Army of None

by Paul Scharre

  Compact car–sized war bots aren’t necessarily unique to Russia, although the Russian military seems to have a casual attitude toward arming them not seen in Western nations. The Russian military isn’t stopping at midsize ground robots, though. Several Russian programs are pushing the boundaries of what is possible with robotic combat vehicles, building systems that could prove decisive in highly lethal tank-on-tank warfare.

  The Uran-9 looks like something straight out of a MechWarrior video game, where players pilot a giant robot warrior armed with rockets and cannons. The Uran-9 is fully uninhabited, although it is controlled by soldiers remotely from a nearby command vehicle. It is the size of a small armored personnel carrier, sports a 30 mm cannon, and has an elevated platform to launch antitank guided missiles. The elevated missile platform that gives the Uran-9 a distinctive sci-fi appearance. The missiles rest on two platforms on either side of the vehicle that, when raised, look like arms reaching into the sky. The elevated platform allows the robot to fire missiles while safely sitting behind cover, for example behind the protective slope of a hillside. In an online promotional video from the developer, Rosoboronexport, slo-mo shots of the Uran-9 firing antitank missiles are set to music reminiscent of a Tchaikovsky techno remix.

  The Uran-9 is a major step beyond smaller robotic platforms like the Platform-M and Wolf-2 not just because it’s larger, but because its larger size allows it to carry heavier weapons capable of taking on antitank missions. Whereas the assault rifle and grenade launcher on a Platform-M would do very little to a tank, the Uran-9’s antitank missiles would be potentially highly lethal. This makes the Uran-9 potentially a useful weapon in high-intensity combat against NATO forces on the plains of Europe. Uran-9s could hide behind hillsides or other protective cover and launch missiles against NATO tanks. The Uran-9 doesn’t have the armor or guns to stand toe-to-toe against a modern tank, but because it’s uninhabited, it doesn’t have to. The Uran-9 could be a successful ambush predator. Even if firing its missiles exposed its position and led it to be taken out by NATO forces, the exchange might still be a win if it took out a Western tank. Because there’s no one inside it and the Uran-9 is significantly smaller than a tank, and therefore presumably less expensive, Russia could field many of them on the battlefield. Just like many stings from a hornet can bring down a much larger animal, the Uran-9 could make the modern battlefield a deadly place for Western forces.

  Russia’s Vikhr “robot tank” has a similar capability. At 14 tons and lacking a main gun, it is significantly smaller and less lethal than a 50- to 70-ton main battle tank. Like the Uran-9, though, its 30 mm cannon and six antitank missiles show it is designed as a tank-killing ambush predator, not a tank-on-tank street fighter. The Vikhr is remote controlled, but news reports indicate it has the ability to “lock onto a target” and keep firing until the target is destroyed. While not the same as choosing its own target, tracking a moving target is doable today. In fact, tracking moving objects is as a standard feature on DJI’s base model Spark hobby drone, which retails for under $500.

  Taking the next step and allowing the Uran-9 or Vikhr to autonomously target tanks would take some additional work, but it would be more feasible than trying to accurately discriminate among human targets. With large cannons and treads, tanks are distinctive military vehicles not easily confused with civilian objects. Moreover, militaries may be more willing to risk civilian casualties or fratricide in the no-holds-barred arena of tank warfare, where armored divisions vie for dominance and the fate of nations is at stake. In videos of the Uran-9, human operators can be clearly seen controlling the vehicle, but the technology is available for Russia to authorize fully autonomous antitank engagements, if it chose to do so.

