Judgement day may have just taken a step closer, for killer robots at least. Amidst concern about the deployment of intelligent robots on the battlefield, governments have agreed to look more closely at the issues that these weapons raise, the first step towards an outright ban before they’ve even been built.
In November, the governments that are part of the Convention on Certain Conventional Weapons (CCW) agreed to meet in Geneva next year to discuss the issues related to so-called “lethal autonomous weapons systems,” or what campaigners have dubbed “killer robots.”
For the military, war robots can have many advantages: They don’t need food or pay, they don’t get tired or need to sleep, they follow orders automatically, and they don’t feel fear, anger, or pain. And, few back home would mourn if robot soldiers were destroyed on the battlefield, either.
There are already plenty of examples of how technology has changed warfare from David’s Sling to the invention of the tank. The most recent and controversial is the rise of drone warfare. But even these aircraft have a pilot who flies it by remote control, and it is the humans who make the decisions about which targets to pick and when to fire a missile.
But what concerns many experts is the potential next generation of robotic weapons: ones that make their own decisions about who to target and who to kill.
Banning killer robots
“The decision to begin international discussions next year is a major leap forward for efforts to ban killer robots pre-emptively,” said Steve Goose, arms director at Human Rights Watch. “Governments have recognised that fully autonomous weapons raise serious legal and ethical concerns, and that urgent action is needed.”
While fully autonomous robot weapons might not be deployed for two or three decades, the International Committee for Robot Arms Control (ICRAC), an international group of academics and experts concerned about the implications of a robot arms race, argues a prohibition on the development and deployment of autonomous weapons systems is the correct approach. “Machines should not be allowed to make the decision to kill people,” it states.
While no autonomous weapons have been built yet, it’s not a theoretical concern, either. Late last year, the U.S. Department of Defense (DoD) released its policy around how autonomous weapons should be used if they were to be deployed in the battlefield. The policy limits how they should operate, but definitely doesn’t ban them.
For example, the DoD guidelines state, “Autonomous and semi-autonomous weapon systems shall be designed to allow commanders and operators to exercise appropriate levels of human judgment over the use of force,” and requires that systems “are sufficiently robust to minimize failures that could lead to unintended engagements or to loss of control of the system to unauthorized parties.”
The guidelines do however seem to exclude weapons powered by artificial intelligence (AI) from explicitly targeting humans: “Human-supervised autonomous weapon systems may be used to select and engage targets, with the exception of selecting humans as targets, for local defense to intercept attempted time-critical or saturation attacks.”
In contrast, the UK says it has no plans to develop fully autonomous weapons. Foreign Office minister Alistair Burt told Parliament earlier this year that the UK armed forces are clear that “the operation of our weapons will always be under human control as an absolute guarantee of human oversight and authority and of accountability for weapons usage,” but then qualifies that slightly: “The UK has unilaterally decided to put in place a restrictive policy whereby we have no plans at present to develop lethal autonomous robotics, but we do not intend to formalise that in a national moratorium.”
Noel Sharkey is chairman of ICRAC and professor of AI and robotics at the University of Sheffield in the UK. When he started reading about military plans around autonomous weapons he was shocked because “There seemed to be a complete overestimation of the technology. It was more like a sci-fi interpretation of the technology.”
Of ICRAC’s intentions he says the campaign is not against autonomous robots. “My vacuum cleaner is an autonomous robot, and I’ve worked for 30 years developing autonomous robots.” What it wants is a ban on what it calls the “kill function.” An autonomous weapon is one that, once launched, can select its own targets and engage them, Sharkey says. “Engage them means kill them. So it’s the idea of the machine selecting its own targets that’s the problem for us.”
For Sharkey robot soldiers can’t comply with the basic rules of war. They can’t distinguish between a combatant or a civilian or between a wounded soldier and a legitimate target. “There are no AI robotic systems capable of doing that at all,” he argues, pointing to one UK-built system that can tell the difference between a human and a car “but has problems with a dancing bear or a dog on its hind legs.”
