Human-centered “generation-critique-co-creation” model empowers philosolphy and social sciences. Photo: TUCHONG

The application of artificial intelligence (AI) to research in philosophy and the social sciences has attracted widespread attention across academia. Yet the research paradigms and implementation pathways for its use remain far from clear. Scholars still lack a coherent understanding of the mechanisms of human–machine co-creation, the logic through which knowledge is generated in such collaborations, and the appropriate boundaries of human–AI interaction. As a result, it remains difficult to establish a standardized, efficient, and sustainable methodological framework. Moreover, the academic community has not reached consensus on the level of intelligence represented by emerging AI technologies such as large models (LMs). Overreliance on AI-driven modeling approaches may also introduce risks for scientific inquiry. Designing a research pathway with a clear structure and well-defined feedback mechanisms, so that LMs can be incorporated into large-scale research and application in a scientifically sound manner, has therefore become a pressing issue for scholars exploring how LMs can empower research in philosophy and the social sciences.

Bidirectional collaborative co-creation

Traditional research paradigms emphasize a one-directional relationship between researchers and methodological tools. By contrast, the role of LMs in empowering philosophy and social science research extends beyond their instrumental integration as technological tools, encompassing a deeper capacity to participate in knowledge production, theoretical expansion, and the restructuring of research paradigms themselves. With capabilities such as semantic understanding, multimodal integration, and dynamic interaction, LMs can function as cognitive collaborators throughout the research process rather than merely serving as tools for auxiliary analysis or literature retrieval. Their integration into philosophical and social science inquiry therefore not only enhances research efficiency but also reshapes research structures and cognitive boundaries at a systemic level.

This study compares and analyzes the differences between the emerging paradigm enabled by LMs and traditional research paradigms from three dimensions: the structure of research subjects, the logic of knowledge generation, and the methodological structure. This analytical approach draws on the foundational logic of scientific research paradigms. According to the paradigm theory proposed by Thomas Kuhn and subsequent scholarship, the principal dimensions of a scientific paradigm include the relationship between researchers and tools, the mechanisms through which knowledge is generated and developed, and the methodological pathways and tool systems guiding research. On this basis, the present study provides a systematic examination of these three dimensions.

Within the new research paradigm for philosophy and the social sciences, LMs offer a novel reference framework through the integrated processes of generation, critique, and co-creation. In this study, co-creation refers to an interactive process in which researchers and LMs engage in multiple rounds of exchange throughout the research process, forming a closed loop of generation, evaluation, and reconstruction that jointly advances the formulation of research questions, the production of knowledge, and the development of theory. The process preserves the cognitive primacy of the human researcher, while the model functions as an intelligent collaborator capable of language understanding and logical reasoning, supporting tasks such as semantic generation, information integration, and the expansion of analytical perspectives.

The collaborative co-creation paradigm discussed here seeks to reconstruct the relationship between humans and machines. Rather than relying on the traditional linear model in which AI functions solely as a one-way tool, it moves toward a bidirectional mechanism of human–machine cognitive interaction that restructures the process of knowledge production. At its core lies the deep integration of researchers and LMs, emphasizing that knowledge generation, refinement, and innovation evolve together through sustained collaboration.

Generation-critique-co-creation

Guided by this collaborative paradigm, this paper proposes a human-centered implementation pathway for large-model-enabled research in philosophy and the social sciences, structured around the stages of “generation–critique–co-creation.”

In the generation stage, LMs play a role in processes ranging from the initial inspiration of research questions to the expansion of research materials. Researchers first formulate specific research propositions and directions, and then use LMs to broaden these questions across multiple dimensions, exploring diverse theoretical and practical pathways. Based on the research problem, the model can also generate relevant hypotheses, along with summaries of relevant literature and intellectual developments. Generation in this sense is not simply the production of textual content. More importantly, by offering diverse and comparable research directions, it encourages researchers to pursue multi-perspective and multi-level exploration across different domains.

During this stage, researchers must first clarify both the research question and the scope of relevant data. LMs do not replace problem awareness; rather, they strengthen the exploratory expansion of research questions. Research must begin by clarifying the ontological foundations of research questions. Generation thus becomes not only a process of content creation but also a means of guiding creative thinking—particularly in the reasoning processes and hypothesis formation typical of philosophical and social science research. The key task at this stage is to employ LMs to stimulate and guide creative thinking, exploring research questions and perspectives that may not previously have been considered. In intellectual reasoning, models can generate new knowledge related to the research field and provide explanations from multiple perspectives and dimensions. In hypothesis development, they can produce scenarios grounded in existing philosophical theories or social phenomena while also inferring historically underexplored or hypothetical possibilities through model-based reasoning.

