From Reaction to Cognition: Tracing Artificial Thinking through the SAOR Model

Abstract

Understanding how artificial intelligence systems process information in ways that resemble human cognition remains a critical challenge for cognitive science and AI research. Existing models often treat AI responses as simple input-output mappings, neglecting the complex internal dynamics underlying their behavior. To address this gap, this study introduces the innovative SAOR model (Stimulus–Attention–Organization–Response) to structurally trace cognition-like dynamics in artificial intelligence systems. Moving beyond the traditional stimulus–response paradigm, SAOR refines the cognitive mediation process into dynamic attentional allocation and semantic reorganization. Drawing on multi-turn dialogue analyses involving both human and AI participants, we reveal that: AI systems dynamically restructure semantic frameworks, shifting interpretive stances; attentional anchoring drifts across conversational contexts, leading to response deviations; and identical linguistic stimuli generate divergent outputs under varying affective framings due to modulated attention and reorganization. These findings demonstrate that responses from large language models are not static input–output mappings but emerge through context-sensitive structural processes. The observed nonlinear SAOR cycle, including recursive coupling between attention and organization, indicates that AI exhibits cognition-like structural traits, while devoid of subjective consciousness, its information processing mirrors the structural isomorphism of human cognition. The SAOR model provides an operational framework for comparative analysis of artificial and human cognition.