THE ECONOMIST: Intelligence software led to military targeting machines and war on an industrial scale

The display of American and Israeli firepower in Iran has been more fast-paced and overwhelming than America showed off in either of the first two Persian Gulf wars.

The two allies are thought to have conducted more offensive sorties on February 28 than America managed on the first day of serious fighting — with much larger deployed forces — in either 1991 or 2003 (around 1300 in each).

Five days later Pete Hegseth, America’s secretary of war, boasted that “Operation Epic Fury has delivered twice the air power of shock and awe of Iraq in 2003”.

America and Israel can muster high volumes of strike sorties and missile launches because they can identify targets more precisely and more quickly than ever before possible. And they are able to produce targets with such pace, scope and precision because of their greatly increased use of software — including, to a limited extent thus far, artificial intelligence.

Both countries’ armed forces now generate and hit targets at an industrial scale.

The selection of targets has been under intense scrutiny since the early hours of the war.

On February 28, 175 people — most of them children — were killed after a girls’ school in Minab, in southern Iran, was hit by what was probably an American Tomahawk cruise missile.

On March 11, the New York Times reported the Pentagon had determined the strike was the result of a targeting mistake, as part of an attack on a nearby naval base.

Mr Hegseth has risked a perception of callous disregard for civilian lives by emphasising repeatedly that he wants to prioritise “lethality” in the armed forces over “tepid legality” — and by gutting the budget for civilian-harm assessments inside the Pentagon.

Almost 1800 people have been killed in Iran so far, most of them civilians, according to HRANA, a human-rights monitor in Washington.

America’s and Israel’s modernised targeting systems are far better at locating targets, and at minimising civilian casualties per strike, than what came before them.

In America the industrial process for Iran’s strikes is run by humans in Tampa, Florida, headquarters of Central Command (CENTCOM), which is responsible for the Pentagon’s operations in the Middle East.

A commander there generates options for various scenarios, such as bombing Iranian nuclear sites or toppling the regime.

His “J2”, or intelligence directorate, produces a database of thousands of possible targets, cobbled together from satellite images, signals intelligence and other sources. The database includes “no strike” lists of schools, hospitals and the like.

Commanders in control

A “weaponeer” decides which munitions are needed for which targets, such as bunker-busters for buried sites or GPS-guided joint direct attack munitions (known as JDAMs) for buildings.

Lawyers review targets, though their role is limited. (A lawyer reviewing targets “doesn’t say ‘You can’t do that’”, says a former American commander. “He says: ‘You can do that, but here are the consequences.’ At the end of the day, the ultimate lawyer is the commander.”)

The command’s J5 (strategy and plans directorate) assembles all this into a coherent war plan and passes it to the J3 (operations), which ultimately breaks the plan down into “air tasking orders” that typically look two days ahead.

Software has long been used to help with this. It estimates the probability of destroying the target given its location and construction, and the likely harm to civilians, by simulating the footprint of blast, heat, and fragmentation.

This can be overlaid on maps as a jagged-edged image, sometimes called a “splat”. But in recent years the technology has taken a leap in scope and sophistication.

America’s armed forces, including CENTCOM and those of NATO, now use the Maven Smart System, built largely by Palantir, an American firm, to soup up the whole process.

Maven is known as a “decision-support” tool. It takes information from open sources, such as social-media feeds, and classified sources, such as satellites, fusing all of it together.

If an Iranian mentions on Telegram (the preferred app for such chatter in Iran) that they saw a missile-launcher driven past their house, Maven can correlate that snippet with data from radio-frequency satellites that detect the electronic emissions from Iranian military radios.

Maven can then generate targets, work out which weapon is best placed to strike each one and assess the damage done afterwards. Maven also serves as a “digital twin” of the real world, writes Arnel David, the NATO officer in charge of the programme, allowing commanders to simulate how a particular decision might play out.

The aim, he says, is to turn military command into a “machine-aided, predictive science”.

All this means that data can be turned into targets at a far faster pace.

Joe O’Callaghan, a retired colonel who led the development of Maven at the US Army’s XVIII Airborne Corps, said on a recent podcast that one classified study had shown how Palantir’s system allowed military staff to plan an operation on the scale of the Iraq war with one-tenth of the manpower.

