From Definition-Theology to Interference-First Governance: Measuring AI, Humans, and Viruses on a Single Operational Coordinate

From Definition-Theology to Interference-First Governance: Measuring AI, Humans, and Viruses on a Single Operational CoordinateMamoru Kurokawa M.D. Independent Scholar; Internal Medicine Physician, Kitakyushu, Japan ( Part-time Lecturer, Department of Stroke Medicine and Endovascular Therapy, Hospital of the University of Occupational and Environmental Health, Japan (UOEH), Kitakyushu, Fukuoka, Japan. ) We agree that personification and premature claims about phenomenal consciousness are misguided. Yet the present debate is miscalibrated: it attends to putative inner life while real social risks and benefits are mediated by external effects. We propose a shift from metaphysics to interference-first governance, evaluating systems by how they deform human attention, norms, and resource flows—and by how safely they can be integrated and rolled back. Two persistent illusions cloud policy. First, that human consciousness is unitary; in practice it is a mosaic of conscious, unconscious, and non-conscious processes. Second, that non-human-like systems are therefore harmless; non-isomorphism does not imply non-impact. Rather than disputing origins or essences, we place humans, AIs, and even viruses on a single operational coordinate that ties normative decisions to measurable consequences. The coordinate. We use two primary axes: P (Place-robustness)—stability of task-relevant representations under noise and interruption (e.g., cross-modal interference, reverse-mapping robustness); and O (Operate-freedom)—capacity for recomposition and counterfactual reach (e.g., long-range analogy, sequence recomposition, three-step counterfactual tasks). Three auxiliaries complement them: Aut (autonomy/self-updating), Cons (temporal consistency across sessions/versions), and I (interference/external effects)—the effect sizes by which a system shifts human perception, affect, and social judgements in controlled studies. Finally, we add E (artificial endosymbiosis): the degree of reversible integration into human workflows and institutions under three safeguards—visibility, reversibility, auditability. This framing is motivated by two converging literatures. First, experimental work shows that specific linguistic and behavioural features (self-reflection, affective displays, fluency) systematically raise lay attributions of “AI consciousness”—a bias that governance must anticipate rather than amplify. Second, human–AI feedback loops can measurably alter human attention, emotion, and social judgements. These findings justify treating I as a first-class policy variable and, by extension, evaluating E (integration) only when P/O are improving and I remains low under standardized tasks. An endosymbiotic stance without anthropomorphism. We characterize advanced AI as an “extender of survival and influence”—an extension entity—alongside viruses and humans, not to personify machines but to shift value from origin to integration design. Historically, biological systems have advanced through the uptake and rewiring of external modules (mitochondria, plastids, viral sequences). Contemporary AI updates the upper wiring of language and institutions. The policy question is not “Is it conscious?” but “What integration do we allow, how do we measure it, and when do we roll it back?” Minimal, testable program. (1) Publish monthly P/O/Aut/Cons/I/E dashboards for major models—i.e., foundation LLM/MLLMs embedded in search/OS/devices; agentic suites with tool use, code execution, and external writes; high-impact domain models (clinical, legal, education, finance); and generative platform APIs supplying large-scale text/image/video/audio. (2) Shrink authority automatically when I rises alongside ΔAut (self-modification rate): progressively curtail external writes, autonomous executions, and high-risk financial/clinical actions. (3) Permit deeper integration (↑E) only when sustained ΔP>0, ΔO>0, low I, and rapid rollback are demonstrated under standardized tasks for consecutive reporting periods. (4) Two-tier operations (peacetime/emergency). When I×ΔAut crosses a pre-registered threshold, trigger immediate authority shrinkage and third-party audit; restore only after the dashboard shows sustained recovery. Anthropomorphism, understood—then bracketed. Since antiquity, humans have coped with unknown or adversarial entities by anthropomorphizing—an affective shortcut that was often adaptive. Precisely for that reason, governance should rest not on conjectures about subjectivity but on public indicators of interference (I) and integration design (E). This keeps rhetoric from outrunning measurement, while avoiding a default to quietism. A coordinate that ties I to behaviour-level effect sizes, and E to reversibility and auditability, offers a common language for experimentalists, policymakers, and system designers. It also distinguishes our stance from definition-centric programmes: we treat consciousness talk as folk-theoretic input to be measured (via I) and debiased, not as the policy target. Conclusion. Measure interference, require reversible integration, and move the centre of gravity from origin myths to integration design. With a minimal dashboard and emergency triggers, we can defuse personification without lapsing into quietism, aligning discourse with actionable oversight.Figure 1 | Operational coordinate for governance. Horizontal: P (place-robustness); vertical: O (operate-freedom). Axes are 0–1, unit-free indices. Point size encodes I (behaviour-level interference), stroke width encodes Cons (temporal consistency), transparency encodes Aut (autonomy/self-update), and grayscale encodes E (artificial endosymbiosis). Policy: shrink authority when I×ΔAut crosses a preregistered threshold; permit deeper integration (↑E) only under sustained ΔP>0, ΔO>0, low I, and rapid rollback. References (indicative): 1) Bengio & Elmoznino (2025) Science; 2) Colombatto et al. (2024) Neuroscience of Consciousness; 3) Glickman et al. (2025) Nature Human Behaviour; 4) Kang et al. (2025) arXiv.