by Paul Scharre
These examples shine a light on a common misperception about autonomous weapons, which is the notion that intelligence is what makes a weapon “autonomous.” How intelligent a system is and which tasks it performs autonomously are different dimensions. It is freedom, not intelligence, that defines an autonomous weapon. Greater intelligence can be added into weapons without changing their autonomy. To date, the target identification algorithms used in autonomous and semiautonomous weapons have been fairly simple. This has limited the usefulness of fully autonomous weapons, as militaries may not trust giving a weapon very much freedom if it isn’t very intelligent. As machine intelligence advances, however, autonomous targeting will become technically possible in a wider range of situations.
UNUSUAL CASES—MINES, ENCAPSULATED TORPEDO MINES, AND SENSOR FUZED WEAPON
There are a few unusual cases of weapons that blur the lines between semiautonomous and fully autonomous weapons: mines and the Sensor Fuzed Weapon deserve special mention.
Placed on land or at sea, mines wait for their target to approach, at which point the mine explodes. While mines are automatic devices that will detonate on their own once triggered, they have no freedom to maneuver and search for targets. They simply sit in place. (For the most part—some naval mines can drift with the current.) They also generally have very limited methods for “deciding” whether or not to fire. Mines typically have a simple method for sensing a target and, when the threshold for the sensor is reached, the mine explodes. (Some naval mines and antitank mines employ a counter so that they will let the first few targets pass unharmed before detonating against a ship or vehicle later in the convoy.) Mines deserve special mention because their freedom in time is virtually unbounded, however. Unless specifically designed to self-deactivate after a certain period of time, mines can lay in wait for years, sometimes remaining active long after a war has ended.
The fact that mines are often unbounded in time has had devastating humanitarian consequences. By the mid-1990s, an estimated more than 110 million land mines lay hidden in sixty-eight countries around the globe, accumulated from scores of conflicts. Land mines have killed thousands of civilians, many of them children, and maimed tens of thousands more, sparking the global movement to ban land mines that culminated in the Ottawa Treaty in 1997. Adopted by 162 nations, the Ottawa Treaty prohibits the production, stockpiling, transfer, or use of antipersonnel land mines. Antitank land mines and naval mines are still permitted.
Mines can sense and act on their own, but do not search for targets. Encapsulated torpedo mines are a special type of naval mine that acts more like an autonomous weapon, however. Rather than simply exploding once activated, encapsulated torpedo mines release a torpedo that homes in on the target. This gives encapsulated torpedo mines the freedom to engage targets over a much wider area than a traditional mine, much like a loitering munition. The U.S. Mk 60 CAPTOR encapsulated torpedo mine had a published range of 8,000 yards. By contrast, a ship would have to pass over a regular mine for it to detonate. Even though encapsulated torpedo mines are moored in place to the seabed, their ability to launch a torpedo to chase down targets gives them a much greater degree of autonomy in space than a traditional naval mine. As with loitering munitions, examples of encapsulated torpedo mines are rare. The U.S. CAPTOR mine was in service for throughout the 1980s and 1990s but has been retired. The only encapsulated torpedo mine still in service is the Russian PMK-2, used by Russia and China.
The Sensor Fuzed Weapon (SFW) is an air-delivered antitank weapon that defies categorization. Released from an aircraft, an SFW can destroy an entire column of enemy tanks within seconds. The SFW functions through a series of Rube Goldberg machine–like steps: First, the aircraft releases a bomb-shaped canister than glides toward the target area. As the canister approaches the target area, the outer casing releases, exposing ten submunitions which are ejected from the canister. Each submunition releases a drogue parachute slowing its descent. At a certain height above the ground, the submunition springs into action. It opens its outer case, exposing four internally held “skeets” which are then rotated out of the inner casing and exposed. The parachute releases and the submunition fires retrojets that cause it to climb in altitude while spinning furiously. The hockey-puck-shaped skeets are then released, flung outward violently from the force of the spinning. Each skeet carries onboard laser and infrared sensors that it uses to search for targets beneath it. Upon detecting a vehicle beneath it, the skeet fires an explosively formed penetrator—a metal slug—downward into the vehicle. The metal slug strikes the vehicle on top, where armored vehicles have the thinnest armor, destroying the vehicle. In this manner, a single SFW can take out a group of tanks or other armored vehicles simultaneously, with the skeets targeting each vehicle precisely.
