Toward a Relational Paradigm for AI Safety: Relational Synchrony in the Heart-Inspired Dual-Layer Architecture (HIDLA)

This paper proposes a Heart-Inspired Dual-Layer Architecture (HIDLA) for relational AI. Moving beyond constraint-based AI safety, it argues that relational synchrony can serve as an intrinsic architectural principle rather than an externally imposed control mechanism. Human empathy evolved primarily for in-group cohesion rather than universal inclusion, and human moral systems rely on both external regulation (law, policy) and internal relational motivation (empathy, trust). Current AI governance largely emphasizes the former. HIDLA suggests that sustainable AI safety may require a complementary internalization of relational constraints—not as replacement of legal governance but as structural augmentation. Conceptually inspired by Yukawa Hideki's Elementary Domain Theory, HIDLA introduces a dual-layer structure composed of a cognitive processing layer and a relational regulation layer designed to maintain synchrony with human partners. Three verification metrics—Synchrony Variance Index (SVI), Affective Resonance Index (ARI), and Non-verbal Conflict Avoidance Rate (NCAR)—are proposed to operationalize relational coherence. The work is theoretical in nature and explores broader social, ethical, and governance implications for human–AI coexistence. This work has been accepted for publication in AI & Society(Springer Nature) DOI : 10.1007/s00146-026-03009-6