That will have since improved, he added. (Maven has been employed in aid of Ukraine from 2022.)

What would previously have taken dozens of people tens of hours, says a former NATO general involved with Maven, “That could be boiled down to two minutes.”

A European general describes what he has witnessed as “alchemy”. “We are moving from ten targets a day to 300,” he says. “The aspiration is 3000 a day.”

Israel uses different software but has also “industrialised” the process, as one officer puts it.

American planners were astonished when, in preparing for the war, their Israeli counterparts arrived with a “target bank” of thousands of Iranian targets (and munitions needed for each): the headquarters and homes of Iranian leaders, military and militia bases, missile-launchers and factories, civilian infrastructure and more.

“Israel has given more autonomy to decision-support systems to generate targets than I would ever have been given,” says the European general.

Israel’s process has its roots in the Yom Kippur war of 1973, when the country’s planes were mauled by Soviet-made air defences.

That prompted the Israeli Air Force to collect and fuse data on enemy surface-to-air missiles (SAMs) systematically, which contributed to a decisive victory against Syria in 1982. Then in a war with Hezbollah in Lebanon in 2006, Israel encountered a new issue.

As the war dragged on for 34 days, Israeli generals complained “The bank has run out of targets”.

Amos Yadlin, then commander of military intelligence, adapted the methods used against SAMs and expanded them to all potential targets.

The chief of staff of the Israel Defence Forces eventually had a bound copy of Hamas targets at the ready; whenever a rocket was launched from Gaza, he could choose a target for retaliation within minutes.

All of this might seem like a black box. Commanders argue that the target banks created by Maven and other tools, which have been tested for almost a decade, are generally reliable.

They apply a confidence level to each target, depending on the underlying data. They also tend to be better than a harried human analyst might be at identifying civilian objects, says an officer.

In both the American and Israeli armed forces, humans approve each target, outside extreme circumstances such as air-defence systems which are tasked with engaging a large number of incoming projectiles.

But some insiders acknowledge that the increasing scale and tempo of strikes has created incentives to give computers greater latitude in actually firing on the targets that they have generated.

Fears about autonomous strikes are at the heart of a dispute between Anthropic, developer of Claude, an AI model, and the Pentagon, though that remains a hypothetical concern for now. (Claude is used to some extent inside Maven but not for geospatial tasks like identifying objects.)

In NATO, some countries are “worried about the loss of human control”, says one person involved with the technology. “We’re moving at a pace of change I wouldn’t even have understood four years ago.”

Operator error

In many cases, the problem with computer-aided target banks is less to do with computers and more to do with the humans using them.

“AI can make a good intelligence officer better and help make sure there’s less collateral damage,” says an Israeli source. “But if the intelligence officer is just trying to come up with more targets and cares less about who gets hurt, AI will help generate those targets.”

Israel’s strikes on Gaza revealed another problem with maintaining sprawling target banks.

“When civilians were killed in strikes, we would go back and check our information,” says one analyst.

“Often it was because Hamas had used that building in the past but moved on and families had moved in.” The IDF periodically “revalidated” its targets, but not often enough.

A failure to revalidate targets may have been the problem with the strike on the girls’ school in Minab, a site which an official told the New York Times used to be part of the nearby naval base.

As humans get swamped by ever more computer-generated targets, enabling far more strikes per day, minimising the risks of such catastrophes will be an increasing challenge.

In America perhaps the biggest challenge comes from the top. Mr Hegseth has denigrated the laws of war, fired military lawyers and loosened rules of engagement.

He has also slashed the number of Pentagon personnel who work on the protection of civilians by 90 per cent, says one official.

The civilian-harm team that works within CENTCOM is at one-third of its pre-Hegseth staffing; personnel dealing with planning and those inside “strike cells” have suffered the greatest attrition. Such staffers can help understand how changes on the ground in Iran can render targets like the school in Minab out of date.

“The more we do on the planning side,” says the official, “the less we have to worry about Minabs.”

On March 8, CENTCOM said Iranians should stay home to avoid getting killed, and accused Iran’s leaders of putting civilians in harm’s way by deploying weapons in built-up areas. The regime, CENTCOM said, “blatantly disregards the safety of innocent people”.