Perspective.

https://www.foreignaffairs.com/united-states/ai-weapons-and-dangerous-illusion-human-control

AI Weapons and the Dangerous Illusion of Human Control

… Although it’s comforting to imagine that living, breathing people will restrain mindless algorithms from killing indiscriminately, this consensus is out of step with technological reality. The AI models driving contemporary autonomous weapons are often too sophisticated for even the best-trained operators to supervise. What is more, every proposed use for autonomous weapons imagines them operating at massive scale, which would require levels of data, speed, and complexity that would make meaningful human intervention impossible. Under normal conditions, expecting a human being to weigh the merits of an AI system’s analysis and suggest courses of action would be challenging. Under combat conditions—marked by severe stress, limited time, and interrupted or nonexistent communications between individuals, units, and higher authorities—doing so would be impossible. Rather than indulge the illusion that humans will have the capacity to control autonomous weapons in wartime, therefore, militaries must build trust in their models for autonomous weapons now—under peacetime conditions—and let them function without excessive human intervention when the shooting starts.