>Kott’s ideas rest on the assumption that countries have obeyed a ban on autonomous weapons “beyond meaningful control”
Yeah, what's the actual incentive for that to happen? In the Last Good Man, by Linda Nagata (which I've already plugged before), one of the opponents has a distinct advantage because it has stripped down its bots' dont-kill-civs AI routines considerably.
I happen to think that an alternative approach is not so much to figure out how to manage all the clutter/civs/friendlies in a land battlezone, which are forecast to become more and more urban. Instead, first aim for autonomous systems in an Air-to-air context.
Anything that doesn't (somehow) light up as friend with your Friend-or-Foe transponders is fair game in a hot war. No need to worry about civs. Just shoot it down.
This is one of the aspects that really worries me about F35s budgetary commitments. We know AI is coming. We know it takes years to train pilots, more so than infantry. We also know that pilots are bound by 8-9G maneuverability meatsack limits. And we know that the other players are going to have a massive incentive to go asymmetric. If we could switch out of whatever gen 5 jets we are fielding when this happens, fine. If we're overspent, not so fine.
Heck, loitering/swarming supersonic flying drones/missiles with go back-to-home-if-unused capability (a la SpaceX boosters) would be 80% of the way there in concept. The hard limitation would then to put together sufficiently performant airframes in large enough numbers. But not having to carry, or train, a human would be a massive facilitator there.
Disclaimer: I am well aware that, so far, Wild Weasels have tended to run rings around SAM systems. But if you can have a bot smart enough for urban combat, then I strongly believe it'd be smart enough for AA. Or at least enough not to make a massive bet against it from peer enemies.