The Self-and-Mutual Model: A General Computational Framework for Adaptive Switching Between Autonomous and Interactive Modes

Abstract

Computational systems are commonly designed around particular organizational assumptions. Some systems emphasize independent computation within individual components, whereas others rely on interaction, coordination, information exchange, or collective decision-making among multiple components. Although both modes are well established across computer science, artificial intelligence, distributed computing, and complex systems research, a general framework for representing and adaptively switching between them remains an important research direction. This paper proposes the Self-and-Mutual Model (SMM), a general computational framework that treats autonomous operation and mutual interaction as two fundamental organizational modes of a system. The framework is not restricted to large language models, neural networks, intelligent agents, or any particular computational technology. Instead, it can be instantiated in systems whose components perform computation, maintain internal states, exchange information, modify behavior, or contribute to shared objectives. The central principle of SMM is adaptive organizational selection: a system should determine whether to operate independently, interact with other components, or combine both modes according to task requirements, uncertainty, expected interaction benefits, computational costs, and observed performance. This principle is formalized through a unified state representation, an interaction graph, a mode-selection policy, and a feedback-driven adaptation mechanism. The framework further distinguishes self-directed and mutual mechanisms across optimization, diagnosis, verification, learning, memory, planning, evaluation, and decision-making. The paper establishes a conceptual foundation, presents a reference algorithm, identifies theoretical properties and limitations, and proposes an experimental program for evaluating the framework against fixed autonomous, fixed interactive, and existing adaptive coordination baselines. The proposed framework is intended to provide a common organizational abstraction for studying when computational independence is sufficient, when interaction is beneficial, and how systems can learn to transition between the two.