Can AI Systems Make Moral Decisions in Warfare?

The speed at which modern militaries can now make lethal decisions has compressed from hours to seconds. Palantir’s Maven Smart System processes thousands of strikes in minutes, feeding a military industrial complex that increasingly relies on algorithms to identify targets, select weapons, and assess damage. But as artificial intelligence becomes embedded deeper into the kill chain, a crucial question emerges: can machines make moral decisions?

The Speed Problem

The Pentagon’s push toward becoming an “AI-first fighting force” centers on accelerating the OODA loop, a decision-making model developed by Air Force Colonel John Boyd that prioritizes speed and adaptability over certainty. During Operation Epic Fury, the recent US war on Iran, Maven helped strike over 1,000 targets on the first day alone. Among them was a primary school in Minab where more than 150 people, mostly children, were killed in what Pentagon investigation suggests was an error stemming from outdated satellite intelligence.

This tragedy illustrates a fundamental problem: faster isn’t always better when lives hang in the balance. The compression of timelines from days to seconds doesn’t just change military operations; it fundamentally transforms the conditions under which moral reasoning can occur.

Elke Schwarz, a political theory professor at Queen Mary University London and author of “Death Machines: The Ethics of Violent Technologies,” argues that meaningful ethical deliberation requires time and a different mode of thinking. “You’re saying: we’re going to sacrifice a more rigorous deliberative process in the interest of speed and scale,” she told Al Jazeera. This frictionless workflow creates what researchers call moral distance, where delegating decisions to AI lends a veneer of legitimacy that makes humans more willing to engage in acts they wouldn’t personally authorize.

The Limits of Silicon Valley Ethics

Large language models like Anthropic’s Claude are now integrated into military targeting systems, querying massive datasets and proposing strike recommendations in natural language. Yet Schwarz is categorical: AI cannot replicate moral reasoning, not because the technology lacks sophistication, but because morality itself is fundamentally human.

“Ethics is a social practice,” Schwarz explains. “It rests on the fact that we take each other’s vulnerability very seriously, and that we trust one another not to violate that unless circumstances dictate. A system is a computational system. It has no concept of the meaning of human life. It has no concept of mourning, of suffering.”

The risks become tangible when you consider what researchers have documented about AI system failures in warfare contexts. The Israeli army’s Lavender targeting system, which identifies Palestinians as potential bombing targets, carries an estimated 10 percent error rate. Human oversight in some cases lasted mere seconds before authorization. International humanitarian law experts warn this practice risks wholesale violations.

Maven’s actual error rate remains undisclosed, but estimates suggest a system far less reliable than Pentagon officials claim. Yet as the Intercept reported, the Pentagon has gutted its Civilian Protection Center of Excellence from 40 staff to just nine, leaning instead on AI tools to conduct harm assessments faster.

Can AI Constrain Bad Actors?

Some argue that properly designed AI systems could impose restraint on unethical militaries from above. Anthropic itself refused to loosen restrictions on autonomous weapons, prompting the Pentagon to blacklist it as a “supply-chain risk.” But Schwarz dismisses this optimism as naive.

“If a military has a system that scales up and speeds up the violence they want to enact, they’re not going to abdicate to a system that says ‘no, you shouldn’t do that,’” she said. “An AI system will not act as a moral decision-maker over an unethical human.”

Military spending on AI applications is projected to nearly double from $11.7 billion today to $19.3 billion by 2030. The momentum is too strong, the institutional investment too deep, for technological guardrails to provide meaningful constraint. The problem requires addressing beforehand, not hoping algorithms will save us from ourselves.

The Pentagon insists humans always make final decisions on what to shoot. But when a system transforms processes that took days into decisions made in seconds, and when institutional cultures increasingly defer to algorithmic recommendations, what does human authorization really mean?

Source: Al Jazeera