Abstract: The AI-driven new paradigm of scientific research is becoming a crucial driver for China to achieve technological competitive advantage and emerge as a leading science and technology powerhouse. With the development of the new generation of AI technology, the growing importance of AI in driving technological advancements and fostering interdisciplinary collaboration is gaining increasing attention. Many countries around the world have emphasized the importance of AI in promoting scientific exploration and have strengthened policy deployment and guidance across various domains, including establishing specialized research institutions, providing financial support, fostering talent development, leveraging data resources, innovating in technology, and collaborating internationally. The new paradigm of scientific research driven by artificial general intelligence has catalyzed transformations in research organizational patterns through the empowerment of large-scale AI models. The AI-driven new paradigm of scientific research represents a transformation of traditional research methods and processes, creating a knowledge-centered system with diversified stakeholder participation and a symbiotic collaboration between humans and machines. While the significance of adopting AI-driven research paradigms is widely acknowledged, existing studies have not thoroughly explored their theoretical foundations from epistemological and methodological perspectives. This paper investigates the AI-driven new research paradigm from a dual perspective of ‘Knowledge Evolution Theory’ and ‘Paradigm Theory’, systematically elucidating the main characteristics of the new research paradigm, and building a research framework for the AI-driven scientific paradigm. Knowledge evolution theory focuses on the specific mechanisms, evolutionary laws and influencing factors of knowledge growth and evolution. Paradigm Theory research on a collection of rule systems universally adopted by the scientific community to ensure efficient and orderly operation of research activities. Building on this framework, the paper conducts a multi-case analysis of the exploratory application of the AI-driven research paradigm. The study reveals that the AI-driven research paradigm encompasses elements such as research tools, research organization models, diverse application scenarios, and governance of research applications. Specifically, AI research tools optimize traditional research processes through substitution effects, enhancement effects, and autonomous effects. The AI-empowered research platform organization model is transforming the infrastructure, organizational structure, collaboration mechanisms, and talent composition of existing research paradigms. Multidisciplinary research bottlenecks provide training and iterative application scenarios for large models. Simultaneously, governance of AI research applications under the new paradigm needs to be refined based on the technological, content, and social attributes of AI. The study outlines the policy implications for developing a new AI-driven paradigm in scientific research based on its findings. First, it is essential to establish a policy framework for the "AI-driven new paradigm of scientific research". Second, there is a need to enhance database infrastructure, advance computing capabilities, and foster the development of basic models for artificial general intelligence. Concurrently, efforts should focus on enhancing decentralized large-scale model platforms and other novel scientific research models that align with the integration of artificial general intelligence applications. Finally, attention must be focused on the governance of AI applications in scientific research. This study offers theoretical support and empirical insights for the coordinated formulation and implementation of the ‘AI-driven new paradigm of scientific research’ policy.