Plan to teach military robots the rules of war

by Tom Simonite for New Scientist

18 June 2009

Technology has always distanced the soldiers who use weapons from the people who get hit. But robotics engineer Ron Arkin at the Georgia Institute of Technology, Atlanta, is working to imagine wars in which weapons make their own decisions about wielding lethal force. He is particularly interested in how such machines might be programmed to act ethically, obeying the rules of engagement. Arkin has developed an "ethical governor", which aims to ensure that robot attack aircraft behave ethically in combat, and is demonstrating the system in simulations based on recent campaigns by US troops, using real maps from the Middle East.

In one scenario, modelled on a situation encountered by US forces in Afganistan in 2006, the drone identifies a group of Taliban soldiers inside a defined "kill zone". But the drone doesn't fire. Its maps indicate that the group is inside a cemetery, so opening fire would breach international law. In another scenario, the drone identifies an enemy vehicle convoy close to a hospital. Here the ethical governor only allows fire that will damage the vehicles without harming the hospital. Arkin has also built in a "guilt" system which, if a serious error is made, forces a drone to start behaving more cautiously. You can see videos of these simulations on Arkin's website. In developing the software, he drew on studies of military ethics, as well as discussions with military personnel, and says his aim is to reduce non-combatant casualties. One Vietnam veteran told him of soldiers shooting at anything that moved in some situations. "I can easily make a robot do that today, but instead we should be thinking about how to make them perform better than that," Arkin says.

Simulations are a powerful way to imagine one possible version of the future of combat, says Illah Nourbakhsh, a roboticist at Carnegie Mellon University, Pittsburgh, US. But they gloss over the complexities of getting robots to understand the world well enough to make such judgements, he says; something unlikely to be possible for decades. Arkin stresses that his research, funded by the US army, is not designed to develop prototypes for future battlefield use. "The most important outcome of my research is not the architecture, but the discussion that it stimulates." However, he maintains that the development of machines that decide how to use lethal force is inevitable, making it important that when such robots do arrive they can be trusted. "These ideas will not be used tomorrow, but in the war after next, and in very constrained situations."

Roboticist Noel Sharkey at Sheffield University, UK, campaigns for greater public discussion about the use of automating in war. "I agree with Ron that autonomous robot fighting machine look like an inevitability in the near future," he told New Scientist. Arkin's work shows the inadequacy of our existing technology at dealing with the complex moral environment of a battlefield, says Sharkey. "Robots don't get angry or seek revenge but they don't have sympathy or empathy either," he says. "Strict rules require an absolutist view of ethics, rather than a human understanding of different circumstances and their consequences." Yet in some circumstances, a strict rule-based approach is valuable. The Georgia Tech group has also made a system that advises a soldier of the ethical constraints on a mission as they program it into an autonomous drone. That kind of tool could see practical use much sooner, says Nourbakhsh: "Similar systems exist to help doctors understand the medical ethics of treatments." Arkin will discuss his latest results at the AUVSI Unmanned Systems conference in Washington, DC, in August.

[I don’t know if I’m thinking RoboCop here or maybe Ed-209. Clearly teaching machines ethics is decades away – at least – and may not be possible at all. I suppose that the military could programme machines with the Rules of Engagement, which is what some of the example above relate too rather than ethics per se, but teaching machines to fight ethically has the underwritten assumption that the particular war, or war in general, is an ethical undertaking in the first place. Would killer robots who accidently kill civilians simply shut down or blow an ethical fuse? How would a machine decide between limited ‘collateral damage’ and saving the lives of its fellow human soldiers? Do we have sophisticated enough programming languages to even code this kind of thing? It’s not exactly my area of expertise but I have serious doubts about it. What I think is a more fundamental question, however, is this: Do we want machines making life and death decisions for us? Are we ready to accept the implications of semi-autonomous machines the sole purpose of which is to kill people? Is that a future we really want our children to live in?]