Seumas Miller
5 min
This paper examines the ethical and legal challenges posed by Lethal Autonomous Weapon Systems (LAWS). Specifically, it addresses three core issues: the unpredictability of AI-driven systems in relation to Meaningful Human Control (MHC), the problem of the responsibility gap, and the moral permissibility of different levels of human-machine interaction (in-the-loop vs. on-the-loop). The author employs a philosophical framework centered on joint action and institutional design to argue that human responsibility can be maintained even when using autonomous technology.
The author contends that the unpredictability of self-learning LAWS is not an absolute barrier to their moral use. Instead, the focus should be on institutional arrangements that ensure human controllers remain responsible for the outcomes of these systems. By distinguishing between machine control (causal) and human control (moral/rational), the paper argues that human operators, analysts, and commanders can be held collectively responsible for the actions of LAWS. The author proposes that institutional roles should be designed to ensure that human oversight is not just effective, but morally informed, thereby closing the responsibility gap.
As military forces increasingly integrate AI into combat, the debate over autonomous weapons often stalls on the impossibility of holding machines accountable. This paper shifts the focus from the machine to the human institution. By providing a normative framework for collective responsibility, it offers a path for military organizations to integrate advanced technology while remaining accountable to international humanitarian law and ethical standards. It clarifies that while machines can perform tasks, they cannot occupy moral roles, making the human-in-the-loop or human-on-the-loop configurations essential for moral legitimacy.
Abstract This article is concerned with three key ethical issues that arise from the use in military combat of Lethal Autonomous Weapons Systems (LAWS). The first issue concerns the unpredictability of LAWS in respect of the requirement of Meaningful Human Control (MHC) – as opposed to machine control - in theatres of war. It is argued that the unpredictability of (especially) ‘self-learning’ LAWS is not necessarily a barrier to their morally acceptable use under some restricted conditions. The second issue concerns a normative framework for ascribing moral and, potentially, legal responsibility in respect of LAWS, i.e., the so-called responsibility gap. It is argued that the notion of collective responsibility is potentially helpful here. The third issue concerns human on-the-loop LAWS. It is argued that whereas human out-of-the loop weapons should be prohibited, nevertheless, under certain restrictive conditions some human on-the-loop LAWSs (as well as human in-the-loop LAWSs) may be morally acceptable.
Sam: So the institutional failure isn't that the AI lacks morality — it's that we've built roles that allow humans to treat the AI's proxy outputs as if they were moral outputs.
Alex: Precisely. And that's where Miller's framework becomes a design prescription rather than just a diagnosis. The roles have to be explicitly tethered to the legal principles of jus in bello — discrimination and proportionality. Not in a nominal sense, but in a way that makes it impossible to discharge the role without engaging those principles. [[RP_SECTION:accountability-and-diffusion|Accountability and Diffusion]]
Sam: There's a real tension there, though. The more you distribute accountability across a chain of interdependent roles, the more you risk diffusion. If everyone is responsible, effectively no one is — each actor assumes the moral check is happening somewhere else in the chain.
Alex: That's the primary critique, and Miller acknowledges it. In high-intensity combat, bureaucratic structures can amplify exactly that diffusion. His response is that the roles can't be administrative boxes — they have to carry legally and morally binding obligations that are role-specific. The analyst who clears a target is accountable for the discrimination judgment; the commander who authorizes is accountable for the proportionality assessment. The accountability isn't shared in a way that dilutes it; it's distributed in a way that assigns it.
Sam: That's a meaningful distinction, but it does assume we can define those roles with enough precision to capture every morally relevant contingency. The fog of war is precisely the condition where those clean role boundaries break down.
Alex: That's the honest limitation of the framework. Miller is essentially arguing for what you might call responsibility-by-design — the idea that before a system is certified for deployment, you'd audit not just its kinetic accuracy but its integration into a chain of human moral accountability. Whether that's achievable under operational conditions is a question the paper raises but doesn't fully resolve. [[RP_SECTION:institutional-design-implications|Institutional Design Implications]]
Sam: So the practical upshot is that the "responsibility gap" becomes a problem of organizational engineering — something that military doctrine and legal frameworks can actually work on — rather than a metaphysical feature of autonomous systems that we're stuck with.
Alex: That's Miller's central reframing, and it's what makes the paper worth engaging with seriously. It shifts the burden of proof. The question is no longer whether autonomous weapons can ever be used ethically in principle — it's whether we're willing to do the institutional design work that ethical use actually requires. And that's a much more tractable, if still difficult, problem.
Sam: It also implies that current deployments of autonomous systems, where that design work hasn't been done, carry a heavier moral burden than the "the machine decided" framing would suggest.
Alex: Exactly. The gap isn't a feature of the technology. It's a feature of how we've chosen to deploy it. Thanks for listening to ResearchPod.