Four Smarter Things to Say about Drone/AI Miltech in Ukraine Ronald Loui Ronald Loui Ronald Loui Published Jun 7, 2026 + Follow Many things to say. Four keep showing up in AI Seminar and rarely in the press. I was watching an interview on DW with a Ukrainian mil tech today and he said three of them! If you saw my BATTLEFIELD AI IN UKRAINE article about four years ago, i was talking about specificity and the Law of Armed Conflict (LOAC). Everyone was talking battlefield AI leaving the human out of the loop, or being held back from fatal decisionmaking, but i was thinking about AI reducing collateral damage. Precision, at least if that's your goal, is achievable with AI and Intel. You like carpet bombing instead? Of course, putin does what he does to civilian targets intentionally. You could even say the possibility of more precision outs the combatants who could be more precise, but really don't care. There was another point in that blurb about concentration of control and distribution of responsibility. Or just dragging your feet, like jury nullification, like Admiral Canaris resisting sHitler. Unfortunately, that lives only on linkedin (https://www.linkedin.com/pulse/battlefield-ai-ukraine-ronald-loui) which is sort of why i am putting my thoughts on linkedin these days. i still like that article and people should read it. i have to point out that much of the terminology i used came from study of these things prior to mainstream journalists discovering them. So people talk about Arkhipov these days, but i always talk Arkhipov, and everyone should. The prose about Lincoln i was also quite proud of. Something about Lincoln being responsible, but a hundred thousand Union soldiers also sharing that responsibility. These the Ukrainian expert did not talk about. But HITL, human-in-the-loop, he did mention prominently. Everyone wants it; everyone says it. People are correctly noticing that a rubber-stamp of AI is not really human-in-the-loop. But what dawns on me today as the Ukrainian expert was talking details: which loop? Some OODA (observe-orient-decide-act) loops are big. Some are tight. Downstream automation is as real as fire-and-forget, or finish-this-target-after-acquisition, even after-designation. Some just launch and pray. Sometimes you have the kill/cancel switch. We actually talked about that defeasance of intent back in the Kosovo air war. Where were your autonomous agent ethics then? The human-in-the-loop (makes me think of Chinese food on DelMar in U City) ... The human-in-the-loop idea needs refinement. It needs to start talking about proximal causality. I know you don't want to talk about that because you don't know what causality is. But we can help you. Law, of all things, talks about it a lot. And contributory negligence. And ascription of responsibility. Causality here is more about probabilistic influence within a discourse scope of admissible possibilities, where intermediaries, confounders, joint effects, counterfactuals, liberty-to-act, and mere correlations are all in focus. (Causality in high school physics and chemistry is a different animal!) You might like my long paper on this, for LAW OF AI 2nd ED (we were in the 1st too!). Took four years to teach myself enough liability terminology to push it out: http://awkscripts.com/papers/24%20Barfield%20proof%201%20edited_Ch%2024%20Loui.pdf ... but yes, you can talk about proximal intervention more intelligently than just saying human-in-the-loop, human-in-the-loop, human-in-the-loop. Right now, HITL is just a way of avoiding responsibility. CYA. Cover. Say that we tried, claim best effort. We had a human make a decision at some point. But yeah, there was a lot of downstream automation. Worry about the upstream AI failures too if you are Palantir. Stale data? We warned you about that as much as the Pentagon warned the Presidents about the Strait of Hormuz. The two more good ones this Ukrainian had spot on: the AI miltech update loop is fast; and these flying mines slash cruise missiles minus minus are going to replace nukes for mutual assured deterrence. That is, if the risk update feedback cycle can get long enough or routine enough that the deterrence is not easily defeated. I mean defeated in the sense of defeasibility, but ironically, defeated's a good word here. So the problem is the drone-counter-drone development has been on a furious timeline. What worked last year doesn't work today. He said six months. I believe him. This is by the way a wonderful example of where a nation of engineers can really put the screws to the nation of gangsters. This means we should be careful talking about nuclear deterrent being replaced by drone deterrent. Because deterrents disappear when people get confident they can defuse. Speaking of which, ABM was a Carter-era invention, not Reagan. And the current set of Iron Dome technologies is increasingly defusing the nuclear ballistic missile threat. Fortunately for those crazy game theorists in polisci in the 60s (fortunately?), that massive overkill of the nuclear MIRV warhead, and that launch-too-close-to-react from a stealthy submarine, leave in place enough doubt that you'd have to be an idiot to try it. Problem is the world's full of idiots. More like narcissist tyrant idiots who want to leave life and leave office with a big boom. Why I never liked the game theory MAD argument. I had Schelling's book in college. Took it to the beach and let the sand and surf destroy it. What. It. Deserved. But it's true that swarm tactics, persistence of threat, and fragility of modern infrastructure with sooooooo many good targets, makes a non-nuclear nation have deterrence options. Cheap too. So hat's off to Iran. You figured it out even before the Ukrainians made it work for them. BTW, this means you no longer need to be a nuclear-weapon-capable nation. Note to supreme leader. If you feel that you need the nuke to make you feel big, ok, i get it. I've seen enough traumatized children on playgrounds feeling small, looking for respect they never earned in other ways, but seek desperately. But in terms of miltech, actual deterrent against invasion, AI and drones are all you need. Some risks there, but any time we can ratchet down from single-error complete-destruction risk to a longer, slower, more intentional burn, with the great specificity that the LOAC desires, hurting mostly lousy humans rather than dolphins and red pandas, I'm making that trade.