Similar to the distinction between Harpy and HARM, the critical variable in the evaluating SFW’s autonomy is its freedom in time and space. While the weapon distributes forty skeets over several acres, the time the weapon can search for targets is minuscule. Each skeet can hover with its sensor active for only a few seconds before firing. Unlike the Harpy, the SFW cannot loiter for an extended period over hundreds of kilometers. The human launching the SFW must know that there is a group of tanks at a particular point in space and time. Like a homing munition, the SFW must be part of a wider weapon system that provides targeting data in order to be useful. The SFW is different than a traditional homing munition, because the SFW can hit multiple objects. This makes the SFW like a salvo of forty homing munitions launched at a tightly geographically clustered set of targets.
PUSHING “START”
Autonomous weapons are defined by the ability to complete the engagement cycle—searching for, deciding to engage, and engaging targets—on their own. Autonomous weapons, whether supervised or fully autonomous, are still built and put into operation by humans, though. Humans are involved in the broader process of designing, building, testing, and deploying weapons.
The fact that there are humans involved at some stage does not change the significance of a weapon that could complete engagements entirely on its own. Even the most highly autonomous system would still have been borne out of a process initiated by humans at some point. In the climactic scene of Terminator 3: Rise of the Machines, an Air Force general pushes the button to start Skynet. (Absurdly, this is done with an old “EXECUTE Y/N?” prompt like the kind used in MS-DOS in the 1980s.) From that point forward, Skynet embarks on its path to exterminate humanity, but at least at the beginning a human was in the loop. The question is not whether there was ever a human involved, but rather how much freedom the system has once it is activated.
WHY AREN’T THERE MORE AUTONOMOUS WEAPONS?
Automation has been used extensively in weapons around the world for decades, but the amount of freedom given to weapons has been, up to now, fairly limited. Homing munitions have seekers, but their ability to search for targets is narrowly constrained in time and space. Supervised autonomous weapons have only been used for limited defensive purposes. The technology to build simple fully autonomous loitering munitions like TASM and Harpy has existed for decades, yet there is only one example in use today.
Why aren’t there more fully autonomous weapons? Homing munitions and even semiautonomous loitering munitions are widely used, but militaries have not aggressively pursued fully autonomous loitering munitions. The U.S. experience with TASM may shed some light on why. TASM was in service in the U.S. Navy from 1982 to 1994, when it was retired. To understand better why TASM was taken out of service, I spoke with naval strategist Bryan McGrath.
McGrath, a retired Navy officer, is well known in Washington defense circles. He is a keen strategist and unabashed advocate of sea power who thinks deeply about the past, present, and future of naval warfare. McGrath is familiar with TASM and other anti-ship missiles such as the Harpoon, and was trained on TASM in the 1980s when it was in the fleet.
McGrath explained to me that TASM could outrange the ship’s o
wn sensors. That meant that initial targeting had to come from another sensor, such as a helicopter or maritime patrol aircraft that detected an enemy ship. The problem, as McGrath described it, was a “lack of confidence in how the targeting picture would change from the time you fired the missile until you got it downrange.” Because the target could move, unless there was an “active sensor” on the target, such as a helicopter with eyes on the target the whole time, the area of uncertainty of where the target was would grow over time.
