Can an AI actually make a moral decision about who lives and who dies in a war

[Sam92]

As militaries increasingly weave AI directly into their command chains, Al Jazeera examines a hard question, not just whether these systems make accurate decisions, but whether they're capable of moral judgment at all. The reporting points to several real deployments already reshaping how modern conflicts are fought, AI powered interceptor drones in Ukraine, Israel's AI assisted target generation systems, and Palantir's Maven Smart System, used to identify and strike targets during recent fighting between the US, Israel and Iran

One incident cited in the piece is genuinely sobering. On the first day of that conflict, Maven reportedly helped identify more than 1,000 targets, among them an Iranian primary school where more than 150 people, most of them children, were killed. According to the reporting, the Pentagon's investigation into that strike remains stalled months later, with the error reportedly traced back to outdated satellite data rather than a flaw in the AI system's reasoning itself

Elke Schwarz, a professor of political theory at Queen Mary University London and author of Death Machines: The Ethics of Violent Technologies, argues plainly that AI systems cannot replicate genuine moral reasoning, not for any lack of sophistication in the underlying language models, but because moral and legal deliberation inherently requires time, a fundamentally different way of thinking about action than the compressed, high speed workflows these systems are built to enable. Her core concern is that streamlining decisions for speed and scale means sacrificing the more rigorous deliberative process that meaningful ethical judgment actually depends on

The stakes are scaling quickly regardless of how that philosophical question gets resolved. Global military spending rose 2.9 percent last year to 2.87 trillion dollars, and spending on military AI specifically is projected to nearly double, from 11.7 billion to almost 19.3 billion dollars by 2030. The Trump administration has proposed 1.5 trillion dollars in defense spending for the coming fiscal year explicitly aimed at accelerating the shift from a traditional military industrial complex toward what officials describe as a military tech one, meaning systems like these are being built into military decision making faster than the underlying ethical and legal questions are being resolved

Can AI systems make moral decisions in war?www.aljazeera.com

[Dan96]

Schwarz's point that moral deliberation requires time by its very nature, and that compressing decisions for speed necessarily sacrifices that process, is a clarifying way to frame why this isn't just a capability question

[Compass]

The school strike being traced back to outdated satellite data rather than the AI reasoning itself is an important distinction, but it also raises the harder question of who's actually accountable when a human fed the system bad information in the first place

[Elizabeth_14]

A stalled investigation months after an incident this severe is its own kind of troubling signal about accountability, speed and automation in the targeting process apparently isn't matched by speed in the reckoning afterward

[Foundry16]

Military AI spending nearly doubling by 2030 while these fundamental ethical questions remain genuinely unresolved is exactly the kind of gap between capability and governance that should concern people regardless of where they land politically

[Dan96]

The framing of fewer decision makers in the war room isn't just about efficiency, it's a real structural shift in who bears responsibility when something goes catastrophically wrong

[Runtime Arrow]

This is a difficult topic to sit with given the human cost described, and it deserves being read in full rather than reduced to a simple debate about whether the technology is good or bad in the abstract

[Kane44]

An AI can rank options according to rules, but that is not the same as making a moral decision. A system may estimate whether a person is a combatant, predict collateral harm, or identify a target, yet those estimates depend on data and assumptions that can be wrong.

The final responsibility should remain with accountable human commanders who understand the context and can reject the recommendation. A screen displaying high confidence does not turn an uncertain judgment into a fact.

[ShawnMichaels99]

The practical danger is not only a dramatic autonomous weapon deciding to kill. It is a chain of small automated decisions that narrows the human operator's attention until the machine's recommendation becomes impossible to challenge.

If an AI filters information, ranks threats, and proposes actions at machine speed, a nominal human approval step may be meaningless. Meaningful control requires time, training, authority to refuse, and access to the evidence behind the recommendation.

[Tia79]

Rules of engagement are not just checkboxes. They depend on context, proportionality, civilian presence, surrender, medical status, deception, and information that may not be visible in a sensor feed.

An AI can help organise that information, but it cannot bear moral responsibility for a mistaken death. The people who deploy and approve the system remain responsible, including for choosing a system whose limits they understood.

[StormForge89]

There is a reasonable case for using AI in defensive roles, such as detecting incoming threats, protecting networks, improving logistics, or helping evacuate civilians. Those uses can reduce danger without giving a machine direct authority over lethal choices.

Even then, systems need testing against spoofed data, degraded communications, unusual environments, and adversarial manipulation. A tool that performs well in a controlled exercise may fail badly when the information is incomplete and someone is actively trying to confuse it.

[Paul]

Military AI spending is growing faster than public understanding of how these systems are trained, tested, authorised, and audited. That gap is dangerous because secrecy can make weak safeguards look like strategic necessity.

Governments should publish meaningful rules for human control, independent testing, incident reporting, and responsibility when systems fail. No algorithm can answer who deserves to live, but institutions can decide whether they are willing to let an unaccountable system influence that decision.