Answers are grounded in this paper and cite their sources.
Existing research on artificial intelligence reasoning has investigated individual model reasoning, adaptive inference-time computation, multi-agent collaborative reasoning, and human–AI reasoning. However, these research directions are frequently examined as separate capabilities or interaction configurations, leaving an important theoretical opportunity for a unified architecture that can dynamically determine whether a reasoning task should be performed by an individual artificial intelligence model, multiple collaborating models, human participants, or combinations of these participants. This paper proposes an integrated extension of the Universal Self-Mutual-Human Model (USMHM), termed the Adaptive Self-Mutual-Human Reasoning Framework (ASMH-RF), to establish a general architecture for adaptive reasoning across heterogeneous human and artificial participants. The framework distinguishes three fundamental reasoning dimensions: Self-Reasoning, in which an individual model or participant performs reasoning independently; Mutual Reasoning, in which multiple models or participants exchange information, distribute reasoning responsibilities, challenge intermediate conclusions, and jointly construct solutions; and Human Reasoning, in which human individuals or groups independently perform or actively contribute to reasoning processes. Its central contribution is not simply the coexistence of these three reasoning mechanisms, but their integration into a unified adaptive decision architecture. The framework enables a reasoning system to determine when to employ individual reasoning, when to activate collaborative reasoning, when to involve human reasoning, and when to combine or dynamically switch among these configurations. Such decisions can be informed by task complexity, uncertainty, confidence, reasoning performance, knowledge requirements, participant capabilities, computational resources, risk, and the need for human judgment. The framework further distinguishes adaptive reasoning-depth adjustment from adaptive mechanism selection, dynamic switching, adaptive combination, and structural reconfiguration. It introduces a formal model, adaptive decision policy, transition architecture, operational lifecycle, governance principles, theoretical propositions, and empirical evaluation methodology. The framework also supports repeated transitions, including Self-to-Mutual, Mutual-to-Human, Human-to-Self, and hybrid configurations, without assuming that any single reasoning mode is universally superior. To the best of our knowledge, this paper is the first to formally propose a unified Self–Mutual–Human reasoning architecture that integrates adaptive selection, dynamic switching, combination, and reconfiguration of reasoning mechanisms within a general mechanism-independent theoretical framework. The proposed approach establishes a conceptual foundation for developing reasoning systems that can adaptively determine not only how deeply to reason, but also who should participate in reasoning and how reasoning responsibilities should be organized.
No takes yet. Share an insight, caveat, or question.
Jincheng Zhang (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: