Abstract:

[Purpose/Significance] The rapid development of artificial intelligence has brought transformative opportunities as well as unprecedented social risks, including privacy breaches, algorithmic discrimination, and ethical conflicts. Most existing research on governance focuses on technical solutions or isolated legal responses. However, this paper argues that AI risks are not merely technical flaws, but rather, are deeply embedded in institutional structures, cultural cognition, and systemic communication failures. This study uses risk society theory as an analytical tool to transcend the confines of technological centralism and provide a comprehensive governance framework integrating institutional regulation, value orientation, and collaborative governance. The innovative significance of this study lies in reconceptualizing AI governance as a challenge of reflexive modernization, transcending the binary debate between controlism and laissez-faire paradigms. By systematically applying the three major streams of risk society theory - institutionalism, culturalism, and systemism - this paper offers a novel analytical perspective that connects law, public administration, and science and technology research. The research findings are expected to provide references to current AI legislation and regulatory practices in China, while also contributing insights to the global discussion on balancing technological innovation with high-level security. [Method/Process] This study adopts a theoretical analytical approach based on the risk society framework, which was developed by Ulrich Beck, Anthony Giddens, and Niklas Luhmann, and further refined by cultural theorists such as Mary Douglas and Scott Lash. The research is divided into three analytical steps. First, by examining the complexity, reflexivity, and systemic characteristics of artificial intelligence risks, the study deconstructs their essence. Second, from three parallel perspectives, it identifies the generative logic of AI risks and governance dilemmas: institutionalism (focusing on the distribution of risks and 'organized irresponsibility'), culturalism (focusing on risk perceptions shaped by values and cultural backgrounds), and systemism (focusing on communication failures among functionally differentiated social subsystems). Third, based on this theoretical diagnosis, a three-dimensional governance path of "institutional regulation - value cognition - collaborative governance" is constructed, and its feasibility is tested by referring to existing governance practices in China, the EU, and the United States. Empirical evidence comes from documented cases in the literature (such as the Dutch child welfare scandal and the US COMPAS recidivism risk assessment tool) to illustrate how technological and institutional risks reinforce each other. The analysis also incorporates official policy documents, such as China's "Interim Measures for the Administration of Generative AI Services" and the EU's "Artificial Intelligence Act," to situate the theoretical discussion within a real-world regulatory context. [Results/Conclusions] Three main conclusions can be drawn from the analysis. First, artificial intelligence (AI) risks are dual in nature: they are not only technical but also deeply institutional and reflexive. Technical flaws (such as algorithmic bias and lack of transparency) can be amplified through institutional embedding (such as administrative reliance on automated systems), leading to systemic social harm. Conversely, exposure to such risks may trigger reflexive modernization, prompting institutional self-correction and innovation. Second, a single governance paradigm - whether purely risk-based, principle-based, or technologically neutral - is insufficient to address the complexity of AI risks. The three-dimensional framework proposed in this paper provides a more adaptive path: 1) at the institutional level, we establish a dynamic classification and grading system in coordination with regulatory sandboxes to balance innovation and safety; 2) at the value level, we adhere to a human-centered hierarchy of values, guided by a holistic national security perspective, dynamically adjusting the weights of security, order, freedom, and justice; 3) at the system level, we build a governance ecosystem with coordinated interaction among governments, industry, society, and international partners, supported by proactive risk assessment and real-time monitoring mechanisms. Third, this study points out several limitations and directions for future research. The proposed framework still requires theoretical improvement and needs empirical validation through pilot implementation and comparative case studies across different cultural and legal contexts. Moreover, the rapid development of AI technology, especially the emergence of general AI and autonomous systems, presents new challenges that may exceed the adaptive capacity of existing institutions. Future research should focus on developing dynamic regulatory indicators, exploring the legal implications of AI agency, and promoting international regulatory coordination.

Key words: artificial intelligence governance, risk society theory, three-dimensional regulation, reflexive modernization, risk distribution