  Russia isn’t stopping at development of the Vikhr and Uran-9, however. It envisions even more advanced robotic systems that could not only ambush Western tanks, but stand with them toe-to-toe and win. Russia reportedly has plans to develop a fully robotic version of its next-generation T-14 Armata tank. The T-14 Armata, which reportedly entered production as of 2016, sports a bevy of new defensive features, including advanced armor, an active protection system to intercept incoming antitank missiles, and a robotic turret. The T-14 will be the first main battle tank to sport an uninhabited turret, which will afford the crew greater protection by sheltering them within the body of the vehicle. Making the entire tank uninhabited would be the next logical step in protection, enabling a crew to control the vehicle remotely. While current T-14s are human-inhabited, Russia has long-term plans to develop a fully robotic version. Vyacheslav Khalitov, deputy director general of UralVagonZavod, manufacturer of the T-14 Armata, has stated, “Quite possibly, future wars will be waged without human involvement. That is why we have made provisions for possible robotization of Armata.” He acknowledged that achieving the goal of full robotization would require more advanced AI that could “calculate the situation on the battlefield and, on this basis, to take the right decision.”

  In addition to pushing the boundaries on robots’ physical characteristics, the Russian military has signaled it intends to use cutting-edge AI to boost its robots’ decision-making. In July 2017, Russian arms manufacturer Kalashnikov stated that they would soon release “a fully automated combat module” based on neural networks. News reports indicate the neural networks would allow the combat module “to identify targets and make decisions.” As in other cases, it is difficult to independently evaluate these claims, but they signal a willingness to use artificial intelligence for autonomous targeting. Russian companies’ boasting of autonomous features has none of the hesitation or hedging that is often seen from American or British defense firms.

  Senior Russian military commanders have stated they intend to move toward fully robotic weapons. In a 2013 article on the future of warfare, Russian military chief of staff General Valery Gerasimov wrote:

  Another factor influencing the essence of modern means of armed conflict is the use of modern automated complexes of military equipment and research in the area of artificial intelligence. While today we have flying drones, tomorrow’s battlefields will be filled with walking, crawling, jumping, and flying robots. In the near future it is possible a fully robotized unit will be created, capable of independently conducting military operations.

  How shall we fight under such conditions? What forms and means should be used against a robotized enemy? What sort of robots do we need and how can they be developed? Already today our military minds must be thinking about these questions.

  This Russian interest in pursuing fully robotic units has not escaped notice in the West. In December 2015, Deputy Secretary of Defense Bob Work mentioned Gerasimov’s comments in a speech on the future of warfare. As Work has repeatedly noted, U.S. decisions may be shaped by those of Russia and other nations. This is the danger of an arms race in autonomy: that nations feel compelled to race forward and build autonomous weapons out of the fear that others are doing so, without pausing to weigh the risks of their actions.

  AN ARMS RACE IN AUTONOMOUS WEAPONS?

  If it is true, as some have suggested, that a dangerous arms race in autonomous weapons is under way, then it is a strange kind of race. Nations are pursuing autonomy in many aspects of weaponry but, with the exception of the Harpy, are still keeping humans in the loop for now. Some weapons like Brimstone use autonomy in novel ways, pushing the boundaries of what could be considered a semiautonomous weapon. DARPA’s CODE program appears to countenance moving to human-on-the-loop supervisory control for some types of targets, but there is no indication of full autonomy. Developers of the SGR-A1 gun and Taranis drone have suggested full autonomy could be a future option, although higher authorities immediately disputed the claim, saying that was not their intent.

  Rather than a full-on sprint to build autonomous weapons, it seems that many nations do not yet know whether they might want them in the future and are hedging their bets. One challenge in understanding the global landscape of lethal autonomy
is that the degree of transparency among nations differs greatly. While the official policies of the U.S. and UK governments leave room to develop autonomous weapons (although they express this differently with the United Kingdom calling them “automated weapons”) countries such as Russia don’t even have a public policy. Policy discussions may be happening in private in authoritarian regimes, but we don’t know what they are. Pressure from civil society for greater transparency differs greatly across countries. In 2016, the UK-based NGO Article 36, which has been a leading voice in shaping international discussions on autonomous weapons, wrote a policy brief critiquing the UK government’s stance on autonomous weapons. In the United States, Stuart Russell and a number of well-respected colleagues from the AI community have met with mid-level officials from across the U.S. government to discuss autonomous weapons. In authoritarian Russia, there are no equivalent civil society groups to pressure the government to be more transparent about its plans. As a result, scrutiny focuses on the most transparent countries—democratic nations who are responsive to elements of civil society and are generally more open about their weapons development. What goes on in authoritarian regimes is far murkier, but no less relevant to the future path of lethal autonomy.

