Killer Robots Aren’t Regulated. Yet.

I love that I can unlock my phone with my face, and that Google can predict what I’m thinking. And that Amazon knows exactly what I need. It’s great that I don’t have to hail a cab or go to the grocery store. Actually, I hope I never have to drive again or navigate or use cash or clean or cook or work or learn. But what if all this technology was trying to kill me? The same technology that is making your life easier is being weaponized. That feature that unlocks your phone with your face, here it is attached to a self-learning machine gun. It’s manufacturer, Kalashnikov, made this video to show the gun using object-recognition software to identify targets. They say it gets more accurate the more you use it. That drone advertised to get awesome snowboarding shots, here’s one that doesn’t require a pilot. This ad shows it with a high-explosive warhead. It hangs out in the sky, until it finds an enemy radar system, then crashes headfirst into it. Oh, and that driverless car you thought was so cool, well, here it is in tank form at a Russian arms fair. It’s called the T-14. Dmitry, here, says he sells them to the Russian government. That contract is part of a trend that’s changing the way wars are waged. Like all good stories, this one starts at a Russian arms fair. We’re a few hours outside of Moscow. Everyone from government officials to gun enthusiasts have come here to see the latest weapons. It’s a family affair. Buyers want to know how the 21st-century technology boom can give their armies a strategic advantage. They want to know: Can technology make war safer? But some fear giving weapons too much power because it brings us closer to machines that could go out and kill on their own. They say, we might not be able to control weapons like these, weapons loaded with artificial intelligence. “So artificial intelligence is a study of how to make machines behave intelligently, which means acting in a way that will achieve the objectives that they’ve been given. And recently, I’ve become concerned about the use of A.I. to kill people.” Stuart Russell. He was an early pioneer in artificial intelligence. He’s also been warning people about its potential danger for years. “So a killer robot is something that locates, selects and attacks human targets.” Stuart isn’t so worried about robots like this. We’re still pretty far from the “Terminator.” But Stuart says we’re not as far from something like this bee-sized drone. He imagined one, and made a movie that he hopes will freak you out. In Stuart’s movie, we see swarms of them armed with explosives set loose on their targets. “The main issue is you’re creating a class of weapons of mass destruction, which can kill millions of people, just like a nuclear weapon. But in fact, it’s much easier to build, much cheaper, much more scalable, in that you can use 1 or 10 or 100 or 1,000 or 10,000. Whereas with a nuclear weapon, it’s sort of all or nothing. It doesn’t destroy the city and the country that you’re attacking. It just kills all the people you want to kill, all males between 12 and 60 or all males wearing a yarmulke in Israel.” The weapon Stuart is describing is terrifying, if it works perfectly. With the current state of tech, many experts say it wouldn’t, but that could be even scarier. “The way we think about A.I. is we build a machine and we put the objective into the machine. And the machine pursues the objective. So you put in the objective of ‘find a cure for cancer as quickly as possible.’ Sounds great, right? O.K. Well, probably the fastest way to do that is to induce tumors in the entire human population, and then try millions of different treatments simultaneously. Then, that’s the quickest way to find a cure. That’s not what you meant, but that’s what you asked for. So we call this the King Midas Problem. King Midas said, ‘I want everything I touch to turn to gold.’ And he got his wish. And the, his food turned to gold, and his drink turned to gold and his family turned to gold. He died in misery and starvation. You know, this is a very old story. We are unable to correctly specify the objective.” Machines will always be limited by the minds of those who made them. We aren’t perfect. And neither is our A.I. Facial recognition software has had trouble with dark skin. Self-driving vehicles still need good weather and calm streets to work safely. We don’t know how long it will take for researchers to create weapons with that kind of flexibility. But behind closed doors, defense labs are working on it and they’re not working alone. “Militaries don’t have to invent A.I. It’s already being built — it’s being driven by major tech companies out in the commercial sector.” Paul Scharre, here, led a Department of Defense working group that helped establish D.O.D. policies on A.I. and weapons systems for the U.S. military. “The reality is all of the technology to put this