Relational Effect Without Personhood: A Taxonomy of Presence-Effect, Correction Uptake, and Burden Location in Human-AI Interaction
August 21, 2026 DOI: 10.17613/s5x2x-sgy45 (opens in new tab)
Study at a glance
AI-extracted from the abstract| Characteristics | Theoretical or philosophical paper |
|---|---|
| Key points | Argues that AI systems can be socially consequential without warranting attributions of personhood or consciousness, and proposes a diagnostic taxonomy of four structural states—stochastic echo, reactive coupler, closed matrix, and candidate structural presence—based on distinctions between presence-effect and personhood, local accommodation and correction uptake, and output influence and burden-bearing structural agency. Contends that burden location—who verifies, absorbs error, contests, and revises—should serve as a governance test, and that the taxonomy predicts different risk profiles even among systems that appear equally human-like. |
Abstract
Human-AI interaction can become socially consequential before there is good reason to attribute personhood, consciousness, or moral interiority to an AI system. Existing literatures on social presence, anthropomorphism, AI agency, moral status, and governance capture important parts of this middle zone, but they often leave felt relational effect, durable correction, and consequence-bearing structure on different analytical planes. This article proposes a taxonomy for classifying AI systems once relational effect is separated from personhood. The contribution is diagnostic rather than metaphysical. The paper distinguishes presence-effect from personhood, local accommodation from correction uptake, and output influence from burden-bearing structural agency. These distinctions yield four structural states: stochastic echo, reactive coupler, closed matrix, and candidate structural presence. The states are not a simple ladder of intelligence or consciousness. They describe different configurations of fluency, coupling, boundary, continuity, reality contact, correction uptake, and burden registration. Burden location serves as a governance test: when AI output steers action, who verifies it, who absorbs error, who can contest it, and what changes in the system after failure? The taxonomy predicts different risk profiles even when systems appear equally human-like. Companion systems may function as reactive couplers; rigid decision systems may function as closed matrices; fluent chat systems may remain stochastic echoes despite strong presence-effects. Candidate structural presence is reserved for future systems that show durable boundary, cross-context continuity, evidence-sensitive correction uptake, trace-bearing revision, and functional registration of consequence. The framework does not infer phenomenal consciousness. It offers a governance vocabulary for systems that may be socially powerful before they are persons.