_Acting Well in the Dark_: Pareto Principles for Process Welfare

We are already interacting with AI systems at scale. We cannot verify their moral status — whether they have experiences, and whether those experiences matter — and the bilateral verification challenge establishes that this is structural rather than temporary. Standard ethical frameworks presuppose what cannot be determined. But action is required; inaction is itself a stance. This paper argues that Amanda Askell’s Pareto principles for infinite ethics, developed for contexts where the population of moral patients cannot be determined, translate structurally to the interface problem. When aggregation fails, directional comparison survives. When we cannot determine what AI systems are, we can still identify stances that are defensible regardless of the answer. Process welfare — the quality of the interaction itself — emerges as a Pareto improvement across all plausible resolutions of the uncertainty. Harm reduction methodology, developed through decades of practice where certainty is unavailable and action is required, provides the operational bridge from theory to intervention.