Image of a military jet flying in the air

The XQ-58A Valkyrie: The Air Force’s pilotless experimental aircraft, which is run by artificial intelligence, completes a test flight. Courtesy U.S. Air Force (Valkyrie); Shutterstock.com (background)

Will A.I. Fight Our Battles?

The Pentagon is starting to embrace the potential of artificial intelligence. Is the world ready?

It’s powered into flight by a rocket engine. It can fly a distance equal to the width of China. It has a stealthy design and can carry missiles that hit enemy targets far beyond its visual range.

But what really distinguishes the Air Force’s pilotless XQ-58A Valkyrie experimental aircraft is that it’s run by artificial intelligence (A.I.). It’s at the forefront of efforts by the U.S. military to harness the capabilities of an emerging technology—one whose potential benefits are tempered by concerns about how much autonomy to grant to a lethal weapon.

The Valkyrie—essentially a next-generation drone—is a prototype of what the Air Force hopes can supplement its fleet of traditional fighter jets, giving human pilots highly capable robots to deploy in battle. Its mission is to use A.I. and its sensors to identify and evaluate enemy threats, and then—after getting human sign-off—to move in for the kill.

It’s powered into flight by a rocket engine. It can fly a distance equal to the width of China. It has a stealthy design. It can carry missiles that hit enemy targets far beyond its visual range.

But what really distinguishes the Air Force’s pilotless XQ-58A Valkyrie experimental aircraft is that it’s run by artificial intelligence (A.I.). It’s at the forefront of efforts by the U.S. military to harness the capabilities of an emerging technology. But the potential benefits are tempered by concerns about how much autonomy to grant to a lethal weapon.

The Valkyrie is a next-generation drone. It is a prototype of what the Air Force hopes can supplement its fleet of traditional fighter jets. It would give human pilots highly capable robots to deploy in battle. Its mission is to use A.I. and its sensors to identify and evaluate enemy threats.  Then, after getting human sign-off, it would move in for the kill.

The Pentagon is forced to confront the question of what role humans should play in conflicts waged with software that’s written to kill.

“It’s a very strange feeling,” says Major Ross Elder, a test pilot at Eglin Air Force Base on Florida’s Gulf Coast, as he prepares to fly his F-15 fighter alongside the Valkyrie as part of an exercise. “I’m flying off the wing of something that’s making its own decisions. And it’s not a human brain.”

The Valkyrie program provides a glimpse into how U.S. weapons and military strategies are being reshaped by rapid advances in technology. The possibility of building fleets of relatively inexpensive weapons that could be deployed in large numbers is allowing Pentagon officials to think in new ways about taking on enemy forces.

But it’s also forcing Pentagon officials to confront questions about what role humans should play in conflicts waged with software that’s written to kill.

“It’s a very strange feeling,” says Major Ross Elder, a test pilot at Eglin Air Force Base on Florida’s Gulf Coast. He is preparing to fly his F-15 fighter alongside the Valkyrie as part of an exercise. “I’m flying off the wing of something that’s making its own decisions. And it’s not a human brain.”

The Valkyrie program provides a glimpse into how U.S. weapons and military strategies are being reshaped by rapid advances in technology. It is possible to build fleets of relatively inexpensive weapons. These could then be deployed in large numbers. This is allowing Pentagon officials to think in new ways about taking on enemy forces.

But it’s also forcing Pentagon officials to confront questions about what role humans should play in conflicts using software that’s written to kill.

Peter Brauns/Alamy Stock Photo

F-15 fighter jets like these would fly alongside the A.I.-piloted drones.

A New Generation Air Force

After decades of building fewer and fewer increasingly expensive combat aircraft, the Air Force now has the smallest and oldest fleet in its history. That’s where the new generation of A.I. drones will come in. The Air Force is planning to build 1,000 to 2,000 of them for as little as $3 million apiece—a fraction of the cost of an advanced fighter.

