Abstract: In recent years, with the rapid iteration of underlying chips and the rapid development of general technologies, the overall technical architecture of artificial intelligence has gradually become clear, becoming the core driving force for a new round of technological revolution and industrial transformation. National laboratories, technology leading enterprises, research universities, and other types of R & D institutions are the core contributors to AI research and innovation. What role do these organizations play in AI research and development? What are the existing R&D models? How can we achieve catch-up and surpass through R&D system layout? These critical issues urgently need to be addressed, and this paper attempts to conduct systematic research on this from a theoretical level. Existing research on R&D models focuses on three aspects: first, exploring suitable R&D models for key core technologies/high-tech breakthroughs from the perspective of technical breakthrough; second, analyzing latecomer enterprises' R&D models from the perspective of technological catch-up; third, identifying multinational organizations' R&D models from the perspective of international R&D. However, there are still certain theoretical gaps in these studies: On the one hand, most related studies focus on enterprise R&D organizations or industrial-academic-research cooperation, and have not yet conducted systematic mode analysis on different types of R&D organizations such as national strategic scientific and technological forces, leading to insufficient discussions and applicability of their R&D roles. On the other hand, these R&D models are mostly aimed at traditional technology fields and have limited guiding significance for emerging technologies such as artificial intelligence. The breakthrough of artificial intelligence technology urgently requires the exploration of new research and development models. This article constructs a theoretical analysis framework for the artificial intelligence research and development model from two dimensions: organizational structure and strategic orientation. Using the multi-case comparative method, this article finds that AI R&D organizations focus on different links of the innovation chain, and have different focuses on theoretical originality, key core technology research, open ecological construction, and industry frontier exploration, forming four types of R&D roles: "innovation explorer", "technology breakthrough leader", "ecological enabler", and "industry leader". Through the analysis of R&D organization’s portraits, research showed that there are differences in the roles of different R&D organizations. The conclusion provides a theoretical basis and practical basis for China to comprehensively promote the layout of artificial intelligence research and development systems at the level of national innovation system, and achieve high-level technological self-reliance and self-improvement through cohesion. Specifically, the diversified R&D model of artificial intelligence demonstrates that not only do various innovative subjects need to carry out in-depth research based on their own advantages and characteristics, but also different R&D models require mutual coordination, thus connecting the innovation chain and industrial chain of artificial intelligence, achieving a comprehensive layout from theoretical originality, key core technology research, open ecological construction, and industry frontier exploration.

摘要： 人工智能科技突破亟须探究研发模式系统布局。本文从研发组织方式和战略导向两个维度构建人工智能研发模式理论分析框架，运用多案例比较法，研究发现：人工智能研发组织聚焦创新链不同环节，在理论原创、关键核心技术攻关、开放生态建设、行业前沿探索上各有侧重，形成“创新探索者”“攻关主导者”“生态赋能者”“行业引领者”四类研发角色。通过对研发角色的画像分析，发现研发主体间存在角色差异性。研究结论对我国从国家创新体系层面推进人工智能研发系统布局、提升人工智能科技创新体系协同效能具有参考意义。