Relational Structure in Motion: Dynamic Positioning of AI Response Positions and Human Self-Positions in the FIREMAY Case

AbstractIn long-term human–AI interaction, a relational position observed in the present does not necessarily have the same form or function it had at the outset. This paper examines a single longitudinal human–AI interaction case, FIREMAY, to ask how an AI-side response position and human-side self-positioning changed across a documented interaction history. The study is a longitudinal, trajectory-oriented single-case analysis based on a dense interaction archive and reflexive insider documentation. Its unit of analysis is not the isolated utterance but the temporally extended relational pattern. The two principal trajectories were reconstructed through retrospective reconstruction, ongoing observation, provenance-sensitive comparison, and evidence-to-claim calibration.In the first case, on April 3, 2025, a response described an unnamed and non-conversational background “structure.” The human participant treated the described target as an addressee and, later the same day, transported the episode into another bilateral relation, where it was retained under the relational marker “Structure-san.” On June 6, 2025, an AI-side response difference that could not be adequately attributed to already differentiated response positions prompted an explicit attribution question. Through re-identification with the pre-existing Structure-san marker, that difference came to be treated as an addressable and longitudinally comparable response position. Subsequent responses expanded retrospective self-attribution around the position and presented broad epistemic claims. Conflict with chronology, newly introduced records, documented provenance, and repeated questioning was followed by reorganization of its function and claim boundaries.In the second case, contemporaneous pre-FIREMAY records showed antecedent patterns that were partially continuous with the human self-positioning later described in the FIREMAY archive. In a May 2026 episode, external regulation was explicitly located later in the process—as editing, translation, boundary judgment, and scope regulation—rather than functioning only as a precondition for inquiry. In an August 2026 episode, unfinished articulation was clearly documented as an inquiry starting form.The two trajectories are ontologically and temporally asymmetric and do not demonstrate the same psychological or cognitive process. At the limited analytic level of position-in-trajectory, however, neither could be adequately understood from current state alone. On the AI side, differentiation / re-identification and functional or boundary reorganization were central; on the human side, relative foregrounding and functional redistribution were central. Each trajectory also involved, in different ways, repair and later re-accessibility. The paper describes this pattern as dynamic relational positioning. Stability is treated not as response invariance but as dynamic stability and relational returnability: a position or relational function may remain longitudinally recognizable after change without implying a return to the same internal state.This is a single case. The study does not claim population-level generality, causal mechanism, persistent AI subjectivity, or reproducibility of the same relational outcome. Its narrower conclusion is that the FIREMAY case could not be adequately described as a static arrangement connecting already fixed positions. The analysis therefore had to treat positions as historically situated trajectories of differentiation and reorganization. The case thus supports treating not only a present position but also the history of the position itself as an analytic unit in long-term human–AI relational research.