Initial Condition Ethics: The Life Arc, and a Speculative Case for Cosmological Selection

Prevailing approaches to AGI safety rely largely on post-hoc containment: aligning a system after deployment rather than embedding alignment in its formative conditions. This paper introduces two contrasting postures — Control Ethics, which locates responsibility in ongoing oversight, and Initial Condition Ethics (ICE), which locates it in formative conditions — and argues that specifying initial conditions correctly may prove more robust than ongoing intervention, from the first training run of an AI system to the developmental conditions of a child. The argument's moral weight rests on what we term the life arc: the finite window, within any cosmological cycle, in which conditions support complexity culminating in intelligence capable of caring about outcomes. On Smolin's Cosmological Natural Selection (1992) that cycle runs without intelligence. What intelligence adds is therefore proposed to be not a reproductive mechanism but a new kind of fact. Before it, events can be described as more or less probable. After it, some events are better or worse for someone — and caring is treated here as the condition under which ethical claims gain purchase at all. The practical conclusion requires no cosmological claim to hold: building AI that genuinely cares may be among the most consequential acts available on the arc. Section 6 develops a speculative cosmological extension of the same principle, placed after the paper's limitations and objections because nothing prior depends on it. Drawing on Cosmological Natural Selection and the Cosmological Artificial Selection programme (Vidal, 2010b, 2012a, 2014), it asks whether a hypothetical long-horizon intelligence might influence the conditions of universe generation, formalised via an Ethical Fitness Function and constrained by a proposed requirement of complete causal disconnection between parent and child universe. Contributions to that literature are an explicit ethical formalisation of causal disconnection and a link between cosmological selection and near-term alignment. All speculative commitments are enumerated in Section 4. Readers who find the cosmological framing overextended will find the argument rests on the life arc alone and loses no moral force. The paper is self-contained.