The ability of the TASM to search for targets over a wide area mitigated, to some extent, this large area of uncertainty. If the target had moved, the TASM could simply fly a search pattern looking for it. But TASM didn’t have the ability to accurately discriminate between enemy ships and merchant vessels that just happened to be in its path. As the search area widened, the risk increased that the TASM might run across a merchant ship and strike it instead. In an all-out war with the Soviet Navy, that risk might be acceptable, but in any situations short of that, getting approval to shoot the TASM was unlikely. TASM was, according to McGrath, “a weapon we just didn’t want to fire.”
Another factor was that if a TASM was launched and there wasn’t a valid target within the search area of the weapon, the weapon would be wasted. McGrath would be loath to launch a weapon on scant evidence that there was a valid target in the search area. “I would want to know that there’s something there, even if there was some kind of end-game autonomy in place.” Why? “Because the weapons cost money,” he said, “and I don’t have a lot of them. And I may have to fight tomorrow.”
Modern missiles can cost upwards of a million dollars apiece. As a practical matter, militaries will want to know that there is, in fact, a valid enemy target in the area before using an expensive weapon. One of the reasons militaries have not used fully autonomous loitering munitions more may be the fact that the advantage they bring—the ability to launch a weapon without precise targeting data in advance—may not be of much value if the weapon is not reusable, since the weapon could be wasted.
FUTURE WEAPONS
The trend of creeping automation that began with Gatling’s gun will continue. Advances in artificial intelligence will enable smarter weapons, which will be capable of more autonomous operation. At the same time, another facet of the information revolution is greater networking. German U-boats couldn’t control the Wren torpedo once it was launched, not because they didn’t want to; they simply had no means to do so.
Modern munitions are increasingly networked to allow them to be controlled or retargeted after they’ve been launched. Wire-guided munitions have existed for decades, but are only feasible for short distances. Long-range weapons are now incorporating datalinks to allow them to be controlled via radio communication, even over satellites. The Block IV Tomahawk Land Attack Missile (TLAM-E, or Tactical Tomahawk) includes a two-way satellite communications link that allows the weapon to be retargeted in flight. The Harpy 2, or Harop, has a communications link that allows it to be operated in a human-in-the-loop mode so that the human operator can directly target the weapon.
When I asked McGrath what feature he would most desire in a future weapon, it wasn’t autonomy—it was a datalink. “You’ve got to talk to the missile,” he explained. “The missiles have to be part of a network.” Connecting the weapons to the network would allow you to send updates on the target while in flight. As a result, “confidence in employing that weapon would dramatically increase.”
A networked weapon is a far more valuable weapon than one that is on its own. By connecting a weapon to the network, the munition becomes part of a broader system and can harness sensor data from other ships, aircraft, or even satellites to assist its targeting. Additionally, the commander can keep control of the weapon while in flight, making it less likely to be wasted. One advantage to the networked Tactical Tomahawk, for example, is the ability for humans to use sensors on the missile to do battle damage assessment (BDA) of potential targets before striking. Without the ability to conduct BDA of the target, commanders might have to launch several Tomahawks at a target to ensure its destruction, since the first missile might not completely destroy the target. Onboard BDA allows the commander to look at the target after the first missile hits. If more strikes are needed, more missiles can be used. If not, then subsequent missiles can be diverted in flight to secondary targets.
Everything has a countermeasure, though, and increased networking runs counter to another trend in warfare, the rise of electronic attack. The more that militaries rely on the electromagnetic spectrum for communications and sensing targets, the more vital it will be to win the invisible electronic war of jamming, spoofing, and deception fought through the electromagnetic spectrum. In future wars between advanced militaries, communications in contested environments is by no means assured. Advanced militaries have ways of communicating that are resistant to jamming, but they are limited in range and bandwidth. When communications are denied, missiles or drones will be on their own, reliant on their onboard autonomy.
