Here’s an idea for an awesome dogfighting aircraft. Make it small, light and fast. Build it out of materials that are hard to detect on radar. Even give it a laser cannon.

Oh, and don’t put a human in the cockpit. In fact, don’t even closely tie the drone to human ground control. Because in an aerial knife fight, a computer-controlled machine will beat a human pilot.

That’s the idea behind a controversial new proposal by U.S. Air Force captain Michael Byrnes, an experienced Predator and Reaper drone pilot. Byrnes is calling for the development of a robotic dogfighter, which he calls the FQ-X, that could blow manned fighters out of the sky.

“A tactically autonomous, machine-piloted aircraft … will bring new and unmatched lethality to air-to-air combat,” Byrnes writes in Air and Space Power Journal.

In Byrnes’ conception, machines have the edge in making the lightning-fast decisions necessary to win a close-quarters aerial battle. “Humans average 200 to 300 milliseconds to react to simple stimuli, but machines can select or synthesize and execute maneuvers, making millions of corrections in that same quarter of a second,” he writes.

Byrnes focuses on famed fighter pilot John Boyd’s classic observe-orient-decide-act decision cycle—the “OODA loop”—which predicts that victory in combat belongs to the warrior who can assess and respond to conditions fastest.

Like a fighter pilot trying to out-turn his opponent in a dogfight, the trick to OODA is quickly making the right decisions while your enemy is still trying to figure out what’s going on.

It’s a battle of wits in which computers are superior, according to Byrnes. “Every step in OODA that we can do, they will do better.”

Byrnes envisions a drone designed from the start to utilize the full potential of an unmanned dogfighter. The FQ-X would be constructed of advanced, difficult-to-detect “metamaterials.” It would have extremely powerful computers that could determine an enemy aircraft’s position from even the scantest of sensor data.

“The principle of ‘first look, first kill’ belongs to the aircraft with the most processing power and the best software to leverage it,” Byrnes writes.

The FQ-X would also have multispectral optics and computer vision software that would it enable it to distinguish friendly from enemy aircraft. The drone would pack a laser or a cannon firing armor-piercing incendiary rounds.

To sweeten the robot’s victory, on-board machine-learning systems would analyze the encounter and transmit tips to other combat drones.