  Looking across the global landscape of robotic systems, it’s clear that many nations are pursuing armed robots, including combat drones that would operate in contested air space. How much autonomy some weapon systems have is unclear, but there is nothing preventing countries from crossing the line to lethal autonomy in their next-generation missiles, combat drones, or ground robots. Next-generation robotic systems such as the Taranis may give countries that option, forcing uncomfortable conversations. Even if many countries would rather not move forward with autonomous weapons, it may only take one to start a cascade of others.

  With no autonomous smoking gun, it seems unnecessarily alarmist to declare that an autonomous weapons arms race is already under way, but we could very well be at the starting blocks. The technology to build autonomous weapons is widely available. Even non-state groups have armed robots. The only missing ingredient to turn a remotely controlled armed robot into an autonomous weapon is software. That software, it turns out, is pretty easy to come by.

  8

  GARAGE BOTS

  DIY KILLER ROBOTS

  A gunshot cuts through the low buzz of the drone’s rotors. The camera jerks backward from the recoil. The gun fires again. A small bit of flame darts out of the handgun attached to the homemade-looking drone. Red and yellow wires snake over the drone and into the gun’s firing mechanism, allowing the human controller to remotely pull the trigger.

  The controversial fifteen-second video clip released in the summer of 2015 was taken by a Connecticut teenager of a drone he armed himself. Law enforcement and the FAA investigated, but no laws were broken. The teenager used the drone on his family’s property in the New England woods. There are no laws against firing weapons from a drone, provided it’s done on private property. A few months later, for Thanksgiving, he posted a video of a flamethrower-armed drone roasting a turkey.

  Drones are not only in wide use by countries around the globe; they are readily purchased by anyone online. For under $500, one can buy a small quadcopter that can autonomously fly a route preprogrammed by GPS, track and follow moving objects, and sense and avoid obstacles in its path. Commercial drones are moving forward in leaps and bounds, with autonomous behavior improving in each generation.

  When I asked the Pentagon’s chief weapons buyer Frank Kendall what he feared, it wasn’t Russian war bots, it was cheap commercial drones. A world where everyone has access to autonomous weapons is a far different one than a world where only the most advanced militaries can build them. If autonomous weapons could be built by virtually anyone in their garage, bottling up the technology and enforcing a ban, as Stuart Russell and others have advocated, would be extremely difficult. I wanted to know, could someone leverage commercially available drones to make a do-it-yourself (DIY) autonomous weapon? How hard would it be?

  I was terrified by what I found.

  HUNTING TARGETS

  The quadcopter rose off the ground confidently, smoothly gaining altitude till it hovered around eye level. The engineer next to me tapped his tablet and the copter moved out, beginning its search of the house.

  I followed along behind the quadcopter, watching it navigate each room. It had no map, no preprogrammed set of instructions for where to go. The drone was told merely to search and report back, and so it did. As it moved through the house it scanned each room with a laser range-finding LIDAR sensor, building a map as it went. Transmitted via Wi-Fi, the map appeared on the engineer’s tablet.

  As the drone glided through the house, each time it came across a doorway it stopped, its LIDAR sensor probing the space beyond. The drone was programmed to explore unknown spaces until it had mapped everything. Only then would it finish its patrol and report back.

  I watched the drone pause in front of an open doorway. I imagined its sensors pinging the distant wall of the other room, its algorithms computing that there must be unexplored space beyond the opening. The drone hovered for a moment, then moved into the unknown room. A thought popped unbidden into my mind: it’s curious.

