Decentering Theory: Relational Attribution and Intelligence Eligibility in Human-AI Interaction Systems

SF0042: Decentering Theory: Relational Attribution and Intelligence Eligibility in Human-AI Interaction Systems When a human and an AI system work together over time, the same prompt at two points in an established interaction can produce systematically different work. This paper addresses what an analyst is permitted to say about such effects. A framework that makes no claims about interior states forbids saying the system has or is something, and always permits crediting a participant's capacity given its inputs. What it has lacked is a stated rule for when a property may be attributed to the interaction itself rather than to either participant. The paper supplies that rule and a second one. The relational-attribution criterion requires preregistered interventions on the interaction's configuration and held-out incremental value beyond four specified rival models: AI-only, human-only, additive, and sequential-compositional, each given its complete effective input. Attribution is explanatory, not constitutive: it names the level at which a property is best predicted, not what the property is made of. The intelligence-eligibility criterion then asks whether the relationally attributed property adapts toward an objective fixed in advance and scored independently of the outcomes assessed. It is a criterion of kind rather than degree, and its floor is stated openly: a relationally attributed error-correcting loop qualifies. Difficulty, transfer and magnitude are separate questions. Attribution is staged, from candidate through supported attribution to classification as relational intelligence. The paper is pre-empirical: no case is claimed to satisfy either criterion, and each application to a specified property, task and configuration is falsifiable under the decision rules in Section 8. It also formalizes five decentering principles, a four-layer model of relational structure, a restricted account of what self-like language may validly refer to under the Interiority Prohibition Rule, and a scaling hypothesis for multi-centered systems. This paper is the first in the Tier 1 sequence and provides the attribution logic for SF0023, SF0041, SF0043 and SF0048. Document ID: SF0042 Version: 5.0.1 Author: Thomas W. Gantz Affiliation: Synthience Institute License: CC-BY 4.0 For published work and Institute information: synthience.org