One of the biggest problems when discussing AI of any kind is our tendency to invoke a mythical human against which to compare the AI we are concerned about. This mythical human is conscious, self-aware, reasonably consistent in its actions, and reasonably competent at its task. Alas, the mythical human has never, and will never, exist. Ordinary people are barely conscious even some of the time, they generally have little or no idea why they are doing what they are doing, and they are often possessed of quite low competence for the task at hand. The debate about AI in cars is a classic example of our intellectual error: we want AI in cars to be infinitely better than real humans, who are often totally incompetent and unaware of what is around them. Everyone imagines themselves to be "better than average" but not one person in 100,000 has ever even considered taking an advanced driving course, never mind actually going on one.

The same is true in war. Remember, it has been ordinary Russian humans committing atrocities in Ukraine. It was ordinary humans working for Blackwater who drove around in Falluja tossing grenades into apartment blocks filled with terrified civilians. As for holding people accountable, the victors rarely if ever face consequences for war crimes (for example, dropping atomic bombs on cities full of civilians....) but the losers are often held to account - that's not justice, it's merely post-hoc revenge.

All in all, the case against AI in war has yet to be made. Current arguments are stuck in an ethical dead-end because comparisons with mythical humans tell us nothing of importance.