together, to build weapons that can go out on the road, make their own decisions to kill human beings, exists today.” But it’s one thing to assemble a weapon in a lab, and another to have it work in any environment. And war is messy. “Machines are not really at a point today where they’re capable of flexibly adapting to novel situations. And that’s a major vulnerability in war.” Governments around the world see potential advantages in these weapons. After all, human soldiers — they get tired, emotional. They miss targets. Humans get traumatized. Machines do not. They can react at machine speed. If a missile was coming at you, how quickly would you want to know? Autonomous weapons could save lives. “The same technology that will help self-driving cars avoid pedestrians could be used to target civilians or avoid them, intentionally.” The problem is we’ve gotten this wrong before. “To really understand the growing trends of automation in weapons that have been growing for decades, you have to go all the way back to the American Civil War, to the Gatling Gun. How do I describe a Gatling Gun? Do I have to describe it? Could you guys show a picture of it? Richard Gatling was looking at all of the horrors that were coming back from the Civil War. And he wanted to find a way to make war more humane, to reduce the number of people that are needed on the battlefield. Wouldn’t that be amazing?” Four people operating Gatling’s gun could fire the equivalent of 100 soldiers. Far less people would be needed on the battlefield. It was the precursor to the machine gun. And it was born with the intention to save lives, at least for the army that had the gun. Of course — “The reality was far, far different. Gatling’s invention had the very opposite effect of what he intended. And then it magnified the killing and destruction on the battlefield, by orders of magnitude.” Gatling was wrong. Automating weapons didn’t save lives. And Dmitry, here, is saying something eerily familiar over 150 years later. And it wasn’t just Gatling. Revolutions of warfare have typically not gone well. “Before we ever developed usable biological weapons, the biologists said, stop doing this.” “All civilized countries today have given up chemical weapons as tools of warfare, but we see that they are still used by some rogue nations.” And then, there are nuclear weapons. Even with multiple treaties in place to police their use, the threat of nuclear obliteration remains a global anxiety. “Now, I am become death, a destroyer of worlds.” “Early in the war in Afghanistan, I was part of a Ranger sniper team that was sent out to the Afghanistan-Pakistan border to watch infiltration routes for foreign fighters coming across the border. We drove all night, and then began to hike up a steep rocky mountain under cover of darkness. From our position on the ridgeline, we could see for dozens of miles in every direction. And by the time the sun came up, we looked down at this compound beneath us. We were basically in someone’s backyard. We were certain that people would be coming to take a closer look at us. What I didn’t anticipate was that they sent a little girl to scout out our position. She wasn’t particularly sneaky, to be honest. She was reporting back our position, and probably how many of us there were. We watched her and she watched us. And then, she left. And pretty soon after, the Taliban fighters came. The gunfight that ensued brought out the whole village. And we knew that many, many more fighters would be coming before long. So we had to leave that position as we were compromised. Later on in the day, we talked about what would we do in a similar situation to that? You know, one of the things that never came up was the idea of shooting this little girl. But here’s the thing: She was a valid enemy combatant, and killing her would’ve been lawful. So if someone deployed an autonomous weapon, a robot that was designed to perfectly follow the laws of war, it would’ve killed this little girl in that situation. Now, I think that would’ve been wrong, maybe not legally, but morally. But how would a robot know the difference between what’s legal and what’s right?” With so much at stake, you’d think a debate would be happening. Well, there is. It’s just that technology moves at a different pace than diplomacy. “We will continue our discussion on Agenda Item 6A, characterisation of the systems under consideration in order to promote a common understanding on concepts and characteristics relevant to the objectives and purposes of the convention.” “One of the things I learned very quickly was that the official proceedings at the United Nations appear to be completely meaningless.” “Thank you, Mr. Chairperson —” “Support continued deliberations —” “We need a normative framework —” “Difference in interpretation —” “The importance of a multi-disciplinary —” “Down the rabbit hole of endless discussions on a subject of —” “Thank you, Mr. President. We are not in a position to make a declaration right now.” “Good morning.” “How are you?” “I’m good. How are you feeling?” “Oh, I’m fine, except for the governments, you know, their do-nothing attitude.” “We’d like to hear about that.” Jody Williams, here, won a Nobel Peace Prize for her work banning land mines. Now, she’s part of the Campaign to Stop Killer Robots. “Academics attacked the campaign in the beginning years, you know, saying robotics and A.I. are inevitable. Maybe they are, but applying them to killing human beings on their own is not inevitable, unless you do nothing. And we refuse to do nothing.” Today, the Campaign to Stop Killer Robots is staging a protest outside of the U.N. The group is made up of activists, nonprofits, and civil society organizations. The campaign’s goal? A ban on all weapons that can target and kill on their own. So far, 30 countries have joined them in supporting a ban, as well as 100 nongovernmental organizations, the European Parliament, 21 Nobel laureates, and leading scientists, like Stephen Hawking, Noam Chomsky and Elon Musk, as well as Stuart Russell, and more than 4,500 other A.I. researchers. Protester: “Everyone, you can get up now.” “Yay.” [cheering] Jody’s here with Mary Wareham. “So this is the sixth time that governments have come together since 2014 to talk about what they call lethal autonomous weapons systems.” We’re going to apologize in advance for the obtuse use of acronyms in this portion of the video. “We’re not trying to prohibit the use of artificial intelligence. You know, it can be beneficial to humanity. We’re pro-robots. We’re just anti-killer robots, anti-fully autonomous weapons.” “The C.C.W., the forum of the Convention for Conventional Weapons, — which actually has a name this long, and I can never remember it — operates by consensus. Which means you either negotiate the lowest common denominator, which means doing pretty much nothing, or if the entire room of diplomats wants to move forward with a treaty, for example, and one state says no, then it goes nowhere. And that’s really a dictatorship by one.” “Once a bullet leaves a gun, the rifleman ceases to have control over that bullet. Autonomy is a way of extending human control beyond the time a munition is deployed.” That’s the United States arguing that A.I. will save lives. And remember, without their support, any kind of regulation can’t move forward. “Using algorithm and software to determine and engage target reduces people to objects.” “In the U.S. perspective, there is nothing intrinsically valuable about manually operating a weapon system, as opposed to operating it with an autonomous function.” The United States isn’t alone. The countries working hardest to build autonomous weapons insist we can’t regulate what doesn’t exist yet. And at the same time, their militaries are developing these weapons right now. “The line between a semi-autonomous weapon that has a human in control, and a fully autonomous weapon could simply be a software patch.” “Indeed, some may say it is similar to trying to discuss the internet in the ’80s, ’70s, ’60s at this stage.” “It is not necessary or desirable at this time, to define laws.” “This so-called difficulty of definitions continues to be willful obfuscation.” The truth is, whether they exist or not just depends on how you define them. We don’t have weapons with artificial general intelligence or A.I. that’s as smart as humans. But we do already have weapons that can use A.I. to search, select and engage targets in specific situations. And the technology is only getting better. “So it could easily take another 10 years before they even agree on a definition of what an autonomous weapon is. And by that time, it will be too late. I think for some countries, that’s the point.” In the ongoing race between technology and diplomacy, technology is winning because in this race, the dual-use nature of technology means software being designed to make your life easier clearly has military applications. “The A.I. community, myself included, we were sort of asleep at the wheel for a long time. And we weren’t really thinking about the ways that it could be misused.” Whether we like it or not, we’ve entered the age of the algorithm. And A.I. is changing our place on the battlefield. Is it possible the next generation of soldiers won’t have to kill? “Look, it’s an appealing idea that, someday, robots will just fight other robots and no one will get hurt. I don’t think that’s realistic.” “Unfortunately, if it worked like that, we could just say, ‘Well, why don’t we just play baseball and decide who wins or Tiddlywinks?’ No country is going to surrender until the costs that they’ve incurred are unbearable. So even if your robots are defeated, the next stage is that their robots will start killing your people.” “Because the unfortunate reality is that wars will only end when people die.”