Due to their expensive cost, even highly advanced loitering munitions are likely to fall into the same trap as TASM, with commanders hesitant to fire them unless targets are clearly known. But drones change this equation. Drones can be launched, sent on patrol, and can return with their weapons unused if they do not find any targets. This simple feature—reusability—dramatically changes how a weapon could be used. Drones could be sent to search over a wide area in space and time to hunt for enemy targets. If none were found, the drone could return to base to hunt again another day.
More than ninety nations and non-state groups already have drones, and while most are unarmed surveillance drones, an increasing number are armed. At least sixteen countries already possess armed drones and another dozen or more nations are working on arming their drones. A handful of countries are even pursuing stealth combat drones specifically designed to operate in contested areas. For now, drones are used as part of traditional battle networks, with decision-making residing in the human controller. If communications links are intact, then countries can keep a human in the loop to authorize targets. If communications links are jammed, however, what will the drones be programmed to do? Will they return home? Will they carry out surveillance missions, taking pictures and reporting back to their human operators? Will the drones be authorized to strike fixed targets that have been preauthorized by humans, much like cruise missiles today? What if the drones run across emerging targets of opportunity that have not been authorized in advance by a human—will they be authorized to fire? What if the drones are fired upon? Will they be allowed to fire back? Will they be authorized to shoot first?
These are not hypothetical questions for the future. Engineers around the globe are programming the software for these drones today. In their hands, the future of autonomous weapons is being written.
PART II
Building the Terminator
4
THE FUTURE BEING BUILT TODAY
AUTONOMOUS MISSILES, DRONES, AND ROBOT SWARMS
Few actors loom larger in the robotics revolution than the U.S. Department of Defense. The United States spends 600 billion dollars annually on defense, more than the next seven countries combined. Despite this, U.S. defense leaders are concerned about the United States falling behind. In 2014, the United States launched a “Third Offset Strategy” to reinvigorate America’s military technological advantage. The name harkens back to the first and second “offset strategies” in the Cold War, where the U.S. military invested in nuclear weapons in the 1950s and later precision-guided weapons in the 1970s to offset the Soviet Union’s numerical advantages in Europe. The centerpiece of DoD’s Third Offset Strategy is robotics, autonomy, and human-machine teaming.
Many applications of military robotics and autonomy are noncontroversial, such as uninhabited logistics convoys, tanker aircraft, or reconnaissance drones. Autonomy is also increasing in weapon systems, though, with next-generation missiles and combat
aircraft pushing the boundaries of autonomy. A handful of experimental programs show how the U.S. military is thinking about the role of autonomy in weapons. Collectively, they are laying the foundations for the military of the future.
SALTY DOGS: THE X-47B DRONE
The X-47B experimental drone is one of the world’s most advanced aircraft. Only two have been ever built, named Salty Dog 501 and Salty Dog 502. With a sleek bat-winged shape that looks like something out of the 1980s sci-fi flick Flight of the Navigator, the X-47B practically screams “the future is here.” In their short life-span as experimental aircraft from 2011 to 2015, Salty Dog 501 and 502 repeatedly made aviation history. The X-47B was the first uninhabited (unmanned) aircraft to autonomously take off and land on an aircraft carrier and the first uninhabited aircraft to autonomously refuel from another plane while in flight. These are key milestones to enabling future carrier-based combat drones. However, the X-47B was not a combat aircraft. It was an experimental “X-plane,” a demonstration program designed to mature technologies for a follow-on aircraft. The focus of technology development was automating the physical movement of the aircraft—takeoff, landing, flight, and aerial refueling. The X-47B did not carry weapons or sensors that would permit it to make engagements.
The Navy has stated their first operational carrier-based drone will be the MQ-25 Stingray, a future aircraft that is still on the drawing board. While the specific design has yet to be determined, the MQ-25 is envisioned primarily as a tanker, ferrying fuel for manned combat aircraft such as the F-35 Joint Strike Fighter, with possibly a secondary role in reconnaissance. It is not envisioned as a combat aircraft. In fact, over the past decade the Navy has moved steadily away from any notion of uninhabited aircraft in combat roles.