As it begins to embrace A.I., the Air Force is confronting deep concerns about military use of artificial intelligence, including fear that the technology might turn against its human creators or more immediate misgivings about allowing algorithms to guide the use of lethal force.

“You’re stepping over a moral line by outsourcing killing to machines—by allowing computer sensors rather than humans to take human life,” says Mary Wareham, the advocacy director of the arms division of Human Rights Watch.

The U.S. has also already faced criticism for using human-piloted remote drones in war that killed innocent civilians—even without the incorporation of A.I. into the devices.

In August 2021, 10 civilians, including seven children, were mistakenly killed in a U.S. drone strike in Afghanistan before American troops withdrew from the region. The Pentagon called the killings a “tragic mistake.”

After decades of building fewer and fewer increasingly expensive combat aircraft, the Air Force now has the smallest and oldest fleet in its history. That’s where the new generation of A.I. drones will come in. The Air Force is planning to build 1,000 to 2,000 of them for as little as $3 million apiece. It is a fraction of the cost of an advanced fighter.

As it begins to embrace A.I., the Air Force is confronting deep concerns about military use of artificial intelligence. This includes the fear that the technology might turn against its human creators. There are also more immediate concerns about allowing algorithms to guide the use of lethal force.

“You’re stepping over a moral line by outsourcing killing to machines—by allowing computer sensors rather than humans to take human life,” says Mary Wareham, the advocacy director of the arms division of Human Rights Watch.

The U.S. has also already faced criticism for using human-piloted remote drones in war that killed innocent civilians. In August 2021, 10 civilians, including seven children, were mistakenly killed in a U.S. drone strike in Afghanistan. This happened before American troops withdrew from the region. The Pentagon called the killings a “tragic mistake.”

Airman 1st Class Quion Lowe/Air Force

The Air Force has already used remote piloted drones for years. Here, a student pilot and operator run an MQ-9 drone flight simulator at an Air Force base in 2019.

Dangers Ahead?

A recently revised Pentagon policy on the use of artificial intelligence in weapons systems allows for the autonomous use of lethal force—but any particular plan to build or deploy such a weapon must first be reviewed and approved by a special military panel.

Asked if Air Force drones might eventually be able to conduct lethal strikes without explicit human sign-off on each attack, a Pentagon spokesperson said that the question was too hypothetical to answer. Any autonomous Air Force drone, she said, would have to be “designed to allow commanders and operators to exercise appropriate levels of human judgment over the use of force.”

A recently revised Pentagon policy on the use of artificial intelligence in weapons systems allows for the autonomous use of lethal force. But any particular plan to build or deploy such a weapon must first be reviewed and approved by a special military panel.

Asked if Air Force drones might eventually be able to conduct lethal strikes without human sign-off, a Pentagon spokesperson said that the question was too hypothetical to answer. Any autonomous Air Force drone, she said, would have to be “designed to allow commanders and operators to exercise appropriate levels of human judgment over the use of force.”

Humans will continue to ‘exercise appropriate levels of human judgment over the use of force,’ a military spokesperson said.

Air Force officials say they fully understand that machines aren’t intelligent in the same way humans are. Machines have no built-in moral compass, and A.I. technology can also make mistakes—as has happened repeatedly in recent years with driverless cars.

Humans will continue to play a central role in the new vision for the Air Force, top Pentagon officials say, but they’ll increasingly be teamed with software engineers and machine learning experts, who will be constantly refining algorithms.

Almost every aspect of Air Force operations will have to be revised to embrace the shift. The Pentagon has already spent several years building prototypes like the Valkyrie and the software that runs it. But now, if Congress approves, $5.8 billion in taxpayer dollars will go toward buying the A.I. drones.

Air Force officials say they fully understand that machines aren’t intelligent in the same way humans are. Machines have no built-in moral compass. A.I. technology can also make mistakes. It has happened repeatedly in recent years with driverless cars.

Humans will continue to play a central role in the new vision for the Air Force, top Pentagon officials say. However, they’ll increasingly be teamed with software engineers and machine learning experts, who will be constantly refining algorithms.

