Ban Killer Robots Before It’s Too Late

Professor Noel Sharkey

Tune into “The Bruce Montalvo Show” as Bruce interviews Professor of Artificial Intelligence and Robotics Noel Sharkey

https://www.spreaker.com/embed/player/standard?episode_id=6506133&autoplay=true As wars become increasingly automated, we must ask ourselves how far we want to delegate responsibility to machines. Where do we want to draw the line?

Weapons systems have been evolving for millennia and there have always been attempts to resist them. But does that mean that we should just sit back and accept our fate and hand over the ultimate responsibility for killing to machines?

Over the last few months there has been an increasing debate about the use of fully autonomous robot weapons: armed robots that once launched can select their own targets and kill them without further human intervention.

Some have argued that robots could be more accurate on the battlefield than human soldiers and save more civilian lives. But this is speculation based on assumptions about future developments of computer hardware and software. It is no more than “hopeware” — since the 1950s, Artificial Intelligence has moved at a snail’s pace compared to what proponents have predicted.

At this point, we cannot rely on machines having the independent facility to conform to international law. Current sensing systems are not up to the task. And even if machines had adequate sensing mechanisms they would still be missing the vital components of battlefield awareness and common sense reasoning to make decisions about who and when it is appropriate to kill.

Robots do not have the agency to decide if striking a target is proportional to the expected military advantage. There is no metric for this. Much of war is art and not science. A military commander must make a qualitative decision about the number of civilian lives that can be risked for a particular military objective. And that commander can be held accountable.

A robot doesn’t have the moral agency to be held accountable. Some would argue that the commander who sends a robot on a mission would be responsible (last point of contact). But that could be unfair since it could be the fault of the mission programmer, the manufacturer or one of dozens of little companies providing components. Maybe it should be the senior staff or policy makers who had the idea to use robots. Or the device could have been tampered with in the industrial supply chain or even damaged in action. Forensics are extremely difficult with such complex devices.

Yet a recent U.S. DoD directive (November 2012) gives a green light to research and development of autonomous weapons systems while presenting a cautious route to their deployment.

This is borne from a culmination of U.S. military road maps dating back to 2002 and it is a bad move. It sends the wrong message to other nations. As the most militarily advanced nation on the planet, the U.S. has the opportunity to take the take the lead in halting these developments.

Thanks to the U.S.’s use of drones, more than 70 other countries have acquired the technology in a new arms race. It is simply blinkered to think that they will not follow suit with autonomous weapons. Is anyone thinking about how an adaptive enemy will exploit the weaknesses of robot weapons with spoofing, hacking or misdirection?

Is anyone considering how unknown computer programs will interact when swarms of robots meet? Is anyone considering how autonomous weapons could destabilize world security and trigger unintentional wars?

In April this year in London, a group of prominent NGOs will launch a large civil society campaign to “Stop Killer Robots.” They are seeking a new legally binding preemptive international treaty to prohibit the development and deployment of fully autonomous robot weapons.

The aim is to stop these weapons getting into the arsenals of the world’s militaries while there is still an opportunity. Once there has been large national investments in the technology, it may be too late.