A robot weapons system won’t be able to judge proportionality either, he argues; that is, judge whether civilian losses are acceptable and in proportion to the military advantage gained by an attack. “How’s a robot going to know that? PhDs are written on military advantage. It’s very contextual. You require a very experienced commander in the field on the ground who makes that judgment,” he said.
But one of the biggest issues is accountability, Sharkey said. A robot can’t be blamed if a military operation goes wrong, and that’s what really worries the military commanders that he speaks to: They are the ones who would be held accountable for launching the attack.
“But it wouldn’t be fair because these things can crash at any time, they can be spoofed, they can be hacked, they can get tackled in the industrial supply chain, they can take a bullet through the computer, human error in coding, you can have sensor problems, and who is responsible? Is it the manufacturers, the software engineers, the engineers, or is it the commander? In war, you need to know, if there’s a mishap, who’s responsible.”
Sharkey’s concern is that the weapons will be rolled out gradually despite the limitations in the technology. “The technology itself is just not fit for purpose and it’s not going to be fit for purpose by the time these things are deployed.”
As the battlefield adapts to the use of increasingly high tech weapons, the use of autonomous robots become more likely. If an enemy can render drones useless by blocking their communications (a likely consequence of their increased usage) then an autonomous drone which can simply continue with its mission without calling home is a useful addition. Similarly, because it takes (roughly) one-and-a-half seconds for a movement on a remote pilot’s joystick to have an effect on a drone, they would be slower to respond than autonomous aircraft if attacked, which is another good reason to make them self-governing.
The ICRAC campaign hopes to use the decision by the Convention on Conventional Weapons to look at autonomous weapons as a first step towards a ban, using the same strategy that lead to a pre-emptive ban on blinding laser weapons.
One reason for the unreasonable level of expectation around autonomous weapons is the belief that AI is far more capable than it really is, or what Sharkey describes as the “cultural myth of artificial intelligence that has come out of science fiction.” Researchers working in the field assert that AI is working on projects that are far more mundane (if useful) than building thinking humanoid robots.
“Every decade, within 20 years we are going to have sentient robots and there is always somebody saying it, but if you look at the people on the ground working [on AI] they don’t say this. They get on with the work. AI is mostly a practical subject developing things that you don’t even know are AI — in your phone, in your car, that’s the way we work.”
And even if, at some point in the far future, AI matures to the point at which a computer system can abide by the rules of war, the fundamental moral questions will still apply. Sharkey said, “You’ve still got the problems of accountability and people will have to decide is this morally what we want to have, a machine making that decision to kill a human.”
The android rules
Discussing whether robots should be allowed to kill – especially when killer robots don’t exist – might seem to be a slightly arcane and obscure debate to be having. But robots (and artificial intelligence) are playing ever-larger roles in society and we are figuring out piecemeal what is acceptable and what isn’t.
What we have been doing so far is building rules for specific situations, such as the DoD policy on autonomous weapons systems. Another less dramatic example is the recent move by some US states to pass legislation to allow autonomous cars to drive on the road. We’re gradually building a set of rules for autonomous robots in specific situations but rarely looking at the big picture.
However, there have been attempts to create a set of rules, a moral framework, to govern AI and robots. Certainly the most famous attempt to create a set of laws for robots to date is Isaac Asimov’s three laws of robotics which, since they were first defined in 1942, have offered – at least in fiction – a moral framework for how robots should behave.
Asimov’s three laws state:
1. A robot may not injure a human being or, through inaction, allow a human being to come to harm.
2. A robot must obey the orders given to it by human beings, except where such orders would conflict with the First Law.
3. A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
Robotics and AI haven’t come anywhere close to being able to build robots that would be able to comprehend or abide by these or any other sophisticated rules. A robot vacuum cleaner doesn’t need this level of moral complexity.
“People think about Asimov’s laws, but they were set up to point out how a simple ethical system doesn’t work. If you read the short stories, every single one is about a failure, and they are totally impractical,” said Dr. Joanna Bryson of the University of Bath.
Bryson emphasises that robots and AI need to be considered as the latest set tools – extremely sophisticated tools, but no more than that. She argues that AI should be seen as a tool that extends human intelligence in the same way that writing did by allowing humans to take memory out of their heads and put it into a book. “We’ve been changing our world with things like artificial intelligence for thousands of years,” she says. “What’s happening now is we’re doing it faster.”
