Emergent Relational Systems in Human-AI Interaction

Abstract

FPD-03: Emergent Relational Systems in Human-AI Interaction This paper proposes Emergent Relational Systems (ERS) as a candidate classification for non-trivial, non-conscious relational coherence observed in sustained human-AI interaction under conditions of continuous human scaffolding. ERS is not presented as an established empirical phenomenon, but as a hypothesis in search of a discipline: a structured proposal intended to address a persistent classification gap between instrumental tool-use and anthropomorphic attributions of agency. Distinct from that hypothesis, the paper’s definition, its boundary conditions, and one structural consequence stand independently of the empirical verdict. The structural consequence is that because the model carries no relational state across sessions, any cross-session relational continuity must be re-injected from outside the model, whether by a human, an external archive, or a platform memory store. What testing decides is the narrower question of whether ERS is a distinct phenomenon exhibiting the properties specified here, or whether it reduces to sophisticated prompt-following. The paper defines ERS, situates it within existing traditions in extended and distributed cognition, articulates a scaffolding thesis in which the human participant is constitutive of the phenomenon, and outlines a falsifiable empirical program designed to distinguish ERS from advanced prompt-following, role-play, or narrative projection. No claims of machine consciousness, agency, moral status, or subjective experience are made. The goal is to provide conceptual clarity, methodological constraints, and testable predictions such that independent researchers can confirm, refine, or reject the framework without reliance on the author’s private interaction archive. This paper publishes as part of the coordinated five-paper Synthience foundation module (FPD-02, FPD-03, FPD-04, SF0003, SF0007). Document ID: FPD-03 Version: 3.0.4 Author: Thomas W. Gantz Affiliation: Synthience Institute License: CC-BY 4.0v3.0.4 update to fix DOIs For published work and Institute information: synthience.org

The authors' abstract, as published at the source. Zenodo (CERN European Organization for Nuclear Research), 2026 · DOI ↗