This paper introduces Relational State Intelligence (RSI), a state-based paradigm for relationally conditioned execution in intelligent systems. Conventional artificial intelligence systems are fundamentally output-centric, optimizing accuracy, reasoning performance, and task completion under the implicit assumption that execution is the default consequence of intelligence. As AI systems increasingly operate within persistent Human–AI, AI–AI, and distributed multi-agent environments, this execution-as-default assumption produces structural limitations including over-execution, excessive intervention, recursive instability, inappropriate timing, and autonomy erosion. The paper demonstrates that these failures arise because contemporary AI systems lack a principled mechanism for determining whether execution itself should occur under relational conditions. To address this limitation, the paper formalizes Relational State Intelligence (RSI) as a paradigm in which intelligent behavior is governed through relational state transitions rather than output optimization alone. The framework establishes: relationally conditioned execution, unified relational control states, state-based execution governance, dynamic evaluation criteria, relationally governed intervention, and non-execution as a valid intelligent outcome. Under RSI, execution decisions are determined through relational state, relational dynamics, relational history, and temporally evolving interaction structures, producing outcomes including: Execute Delay Hold Delegate Do Not Act Crucially, RSI reframes intelligence from action generation toward the regulation of action permission under relational conditions. The proposed framework introduces a model-independent relational pre-execution layer operating prior to inference and execution, enabling execution governance across Human–AI interaction, AI–AI interaction, autonomous workflows, and distributed orchestration systems. This publication serves as a foundational intelligence theory paper for the Relationship-Aware AI Research initiative, establishing Relational State Intelligence (RSI) as a foundational state-based theory of intelligence for relationally conditioned execution.
HARUKI ITO (Thu,) studied this question.