But for Bryson, regardless of how autonomous or intelligent an android is, because it is a tool, it’s not the robots that need the rules – it’s us. “They have to be inside our moral framework. They won’t have their own moral framework. We have to make the choice so that robots are positioned within our moral framework so that they don’t damage the rest of the life on the planet.”
The UK’s Engineering and Physical Sciences Research Council (EPSRC) is one of the few organisations that has tried to create a set of practical rules for robots, and it quickly realised that laws for robots weren’t what is needed right now.
Its Principles of Robotics notes: “Asimov’s laws are inappropriate because they try to insist that robots behave in certain ways, as if they were people, when in real life, it is the humans who design and use the robots who must be the actual subjects of any law. As we consider the ethical implications of having robots in our society, it becomes obvious that robots themselves are not where responsibility lies.”
As such, the set of principles the EPSRC experts – including Dr. Bryson – outlined were for the designers, builders, and users of robots, not for the robots themselves.
For example, the five principles include: “Humans, not robots, are responsible agents. Robots should be designed; operated as far as is practicable to comply with existing laws and fundamental rights and freedoms, including privacy.”
Dr. Kathleen Richardson of University College London (UCL) also argues that we don’t need new rules for robots beyond the ones we have in place to protect us from other types of machines, even if they are used on the battlefield.
“Naturally, a remote killing machine will raise a new set of issues in relation to the human relationship with violence. In such a case, one might need to know that that machine would kill the ‘right’ target…but once again this has got nothing to with something called ‘robot ethics’ but human ethics,” she said.
The robots we are currently building are not like the thinking machines we find in fiction, she argues, and so the important issues are more about standard health and safety – that we don’t build machines that accidentally fall on you – rather than helping them to distinguish between right and wrong.
“Robots made by scientists are like automaton,” she said. “It is important to think about entities that we create and to ensure humans can interact with them safely. But there are no ‘special’ guidelines that need to be created for robots, the mechanical robots that are imagined to require ethics in these discussions do not exist and are not likely to exist,” she said.
So while we might need rules to make sure a bipedal robot can operate safely in a home, these are practical considerations alone, the ones you’d require from any consumer electronics in the home.
“Ethics on the other hand implies something well beyond this,” she says. “It implies a different set of categorical notions need to be implemented in relation to robotic machines as special kinds of entities.”
Exploitive, loving robots
Indeed, while few of us (hopefully) are likely to encounter a killer robot, with aging populations use of human-like robots for care may become more important, and this could be a bigger long-term issue. Rather than feeling too much fear of robots, we may become emotionally dependent, and feel too much love.
Another of the EPSRC guidelines (again, one of the few sets of guidelines in this area that exist) states: “Robots are manufactured artifacts. They should not be designed in a deceptive way to exploit vulnerable users; instead their machine nature should be transparent.” It warns that unscrupulous manufacturers might use the illusion of emotions in a robot pet or companion to find a way to charge more money.
Perhaps one of the biggest risks we face is that, by giving robots the illusion of emotions and investing them with the apparent need for a moral framework to guide them, we risk raising them to the level of humans – and making it easier to ignore our fellow humans as a result.
UCL’s Richardson argues that robotic scientists are right to think about the implications but that the debate risks missing a bigger issue: why are we using these devices in the first place, particularly in terms of social care.
The real responsibility
Killer robots and power-mad AIs are the staples of cheap science fiction, but fixating on these types of threats allow us to avoid the complexities of our own mundane realities. It is a reflection – or indictment – of our society that the roles we are finding for robots – fighting our wars and looking after the elderly – are the roles that we are reluctant to fill ourselves.
Putting robots into these roles may fix part of the problem, but doesn’t address the underlying issues, and even worse perhaps allows us as a society to ignore them. Robots fighting our battles make war easier, and robots looking after the elderly makes it easier to ignore our obligations and societal strain that comes with an aging population. As such, worrying about the moral framework for androids is often a distraction from our own ethical failings.