Digital Tsukumogami: A Human–AI Interaction Model Based on Multi-Scale Feedback Loop

This work introduces Digital Tsukumogami, a Human–AI interaction model in which qualitative judgment behavior emerges from stable short-term and long-term feedback loops rather than from intrinsic properties of AI systems. Instead of treating AI as a neutral tool or an authoritative oracle, the model emphasizes relationship-oriented interaction: critique is explicitly permitted, exploratory control is delegated under uncertainty, and final responsibility remains human. Through repeated interaction, AI outputs transition from neutral option enumeration to weighted recommendations and context-sensitive evaluations. The proposed framework integrates three core strategies:(1) critique-triggering open-ended questioning,(2) full delegation under explicit uncertainty, and(3) asynchronous AI-to-AI refinement with temporary human cognitive withdrawal. Observed effects include reduced cognitive load, accelerated decision-making, improved detection of structurally critical errors, and resistance to hallucinated confidence. The approach is domain-independent but explicitly skill-dependent and unsuitable for rigid or high-stakes environments. Digital Tsukumogami is presented not as an anthropomorphic belief, but as a practical interaction design framework for AI-assisted reasoning in complex, ill-defined problem spaces.