TAN Xinyu
Large generative AI models (LGAIMs), exemplified by ChatGPT, have not only brought tremendous changes to human social life but also become deeply integrated into the modern social governance system, continuously driving the evolution of government structures and the innovation of governance models. While the original intention and essence of embedding LGAIMs into social governance is to leverage artificial intelligence (AI) technology to safeguard public interests and pursue public values, these technologies inevitably cause new risks and challenges in practice. Existing literature has explored the risk patterns and governance pathways associated with the embedding of LGAIMs in social governance, focusing on the conflict between political values and instrumental values, as well as the balance between public authority and algorithmic power. However, the interactive relationship between LGAIMs, government organizational structures, and social governance orders has not been explored in depth. Furthermore, there has not been a precise analysis of the risk patterns and regulatory pathways, considering the typical empowerment scenarios of LGAIMs.
Drawing on the inter-construction theory of technology and organization, as well as technology and society, this study delves in the empowerment scenarios, risk patterns, and regulatory pathways of LGAIMs embedded into social governance. First, LGAIMs contribute to the construction of algorithmic decision-making systems in social governance, facilitating the shift from bureaucratic decision-making driven by human emotions to algorithmic rationality. However, this transformation faces challenges related to moral and responsibility judgments, necessitating a technology-for-good approach to guide the development of algorithmic decision-making models and intelligent systems in governance. Second, LGAIMs break the “information isolated island” phenomenon that often arises in traditional social governance, enabling effective cross-departmental and cross-level human-computer collaboration. Nonetheless, there is a risk of technological alienation, which makes it essential to ensure the primacy of human agency in social governance activities. Third, LGAIMs enable the scientific calculation of public service demands and accurate identification of social governance contradictions, advancing the precise matching of “supply and demand” in social governance. Nevertheless, this can also lead to the risk of social stereotyping and bias, making it essential to adhere to public value principles of fairness, justice, democracy, and equality. Fourth, LGAIMs support the construction of smart social governance scenarios characterized by virtual-physical symbiosis and human-computer interaction; however, there is a risk of undermining citizens' rights, necessitating the establishment of a multi-agent collaborative governance model focused on safeguarding civil rights.
This study expands the theoretical framework in two aspects. First, this study, based on the inter-construction theory of technology and organization, as well as technology and society, profoundly reveals the operational patterns of embedding LGAIMs into social governance. In particular, this study proposes a technology-for-good approach and a collaborative governance model to balance the conflict between political values and instrumental values during the inter-construction process between technology and organization. Similarly, this study advocates for humanistic principles and public value objectives to balance the tension between public authority and algorithmic power in the inter-construction of technology and society. Second, this study highlights key social governance scenarios, examining the empowerment effects, risk patterns, and regulatory pathways arising from the integration of LGAIMs. It further advances the contextualized study of embedding LGAIMs with social governance. This study provides policy recommendations for government agencies seeking to deploy LGAIMs in decision-making, strengthen interdepartmental collaboration, analyze and address governance contradictions, and optimize service provision. These recommendations include the formulation of target value orientations, the establishment of a risk regulatory framework, the development of an ethical system, and the promotion of a collaborative governance model.