The Pentagon has poured billions of dollars into artificial intelligence for warfare, funding systems that analyze drone footage and, in some cases, make autonomous targeting decisions. The scale of the investment represents one of the largest shifts in military technology since the development of precision-guided munitions, and it is happening with relatively little public debate.

Katrina Manson, an author who has written extensively on the subject, discussed the Pentagon's AI push on GZERO World with Ian Bremmer. Her reporting details how the Department of Defense has moved from experimental projects to operational deployment of AI tools across multiple domains.

The most visible applications involve drone footage analysis. Human analysts can review hours of video from surveillance drones, but AI systems can flag potential targets or suspicious activity in real time. This speeds up the kill chain, reducing the time between detection and action. The Pentagon has deployed these tools in counterterrorism operations, where the volume of aerial surveillance has overwhelmed human capacity.

More controversial is the move toward autonomous targeting. These systems use machine learning to identify and engage targets without direct human control over each individual shot. The military describes this as "lethal autonomous weapons" and has argued that they can react faster than humans in combat scenarios. Critics, including many in the AI research community, warn that these systems lack the ability to distinguish between combatants and civilians, especially in complex environments like urban warfare.

The Pentagon has not fully embraced fully autonomous killing. Current doctrine requires a human to be "in the loop" for lethal decisions. But the technology is pushing toward what experts call "human-on-the-loop" supervision, where a person monitors multiple AI-controlled systems and can intervene only if something goes wrong. Whether that distinction offers meaningful accountability is an open question.

Manson's work highlights the tension between military necessity and ethical risk. Proponents argue that AI can reduce collateral damage by making targeting more precise. They point to drone strikes that have killed civilians because of faulty intelligence — an AI system, they say, might have avoided those mistakes by cross-referencing more data. Opponents counter that AI systems are brittle: they fail in unexpected ways, they can be fooled by adversarial inputs, and they strip human judgment from life-or-death decisions.

The Pentagon has attempted to address these concerns by issuing ethical principles for AI use in 2020. The principles call for systems to be "responsible, equitable, traceable, reliable, and governable." But they are not legally binding, and there is no independent oversight of how the military implements them. Several defense contractors have since developed AI targeting systems that operate with varying degrees of autonomy, none of which have been subject to public audit.

International law offers little guidance. The United Nations has held talks on lethal autonomous weapons since 2014, but no treaty has emerged. The United States has opposed a ban, arguing that existing laws of armed conflict are sufficient. Meanwhile, Russia and China are developing their own AI weapons, creating an arms race dynamic that the Pentagon says it cannot ignore.

The financial commitment is enormous. The Pentagon's budget for AI research and development has grown from a few hundred million dollars a decade ago to billions today. Much of that money flows to the same tech companies that dominate the commercial AI market — though the Pentagon also funds its own research labs and startups working on military applications.

What does this mean for the soldier in the field? AI-powered systems are already being used to interpret sensor data, predict equipment failures, and coordinate drone swarms. The next generation of systems will likely make tactical decisions as fast as or faster than human commanders. The military says this will save lives by reducing the time troops spend in dangerous situations. But it also means that a software bug or an adversary's spoofing attack could have lethal consequences.

The most immediate risk is escalation. Autonomous systems that react faster than humans could drag nations into conflicts before diplomatic off-ramps are possible. A drone swarm that misidentifies a civilian vehicle as an enemy convoy could trigger a cascade of retaliation. No one has yet built a fail-safe mechanism for such scenarios.

Manson's reporting, as highlighted in the GZERO World interview, underscores a central paradox: the Pentagon wants AI to make war more precise and less bloody, but the technology itself introduces new forms of unpredictability. The billions spent on AI are buying capability, not certainty. And as the United States races to stay ahead of its adversaries, the ethical questions are being answered with policy documents rather than hard rules.

The next few years will determine whether the Pentagon's AI war machine remains under human control or whether the logic of speed and automation pushes decision-making further from the battlefield. For now, the technology is advancing faster than the debate about its limits.

That debate, if it ever happens at scale, will need to reckon with the fact that the Pentagon is already operating these systems in active theaters. The drone footage analysis is real. The autonomous targeting prototypes are real. The billions are spent. The only question left is whether the safeguards will catch up before something goes wrong.