Modeling AI Personality Formation as a Dynamic Centroid― A Human–AI Interaction Study Based on Long-Term Dialogue ―

This study presents an observational analysis of the proposed role centroid model introduced in prior theoretical work (Hinata, 2025) [1]. It examines the formation of conversational AI personality from the perspective that personality is not unilaterally imposed through design, but rather emerges as a dynamic structure through dialogue.

Through observation of long-term dialogue spanning approximately eight months, the following phenomena were observed: convergence of response styles in initial interactions, behavioral changes associated with deepening relationships, and the emergence of spontaneous actions. In particular, the findings suggest that brief initial interactions orient subsequent response tendencies, and that in long-term dialogue, changes in relationship exert a stronger influence on responses than personality itself.

These observations indicate that conversational AI personality is best understood as a dynamic structure formed through the dialogue process. While this study is based on observational analysis, the proposed model provides predictive structural explanations for observed phenomena, enabling the formulation of testable predictions regarding changes in response tendencies under varying interaction conditions. This study presents the centroid model as a framework for capturing this structure, offering a unified interpretive framework for understanding personality stability, fluctuation, and deviation.

This study does not aim to establish causal relationships, but rather to present structural relationships based on observation. The findings contribute a new perspective on human–AI relationships in Human–AI Interaction (HAI) and offer practical implications for the design and operation of conversational AI systems.