evidence fit that scenario, ignoring alternative interpretations.
But , a lecturer at Harvard Law School and a researcher at Human Rights Watch, points out the limitations of artificial intelligence, particularly in complex war zones like Iraq. “Robots wouldn’t have the capacity to fully understand the complexities of a situation or interpret the nuances of human behaviour,” she says. “They would find it difficult to distinguish soldiers from civilians.”
Send in the drones
But we may have to let robots off the leash sooner than we might like. New tactical jamming systems with names like DroneDefender and DroneShield, are designed to break the radio link between operator and drone, and are appearing more frequently on the battlefield. US drones donated to Ukraine . If remote control isn’t possible, uncrewed systems will need to be fully autonomous. (See “Russia’s robot army“)
Arkin has a better solution. Rather than simply leaving robots to follow orders as they see fit, armed bots could be equipped with “” software that takes into account international law, he says.
For example, if a potential target cannot be confirmed as valid, the ethical governor may demand that the robot take further actions, such as moving in for a closer look. It could also veto an attack altogether, if say, civilians are in the area.
Arkin from Iraq showing a helicopter crew complying with orders to shoot and kill a wounded insurgent. A robot with an ethical governor would decline that order, citing its legal reasons, he says. Unlike a human, a robot would not feel the need to react illegally out of a desire for vengeance, or from fear of authority. Killer robots might be less of a risk than killer humans, with their long history of committing war atrocities. As with driverless cars, machines may better uphold laws humans are inclined to break.
The ethical governor concept is in its , though Arkin thinks future developments in AI will make it feasible. For some though, the issue is less a technical one than a fundamental question of ethics.
“Fully autonomous weapons raise a host of other concerns that do not relate to the ability to distinguish between civilian and military objects,” says Docherty. “Delegating life-and-death decisions to machines crosses a moral threshold.”
But that threshold has already been crossed by cars on the market, such as Tesla vehicles that can change lanes without human intervention. , a philosopher at California Polytechnic State University, says this exposes a double standard in those wanting to ban killer robots.
“I don’t think I’ve heard a single person call for meaningful human control in autonomous cars,” says Lin. “If it is a real ethical requirement for killer robots, then it would logically extend to other machines that can kill, such as robot cars.”
Clearly, killer robots are designed to be lethal, while a robot car might only kill by accident. But Lin argues this distinction shouldn’t matter.
“Robot cars may accidentally kill more people per year than military robots, given that robot cars would be far more prevalent,” he says. If we allow driverless cars to make life-or-death decisions, then we should also accept killer robots – or reject both, he says.
at the Massachusetts Institute of Technology studies the ethical issues associated with driverless cars, and also sees direct parallels. She suggests an ethical policy for autonomous weapons should follow the pattern set by her work on self-driving vehicles.
“We start with a psychological question,” says Levine. “How do people perceive autonomous weapons and how do they attribute blame and causal and moral responsibility to them?”
The answers should inform our policy, she says. Her work on driverless cars highlighted inconsistent beliefs: we want cars in general to minimise harm, but we want our car to protect us rather than others. Public attitudes to killer robots may turn out to be just as paradoxical.
“If we accept driverless cars making life-or-death decisions, then we should accept killer robots”
As such, we might want to rethink a ban on killer robots. “We should be in a hurry to regulate them, but a ban is premature,” says Arkin. Arming machines could mean less risk to soldiers, and less chance of killing innocent civilians, but only if we develop the technology for adequate safeguards. If an outright ban is implemented, that won’t happen – and civilian deaths will continue as before.
Russia’s robot army
While Russia has lagged in aerial drone technology compared with the US military, it is hoping to win the ground war. An initiative kicked off in 2013 has seen armed ground robots proliferate, from small bots that can operate indoors, up to , some of which can .
For example, the armed with cannon or anti-tank missiles; 22 have been delivered to the Russian army and they should enter service next year. Uran-9 can be controlled remotely, but also has a to detect, identify and shoot targets on its own.
Vadim Kozyulin of the PIR Centre for Policy Studies, a Russian security think tank, says the profusion of armed robots is slow, as many get no further than the proposal stage. The Russian military has not ordered autonomous killing machines as such, says Kozyulin, but semi-autonomous machines with the capacity to operate autonomously if they lose contact with human controllers as a result of radio jamming countermeasures by the enemy.
This need for autonomy may mean that, as with landmines and cluster bombs, Russia will not sign up for a ban that could reduce its military capability (see main story).
“At the moment Russia is reluctant to support the ban. They are evaluating the situation and this position might change,” says Kozyulin.
This article appeared in print under the headline “Lethal logic”