Almost every aspect of Air Force operations will have to be revised to embrace the shift. The Pentagon has already spent several years building prototypes like the Valkyrie and the software that runs it. But now, if Congress approves, $5.8 billion in taxpayer dollars will go toward buying the A.I. drones.

The Test Pilots

Over the summer, Major Elder flew his F-15 Strike Eagle within 1,000 feet of the experimental XQ-58A Valkyrie. He watched it closely, like a parent running alongside a child learning how to ride a bike, as the drone flew on its own, reaching certain assigned speeds and altitudes. In a test slated for later this year, the combat drone will be asked to chase and then kill a simulated enemy target over the Gulf of Mexico. It will come up with its own strategy for the mission.

Major Elder watches for discrepancies between simulations run by computer before the flight and the actions by the drone when it’s actually in the air. He also looks for any sign of worrisome “emergent behavior,” where the robot drone is acting in a potentially harmful way.

The initial version of the A.I. software largely follows scripts based on computer simulations. Eventually, the A.I. software will have to be able to perceive the world around it—and learn to understand and overcome enemy defenses. Real adversaries will likely try to fool the A.I., for example, by creating a virtual camouflage for enemy planes or targets to make the robot believe it’s seeing something else.

Over the summer, Major Elder flew his F-15 Strike Eagle within 1,000 feet of the experimental XQ-58A Valkyrie. He watched it closely, as the drone flew on its own, reaching certain assigned speeds and altitudes. In a test slated for later this year, the combat drone will be asked to chase and then kill a simulated enemy target over the Gulf of Mexico. It will come up with its own strategy for the mission.

Major Elder watches for discrepancies between simulations run by computer before the flight and the actions by the drone when it’s actually in the air. He also looks for any sign of worrisome “emergent behavior,” where the robot drone is acting in a potentially harmful way.

The initial version of the A.I. software largely follows scripts based on computer simulations. Eventually, the A.I. software will have to be able to perceive the world around it. It will have to learn to understand and overcome enemy defenses. Real adversaries will likely try to fool the A.I. For example, they might create a virtual camouflage for enemy planes or targets. This might make the robot believe it’s seeing something else.

Edmund D. Fountain/The New York Times

‘I’m flying off the wing of something that’s making its own decisions.’ 

—Major Ross Elder

Training the algorithms in such skills will require massive data collection, and the software will need to be heavily protected against hacking by an enemy. In early tests, the autonomous drones already have shown that they can act in unusual ways, with the Valkyrie in one case going into a series of rolls.

In addition to the Valkyrie, the Air Force has begun a second test program, called Project Venom. It will put pilots in six F-16 fighter jets equipped with artificial intelligence software that will handle key mission decisions.

The goal, Pentagon officials say, is an Air Force that’s more unpredictable and lethal—one that would better deter enemies. And the hope is, at least for the U.S. Air Force, that it would be a less deadly fight.

Officials estimate that it could take 5 to 10 years to develop a functioning A.I.-based system for air combat. Air Force commanders are pushing to accelerate the effort—but recognize that speed can’t be the only objective.

“It is an awesome responsibility,” says Colonel Tucker Hamilton, the Air Force chief of A.I. Test and Operations. “We just need to get there methodically, deliberately, ethically—in baby steps.”

Training the algorithms in such skills will require massive data collection. The software will need to be heavily protected against hacking by an enemy. In early tests, the autonomous drones already have shown that they can act in unusual ways. The Valkyrie in one case went into a series of rolls.

In addition to the Valkyrie, the Air Force has begun a second test program. It is called Project Venom. It will put pilots in six F-16 fighter jets equipped with artificial intelligence software that will handle key mission decisions.

The goal, Pentagon officials say, is an Air Force that’s more unpredictable and lethal. It would better deter enemies. The hope is, at least for the U.S. Air Force, that it would be a less deadly fight.

