双重加工视角下的人机协同认知:基于大语言模型的动态认知分工OA

Drawing on dual-process theory,this paper examines how reasoning-enhanced large language models(LLMs)can support human-machine collaborative cognition in educational contexts.Supported by mechanisms such as test-time computation,re-inforcement learning,and chain-of-thought reasoning,LLMs are able to combine rapid response generation with stepwise reasoning,thereby displaying processing tendencies analogous to System 1 and System 2.This dual tendency challenges the conventional division of cognitive responsibility,in which AI is expected to provide information while humans engage in higher-order thinking,and gives rise to emerging concerns such as cognitive outsourcing,capability substitution,and collaborative redundancy.By clarifying the func-tional correspondence between these two processing tendencies and their educational implications,this paper develops a human-ma-chine collaborative cognitive model tailored to educational contexts.The model is organized around four dimensions:collaborative principles,learning goals,interaction mechanisms,and safeguards.Based on task structure,learning phase,and cognitive load,the model proposes the conditional activation,intensity regulation,and gradual fading of"fast-answer"and"deep-reasoning"modes.At the interaction level,it incorporates metacognitive mechanisms that enable learners to monitor,question,and reprocess AI-generated reasoning.Building on this framework,the paper discusses its application in representative instructional scenarios,including struc-tured STEM problem solving,argumentative writing,inquiry-based learning,and real-time classroom diagnosis.It argues that LLMs should not only serve as answer-generation tools,but should function as explainable,controllable,and gradually fading cognitive scaf-folds.The study concludes that the design of human-machine collaborative learning in the AI era should prioritize the development of learners' System 2 capacities,while enhancing task efficiency and sustaining learners' independent thinking,argumentative reflection,and transfer of learning.