It might be old news that “war is the father of all” (Heraclitus, circa 500 BCE), but the conflict in Ukraine has birthed high-tech horrors of which we do not seem to be sufficiently afraid. There have been grim soundbites—the Ukrainian battlefield as “a mash-up of All Quiet on the Western Front and Blade Runner,” in the words of writer Max Boot—and reports of the gruesome rounds of technological one-upmanship as Russia and Ukraine race to out-invent each other in a deadly wizard’s war.
The battlefield incubator has spawned a generation of drones that are terrifyingly fast, cheap, and legion, yes, but it is becoming more and more likely that one outfitted with artificial intelligence has already targeted and killed a human on its own—that is, of its sovereign determination—a line we seem to have slipped (or be slipping) over without much, if any, fanfare.
Whether or not a drone has yet followed through on a decision to kill without human oversight, this seems to be the human-appointed goal: to offload not just the work but the ethical calculations of warfare to machines.
Self-determining drones are currently learning to perform tasks on their own—to fly, see, target, and attack without human control—so that more of the “kill chain,” to use military parlance, can become “automated”: An assassination drone using facial recognition to find someone particular; a drone picking a target all its own, “pixel locking” not just onto a position, ship, or storage tank, but a moving vehicle or person; a drone whose mission is to decide when and whom to kill.
Three years ago, during the bloody conflict in Bakhmut, a Ukrainian drone pilot stared into the blue shellshocked eyes of a thirty-year-old former manager of a liquor store who had been drafted into the Russian army and now appeared in high resolution on his screen. Wounded, alone, hunted for hours by quadcopters, the soldier had come out of his hole and was trying to surrender to the drone. Instead of dropping the grenade he’d flown out to the trench to do the job, the pilot led the man to safe capture, saying, “Despite that he is an enemy, even though he has killed our boys, I still felt sorry for him.”
Of course an autonomous drone never shows mercy. I spent nearly a decade writing a cultural history of the aerial imagination—its pretense of mastery, how it erases as much as it reveals, and how quickly god-sight slips into bomb-sight. In 1910, only a year out of office, Theodore Roosevelt became the first U.S. president to fly in an plane; upon landing, Roosevelt exclaimed, “Bully!”, but before, while aloft, he kept waving and talking, though the pilot could only make out a few words, among them “bomb,” “army,” and “war.” Today in Ukraine, innovation is being fed by investors, coders, manufacturers, and governments, both within and without—a hellish laboratory for the future face of war.
A previous Chairman of the Joint Chiefs of Staff has said that within a decade most of the leading world militaries could be one-third robotic. Even if humans are “kept in the loop,” that is, still have some part in the decision-making process—retain the ability to abort or approve an auto-generated target, say—history suggests we will give the machine too much credit, mistake automation for truth, fall for the sheen of seemingly dispassionate analysis, when in reality we have made the algorithm in our own fallible image, baked our biases and shortcomings right into the code.
Machine vision is not only less accurate than our still rather flawed powers of human analysis, but also biased against certain populations (due to the data it trained on), prone to mistakes in common conditions and environments (snowy vs sunny skies, for example), and unable to contend with the unexpected (algorithms are based on predictability, something the modern battlefield is decidedly not). Artificial intelligence can sift through data, “seeing” more and targeting faster than humanly possible. But if AI still can’t reliably identify a face—or drive a car—why should we let it pilot our bombs? Are we so willing to sacrifice accuracy for speed? Or is it expediency? Maybe the real dream is that we can abdicate accountability—mistakes will be made, but now we can blame the machine.
Some hope AI will lead us to peace: free from our baser impulses (greed, hatred, ignorance), the computer might see what humans have chosen to ignore all along—that the cost of modern war is simply too high. As the rogue supercomputer in the 1983 movie WarGames learned, when it took control of the U.S. missile systems and tried to win a simulated global thermonuclear conflict: “A strange game. The only winning move is not to play.” Unfortunately, in recent rounds of tests, AIs have proven trigger-happy—quick to fire a nuke. An old fantasy, one technology to end all wars. As Orville Wright mused in 1917: “When my brother and I built and flew the first man-carrying flying machine, we thought we were introducing into the world an invention which would make further wars practically impossible.”
By now we should know that people stop wars, not machines.
Military strategists liken the arrival of AI on the battlefield to the advent of the machine gun. Lethal efficiency that remakes the rules of engagement. But Alexander Karp, one of the cofounders of Palantir Technologies, which develops AI platforms for the U.S. military, among others, most often reaches for the analogy of the nuclear bomb. No surprise that in January the thinkers at the Bulletin of the Atomic Scientists moved the hands of the Doomsday Clock the closest they’ve ever been to the eve of our destruction, including during the hair-trigger days at the height of the Cold War. We’ve trod this path before. The cold, closed logic of an arms race: development for development’s sake, paranoia, greed, faith in a techno-fantasy seemingly bent on mutually assured destruction.
Eighty years ago, with the birth of the bomb, we became destroyers of worlds. Now what we might finally obliterate, what we’re chipping away one algorithmic iteration at a time, is the last vestige of our humanity.