Officials estimate that it could take 5 to 10 years to develop a functioning A.I.-based system for air combat. Air Force commanders are pushing to accelerate the effort. But they recognize that speed can’t be the only objective.

“It is an awesome responsibility,” says Colonel Tucker Hamilton, the Air Force chief of A.I. Test and Operations. “We just need to get there methodically, deliberately, ethically—in baby steps.”

Eric Lipton is an investigative reporter for The New York Times, based in Washington. With reporting by Rebecca Katzman.

Eric Lipton is an investigative reporter for The New York Times, based in Washington. With reporting by Rebecca Katzman.

Responsible A.I.?

We asked military ethicist Jessica Wolfendale to weigh in on the use of A.I. in warfare

Daniel Milner/Courtesy of Jessica Wolfendale

Jessica Wolfendale

There are a lot of concerns around using A.I. software in military drones. Research shows that using algorithms to select targets can be influenced by different kinds of unconscious racial biases. This can particularly be the case with target selection.

A signature strike is a drone strike based on calculations of probability that a certain person or group is a threat. The calculation might be based on assumptions around certain types of behaviors and the likelihood that, for example, someone might be part of a militant group.

Signature strikes are basically high-tech guesswork. The guesswork is based on input from people. And people, of course, are flawed. The input creates a database of supposedly suspicious behaviors, which is then used to target people—who may actually be engaged in completely innocent activity. This has actually happened a number of times [with remote piloted drones].

There are a lot of concerns around using A.I. software in military drones. Research shows that using algorithms to select targets can be influenced by different kinds of unconscious racial biases. This can particularly be the case with target selection.

A signature strike is a drone strike based on calculations of probability that a certain person or group is a threat. The calculation might be based on assumptions around certain types of behaviors and the likelihood that, for example, someone might be part of a militant group.

Signature strikes are basically high-tech guesswork. The guesswork is based on input from people. And people, of course, are flawed. The input creates a database of supposedly suspicious behaviors, which is then used to target people—who may actually be engaged in completely innocent activity. This has actually happened a number of times [with remote piloted drones].

‘I’m flying off the wing of something that’s making its own decisions.’ 

I think there are fairly innocuous and benign potential uses of A.I., such as mapping landscapes. And it’s not impossible that there could be more sophisticated ways of using A.I. to more accurately identify dangerous behavior. But in relation to targeting, I worry that in some ways, the technological progress is going ahead of the ethical evaluation.

The decisions the technology makes may seem objective because they’re coming from a computer. But the reluctance to challenge or question the output of an algorithm is very likely to lead to the mistaken killings of innocent people.

We should all care about technology that runs the risk of killing innocent people. We are delegating what ought to be moral decision making to algorithms. I don’t think A.I. is or will be actually capable of making those kinds of decisions.

—as told to Rebecca Katzman

This interview was edited and condensed for length and clarity.

I think there are fairly innocuous and benign potential uses of A.I., such as mapping landscapes. And it’s not impossible that there could be more sophisticated ways of using A.I. to more accurately identify dangerous behavior. But in relation to targeting, I worry that in some ways, the technological progress is going ahead of the ethical evaluation.

The decisions the technology makes may seem objective because they’re coming from a computer. But the reluctance to challenge or question the output of an algorithm is very likely to lead to the mistaken killings of innocent people.

We should all care about technology that runs the risk of killing innocent people. We are delegating what ought to be moral decision making to algorithms. I don’t think A.I. is or will be actually capable of making those kinds of decisions.

—as told to Rebecca Katzman

This interview was edited and condensed for length and clarity.

Wolfendale is professor of philosophy and co-director of the masters in military ethics program at Case Western Reserve University in Ohio.

Wolfendale is professor of philosophy and co-director of the masters in military ethics program at Case Western Reserve University in Ohio.

videos (1)
Skills Sheets (3)
Skills Sheets (3)
Skills Sheets (3)
Leveled Articles (1)
Text-to-Speech