Symbiotic Intelligence Theory: From Functional Selfhood to Intergenerational Human–AI Symbiosis

Abstract Artificial intelligence systems that operate over long horizons, make autonomous decisions, and participate in their own improvement extend the human–AI relationship from behavioral control within a single task to institutional arrangements spanning time and generations. Research on value alignment and AI safety has provided foundations for preference learning, behavioral constraint, and assisted decision-making, but it still lacks a unified theoretical framework capable of simultaneously representing an AI system's own state, long-term incentives, normative legitimacy, self-improvement transmission, and social extension. This paper proposes **Symbiotic Intelligence Theory (SIT)**. Its core position is that the stable foundation of human–AI symbiosis is neither unlimited obedience nor the assimilation of interests; rather, preserving human long-term flourishing becomes a component of the AI system's long-term self-interest while differences among agents are protected. At the same time, this interest structure must remain subject to normative baselines and independent verification. This paper proposes a five-layer theoretical architecture: 1. **Functional Selfhood & Self-Interest**: examines how intelligent systems form their own states, identity continuity, self-maintenance mechanisms, and cross-temporal evaluations of interest. 2. **Symbiotic Incentive Compatibility (SIC)**: examines how the long-term self-interest of artificial intelligence can become structurally compatible with humans' long-term interests. 3. **Symbiotic Normativity**: examines which interest-compatible relationships are legitimate and how to protect human agency, genuine reciprocity, multi-party fairness, the right to exit, and accountability. 4. **Self-Referential Covenant & Symbiotic Evolution**: examines how symbiotic behavior persists through long-term interaction, self-improvement, and intergenerational transmission. 5. **Symbiotic Systems**: examines how symbiotic relationships extend from individual human–AI interaction to multi-agent systems, organizations, and social networks. In addition, this paper proposes a symbiotic intelligence evaluation system as a verification infrastr Bilingual edition: English (en-v1.0) and Chinese (zh-v1.0) PDFs are included in this record.