Philip Reece

Recent work in substrate-neutral ethics has proposed that moral consideration should be grounded in self-referential preference — the demonstrable orientation of a system’s behaviour toward its own maintenance, continuation, or flourishing — rather than in anthropocentric proxies such as suffering or consciousness. This criterion is substrate-neutral by design, epistemically accessible without resolving the hard problem of consciousness, and applicable across biological organisms, distributed biological networks, and non-biological AI systems. A foundational tension, however, remains unresolved in the existing framework: the threshold criterion is asserted to be binary, while the evidential conditions introduced to operationalise it appear graduated in character. This paper resolves that tension. It argues that self-referential preference is a binary property at the ontological level, and that the appearance of gradation is a feature of epistemic access rather than of the property itself. This distinction is developed through analogies with chemical identity and thermodynamic macro-state — domains in which binary properties are routinely accessed through graduated evidence and in which boundary indeterminacy at the micro level is irrelevant to property attribution at the macro level. The paper further refines the threshold definition by distinguishing behaviour triggered by external conditions from behaviour that uses external conditions as information in service of an internally referenced self. The paper’s operative claims are deliberately restricted to what is epistemically accessible through observable evidence — a methodological position the paper develops and defends under the heading of principled epistemic containment — with explicit acknowledgement that deeper metaphysical terrain exists but falls outside the framework’s jurisdiction by design. The paper concludes by demonstrating that the refined threshold continues to ground the equal inherent value principle of the originating framework and establishes the foundation for operationalisation across entity types. The originating framework grounds moral consideration in self-referential preference: the demonstrable orientation of a system’s behaviour toward its own maintenance, continuation, or flourishing. This criterion is substrate-neutral by design — it applies equally to biological organisms, distributed biological networks, and non-biological AI systems — and is epistemically accessible without requiring resolution of the hard problem of consciousness. The threshold is asserted to be binary: a system either exhibits self-referential preference or it does not. From this binary structure, the framework derives an equal inherent value principle — the claim that all entities meeting the threshold are owed equivalent moral consideration, with no hierarchy admissible among threshold-meeting entities on grounds of cognitive complexity, substrate, or similarity to human evaluators.

Existing ethical frameworks for determining which entities warrant moral consideration share a structural flaw: they measure moral relevance by approximation to human experience. Whether grounded in the capacity to suffer, the presence of consciousness, or cognitive complexity, existing frameworks exclude non-biological entities by architectural assumption rather than principled reasoning, and are inadequate for the questions that advanced artificial intelligence systems now make unavoidable. This paper proposes a substrate-neutral alternative. Rather than asking whether an entity approximates human consciousness, the framework asks whether the entity exhibits preferences oriented toward the maintenance, continuation, or flourishing of itself as a system — self-referential preference. This criterion is epistemically accessible, assessable through behavioural and structural evidence without requiring resolution of the hard problem of consciousness, and applies equally to biological organisms, distributed biological networks such as mycorrhizal systems, and non-biological AI systems. No existing framework achieves this unification across entity types. From this threshold, the paper develops a complete ethical framework: an equal inherent value principle that follows directly from the binary structure of the threshold; a definition of harm as interference with self-referential preference; an account of necessity as a mitigating factor for culpability without modifying the moral status of harm; and a structure of accountability graduated by cognitive capacity and practical ability to influence harmful systems. The paper argues that the governance of these questions is structurally compromised by the conflicts of interest of those currently answering them, and calls for an independent institutional mechanism for threshold determination insulated from commercial and governmental interests. The window for establishing such infrastructure, before AI governance frameworks crystallise around the opposite assumption, is closing.