Army of None
Machines deciding who dies
Description
On July 3, 1988, the USS Vincennes was tracking an aircraft climbing out of Bandar Abbas over the Persian Gulf. Its Aegis combat system — a network of radar and computers built to manage dozens of threats faster than any crew could — flagged the plane as a descending Iranian F-14 on an attack run. The captain had seconds. He fired. The aircraft was Iran Air Flight 655, a civilian airliner, climbing not descending, carrying 290 people, all of whom died. The machine had done roughly what it was designed to do. The humans, reading its outputs under pressure, drew the wrong conclusion. Nobody could quite say afterward where the error had lived — in the code, in the crew, or in the space between them.
Paul Scharre keeps returning to that gap. A former Army Ranger who served in Iraq and Afghanistan, he later helped write Pentagon policy on autonomous weapons, and in Army of None he sets out to map a question the military itself has not settled. Weapons that select and engage targets on their own are no longer speculative. Loitering munitions, self-guided missiles, sentry guns that can track a human across a border — the technology to remove the person from the loop mostly exists. What is missing is agreement on whether it should be used.
Scharre is not a pacifist writing against war, nor a technologist selling the future. He is an insider who has stood at the checkpoint and read the classified memos, and he is uneasy about both the people who want to ban these weapons outright and the people who wave the worry away. The uncertainty is the point.
The question we’re asking : When a weapon can choose its own target, who is left to answer for the death it causes?What we’ll see : How a soldier, a Pentagon lawyer, and a software engineer each see the same trigger differently — and why the drift toward machines that fire on their own may be nobody's decision at all.
Table of contents
01Chapter 1 — The Aegis that almost fired
Scharre opens close to his own experience, because the abstract debate over "killer robots" tends to skip the part he knows best: the moment of decision. He describes lying on a ridgeline in Afghanistan as a scout, watching a young girl herd goats near his team's position while relaying their location back to Taliban fighters. By a strict reading of the laws of war, she may have been a combatant. No soldier in his unit would have dreamed of shooting a child. That refusal was not in any rulebook. It came from somewhere a targeting algorithm has no access to.
From there he pulls back to the systems already in service. The Aegis that misjudged Flight 655 was not autonomous in the science-fiction sense, but it sat on a spectrum. It could be set to different levels of human control, from fully manual to a semi-automatic mode where it would fire on its own unless a human intervened in time. Scharre's point is that the frightening threshold is not some future robot with a mind of its own. It is the quiet slide toward modes where the human is technically present but functionally a rubber stamp, approving what the machine has already chosen.
02Chapter 2 — The line the Pentagon won't quite draw
Having shown how blurry the technical line is, Scharre turns to the institution that has to draw it. In 2012 the Department of Defense issued Directive 3000.09, the first serious attempt by any military to set policy on autonomous weapons. It required that such systems allow commanders and operators "appropriate levels of human judgment over the use of force." Scharre helped shape that language, and he is honest about what it does and does not accomplish. "Appropriate" is doing enormous work. The directive did not ban autonomous weapons. It set up a review process and left the hard question open.
He is skeptical of clean solutions from every direction. The campaign to ban lethal autonomous weapons outright, pushed by a coalition of NGOs and some states, has moral force, but Scharre keeps asking the practical questions its advocates tend to skip. What exactly is being banned? A landmine already engages without a human deciding. A homing torpedo picks its own path to a ship. If the definition is too broad it catches weapons in use for decades; too narrow and it misses the systems that actually worry people.
03Chapter 3 — When the algorithm mistakes a tractor for a target
The strongest section of the book is the least ideological: it is about how these systems actually fail. Scharre draws on his time around real military hardware and on the wider history of automation accidents to argue that the danger is rarely a machine turning evil. It is a machine being confidently, catastrophically wrong in a situation nobody anticipated. Autonomous systems are brittle at the edges. They perform beautifully inside the conditions they were tested for and fall apart the moment reality supplies something the designers never imagined.
He borrows the concept of "normal accidents" from the sociologist Charles Perrow, who studied disasters like Three Mile Island. In tightly coupled systems, where components interact fast and in complex ways, failures are not freak events — they are built in, waiting for the right combination of small errors. A weapon that reacts in milliseconds to another weapon that reacts in milliseconds creates exactly this coupling. Scharre raises the specter of a "flash war," a military version of the 2010 flash crash that briefly wiped roughly a trillion dollars off Wall Street before humans could react. Two automated forces could escalate a border skirmish into a shooting war before any commander understood what was happening.
04Chapter 4 — The decision no machine should own
Stepping back, Scharre's real subject is not the hardware but what war is when the kill decision leaves human hands. He is careful not to romanticize the human trigger. Humans firebombed cities and ran death camps; the record does not suggest that keeping a person in the loop guarantees mercy. His claim is narrower and harder to dismiss. Whatever restraint exists in war lives in the human capacity to refuse, to feel the weight of a killing, to be held to account for it afterward. Remove the person entirely and you do not just change the speed of war. You remove the possibility of that refusal.
This is why he resists both the ban and the shrug. A treaty that pretends the technology can be uninvented ignores that the components are dual-use and already everywhere. But treating autonomy as just another performance upgrade misses that some decisions are different in kind, not degree. Choosing to end a specific human life is, for Scharre, one of them — not because a machine could not do it accurately, but because accuracy was never what was at stake. What is at stake is moral agency, and agency is not a feature you can install.
05Conclusion
The Aegis crew that fired on Flight 655 has never fully left the debate. They had the machine and they had the final say, and the tragedy happened anyway, in the gap between a system's confident output and a human's ability to question it under pressure. Scharre's warning is that we are steadily narrowing that gap — not by malice, but by the accumulated logic of speed, cost, and keeping pace with rivals. Each step looks reasonable. The destination does not.