  It’s silly to impart such a human trait to a drone. Yet it comes so naturally to us, to imbue nonhuman objects with emotions, thoughts, and intentions. I was reminded of a small walking robot I had seen in a university lab years ago. The researchers taped a face to one end of the robot—nothing fancy, just slices of colored construction paper in the shape of eyes, a nose, and a mouth. I asked them why. Did it help them remember which direction was forward? No, they said. It just made them feel better to put a face on it. It made the robot seem more human, more like us. There’s something deep in human nature that wants to connect to another sentient entity, to know that it is like us. There’s something alien and chilling about entities that can move intelligently through the world and not feel any emotion or thought beyond their own programming. There is something predatory and remorseless about them, like a shark.

  I shook off the momentary feeling and reminded myself of what the technology was actually doing. The drone “felt” nothing. The computer controlling its actions would have identified that there was a gap where the LIDAR sensors could not reach and so, following its programming, directed the drone to enter the room.

  The technology was impressive. The company I was observing, Shield AI, was demonstrating fully autonomous indoor flight, an even more impressive feat than tracking a person and avoiding obstacles outdoors. Founded by brothers Ryan and Brandon Tseng, the former an engineer and the latter a former Navy SEAL, Shield AI has been pushing the boundaries of autonomy under a grant from the U.S. military. Shield’s goal is to field fully autonomous quadcopters that special operators can launch into an unknown building and have the drones work cooperatively to map the building on their own, sending back footage of the interior and potential objects of interest to the special operators waiting outside.

  Brandon described their goal as “highly autonomous swarms of robots that require minimal human input. That’s the end-state. We envision that the DoD will have ten times more robots on the battlefield than soldiers, protecting soldiers and innocent civilians.” Shield’s work is pushing the boundaries of what is possible today. All the pieces of the technology are falling into place. The quadcopter I witnessed was using LIDAR for navigation, but Shield’s engineers explained they had tested visual-aided navigation; they simply didn’t have it active that day.

  Visual-aided navigation is a critically important piece of technology that will allow drones to move autonomously through cluttered environments without the aid of GPS. Visual-aided navigation tracks how objects move through the camera’s field of view, a process called “optical flow.” By assessing optical flow, operating on the assumption that most of the environment is static and not moving, fixed objects moving through the camera’s field of vision can be
used as a reference point for the drone’s own movement. This can allow the drone to determine how it is moving within its environment without relying on GPS or other external navigation aids. Visual-aided navigation can complement other internal guidance mechanisms, such as inertial measurement units (IMU) that work like a drone’s “inner ear,” sensing changes in velocity. (Imagine sitting blindfolded in a car, feeling the motion of the car’s acceleration, braking, and turning.) When IMUs and visual-aided navigation are combined, they make an extremely powerful tool for determining a drone’s position, allowing the drone to accurately navigate through cluttered environments without GPS.

  Visual-aided navigation has been demonstrated in numerous laboratory settings and will no doubt trickle down to commercial quadcopters over time. There is certain to be a market for quadcopters that can autonomously navigate indoors, from filming children’s birthday parties to indoor drone racing. With visual-aided navigation and other features, drones and other robotic systems will increasingly be able to move intelligently through their environment. Shield AI, like many tech companies, was focused on near-term applications, but Brandon Tseng was bullish on the long-term potential of AI and autonomy. “Robotics and artificial intelligence are where the internet was in 1994,” he told me. “Robotics and AI are about to have a really transformative impact on the world. . . . Where we see the technology 10 to 15 years down the road? It is going to be mind-blowing, like a sci-fi movie.”

  Autonomous navigation is not the same as autonomous targeting, though. Drones that can maneuver and avoid obstacles on their own—indoors or outdoors—do not necessarily have the ability to identify and discriminate among the various objects in their surroundings. They simply avoid hitting anything at all. Searching for specific objects and targeting them for action—whether it’s taking photographs or something more nefarious—would require more intelligence.