War by Algorithm

M A Hossain, 

There is a moment in the long, unglamorous history of weapons development when a technology stops being a tool and starts being a doctrine. The machine gun began as a curiosity, then became a tactical instrument, then remade the entire architecture of European civilization — burying a generation in Flanders mud. The atomic bomb was sold as a way to end one war. Within a decade it had restructured global politics entirely, producing a permanent condition of mutual dread that we still inhabit today.

We are at that inflection point again. This time, the technology is artificial intelligence, and it is already deployed on real battlefields, identifying real targets, accelerating real decisions about who lives and who dies. The question is no longer whether AI will transform warfare. It already has. The question — the urgent, uncomfortable one that democratic societies are conspicuously failing to ask — is what kind of war we are choosing to wage, and whether we ever actually chose it at all.

Project Maven, the Pentagon's flagship AI targeting program, was not born from a grand strategic vision. It emerged from a specific, practical frustration: the U.S. military was drowning in drone footage it could not process. Analysts stared at screens for hours, exhausted and error-prone, watching hours of surveillance video and flagging maybe a fraction of what mattered. Maven's original proposition was modest — let computer vision handle the sorting, free the humans for judgment. Track motorcycles. Flag patterns. Save the analyst's eyes for what counts.

That is a reasonable argument. It is also, historically, how every incremental weapons revolution begins: with a limited use case and a sensible justification.

The numbers, however, have a way of growing. By last year, the National Geospatial-Intelligence Agency was boasting that AI-assisted targeting could process a hundred targets a day without the technology — and a thousand with it. Add large language models into the targeting cycle, they said, and you could plausibly reach five thousand targets a day. Five thousand. The human analyst who once strained to review fifty images is now nominally "in the loop" on decisions being generated at industrial scale. The phrase "human in the loop" — repeated like a liturgical chant by military commanders — begins to look less like a safeguard and more like a public relations strategy.

Jim Mattis, who has forgotten more about strategy than most generals ever learn, put it plainly: targetry is no substitute for strategy. The ability to identify and strike five thousand things a day tells you nothing about whether striking them advances any coherent goal. It took the United States years of punishing experience in Iraq and Afghanistan to internalize that lesson. The temptation, with AI, is to forget it again — to mistake speed and scale for effectiveness, and efficiency for wisdom.

Consider the Iran operations, which have functioned as a real-world demonstration of AI-enabled warfare at scale. Central Command confirmed publicly that AI tools had compressed decision cycles that used to take days down to seconds. Impressive. And yet — what has been achieved? The regime did not fall. No popular uprising materialized. The strategic position of the United States in the region is not obviously improved. The argument that precision and speed produce strategic results has yet to be proven; in Iran, as in much of modern American military experience, the machines worked and the strategy did not.

The civilian dimension is worse. There have been credible reports of a school struck, lives lost — the kind of incident that, in an earlier era, might have provoked congressional hearings, public reckoning, genuine accountability. In the AI era, the conversation shifts almost immediately to the process. Was the algorithm correctly trained? Was the data properly labeled? The dead become a debugging problem.

This is not hypothetical squeamishness. The laws of armed conflict exist precisely because wars fought without restraint tend not to end — they metastasize. The United States spent decades building the normative architecture of proportionality, distinction, precaution. That architecture is not a concession to weakness. It is the thing that makes the United States a country whose wars can be sustained politically over time. Undermine it, and the strategic cost compounds.

There is a serious counterargument, and it deserves honest engagement. Friendly fire incidents — the agonizing deaths that result from human confusion, exhaustion, and imperfect information — have plagued every war in history. If AI can genuinely reduce the rate at which soldiers kill their own comrades, or at which pilots mistake wedding parties for insurgent formations, that matters enormously. The advocates of Project Maven were not cynics or warmongers; many were veterans haunted by preventable deaths.

The problem is the gap between what AI was designed to do and what it is increasingly being used for. The early, cautious ambition — don't use AI for casualty estimation, limit it to tracking motorcycles — give way, step by step, to something far larger. One Pentagon program rejected the idea of using AI to estimate how many civilians were in a strike zone because "nobody wanted to stake their career on an AI-generated count." That instinct was sound. The worry is whether it survived contact with institutional pressure to go faster.

Ukraine has offered a different model. The Delta system — Ukraine's battlefield awareness platform — has demonstrated that near-total situational awareness can be achieved without surrendering targeting decisions to autonomous systems. Drone operators can work from anywhere in the world. Every unit sees the common operating picture. The humans remain, genuinely, in the loop — not as rubber stamps on machine-generated kill lists, but as decision-makers with real information. It is not a perfect system. But it suggests that the choice is not binary, between AI targeting and fog of war.

The deeper problem is political, not technical. The transformation of American warfare through AI has happened largely outside democratic deliberation. Congress has not passed legislation governing autonomous weapons. The public has not been asked whether it consents to wars fought at machine speed. The phrase "appropriate levels of human judgment" — the Pentagon's current formulation, notably absent from any mention of a "human in the loop" — was written by officials, not debated by legislatures. History is not encouraging at this point. 

Technologies that reshape the conduct of war tend to outpace the political institutions meant to govern them. The drone program expanded quietly for years before public debate caught up. Nuclear weapons produced the doctrine of mutual assured destruction before strategists had fully thought through its implications. Each time, the technology moved first, and democratic accountability scrambled to follow.

AI is moving faster than any of them. The question is not whether algorithms belong anywhere near the battlefield. The question is who decides where the line is — and whether anyone is really drawing it at all. That is not a technical question. It is a political one. And so far, the answer looks uncomfortably like: nobody in particular, moving very fast, in a direction nobody formally chose. That should bother us. It should bother us a great deal.

M A Hossain is a senior journalist and international affairs analyst, based in Bangladesh. He can be reached at: writetomahossain@gmail.com

  This article published at :

1. The Nation, Pak : 25 May, 26