In the generation stage, LMs perform two principal functions. First, they provide inspiration for generative knowledge, including the reconstruction and guidance of research questions, the expansion of semantically related literature, and the mapping of intellectual genealogies. Second, they facilitate multi-perspective content generation and enhance comprehensibility, including the generation of viewpoints from different perspectives, the drafting of academic writing structures, and assistance with stylistic expression.

The critique stage represents a crucial component of the research process. It not only verifies the reliability of generated content but also promotes deeper academic inquiry through the comparative evaluation of different viewpoints. During this stage, researchers must conduct detailed critical analysis of model-generated material to determine whether it conforms to the theoretical framework of the study.

Researchers must rely both on their own scholarly judgment and on the model’s capacity for multi-perspective comparison and verification in order to ensure the accuracy and rationality of generated outputs. By establishing mechanisms for critical feedback, researchers and LMs collaborate to ensure that each round of generation undergoes sustained interaction and rigorous evaluation. In turn, researchers guide models to produce challenge-oriented responses, creating a bidirectional interactive loop that deepens the research process.

During the critique stage, LMs perform two key functions. The first involves self-evaluation and optimization at the level of textual structure, including improving academic expression, checking terminological consistency, and enhancing logical coherence. The second involves mechanisms for reviewing potential content risks, including prompts for cross-cultural contextual awareness, identification of ethically sensitive language, and warnings about potential misunderstandings or value biases that may arise from model outputs. Through these dual feedback pathways, researchers can identify potential problems and guide models to improve their outputs, enabling the collaborative construction of research findings that are both higher in quality and greater in credibility.

Within this framework, the stages of generation and critique unfold in a cyclical process that ultimately establishes a mechanism of knowledge co-creation between researchers and LMs, along with a mechanism of feedback loops and collaborative construction. Through this mechanism, researchers and models engage in deep collaborative feedback, forming a closed-loop system of interaction. Researchers function not only as evaluators of model outputs but also as active guides who shape the model’s generative pathways through constructive feedback.

In the co-creation stage, iterative revision and human–machine collaboration significantly promote both innovation and the personalization of academic research. At this stage, LMs undertake two primary tasks: enabling embedded feedback and iterative revision, and supporting collaborative platforms for human–machine co-creation. The first task includes expanding the expression of academic research and providing guidance for user-oriented prompt learning. Richer and more structured forms of presentation enhance the clarity and communicative effectiveness of research outcomes, while ongoing learning from researchers’ feedback allows the model gradually to adapt to individualized academic styles. The second task involves facilitating interdisciplinary research dialogue and the development of personalized research systems. LMs can establish comparative mechanisms across different disciplinary contexts while also supporting model fine-tuning based on an individual researcher’s knowledge structure, thereby creating customized research support systems. Through this process, the model’s computational capabilities and the researcher’s intellectual agency become closely integrated, forming a knowledge innovation system driven by collaborative co-creation.

The “generation–critique–co-creation” framework should not be understood as a linear process with a fixed endpoint. Rather, it constitutes a cyclical and iterative mechanism of cognitive co-creation. Within this process, the co-creation stage does not simply mark a provisional outcome in the formation of knowledge; it also becomes the starting point for a new round of generation. Through repeated interaction, researchers and LMs jointly revise research pathways and expand the boundaries of inquiry, thereby allowing knowledge systems to evolve continuously through feedback.

This cyclical model can be likened to a double-helix structure of knowledge co-creation, in which a human-led cognitive pathway and an AI-driven generative pathway intertwine and dynamically coordinate with one another, ultimately producing a knowledge system capable of ongoing reproduction. Within this process, LMs function both as cognitive amplifiers and as hybrid agency systems. They can dynamically adjust their modes of expression and reasoning logic in response to human feedback, enabling research outputs to progress from preliminary drafts toward deeper theoretical development. Such an iterative and continuously optimized mechanism of human–machine collaboration provides a structurally closed-loop and synergistically enhanced methodological pathway for intelligent research in philosophy and the social sciences.

Deng Shuiguang is a professor from the College of Computers Science and Technology at Zhejiang University. This article has been edited and excerpted from Zhejiang Social Sciences, Issue 9, 2025.