China’s Approach to AI: Tracking Temporal Shifts in Policy Documents with NLP

• In Event: Digital Authoritarianism

Abstract

In recent years, many governments have begun making significant investments into the development of artificial intelligence (AI) and machine learning systems. The narrative frame for doing so has also evolved, particularly for authoritarian governments pursue both economic growth and social control. When it comes to new technologies, the Chinese government often tries to optimize for economic growth while preserving the ability to use those technologies for surveillance, censorship, and government capacity. But how the People's Republic of China (PRC) frames technology develop has changed over time. What explains changing patterns in government interest in AI technologies, and how has the PRC’s frame for AI-related investments evolved over time? Can we use machine learning itself to understand a large corpus of state documents and reveal what “the state” thinks about technology innovation? We examine the PRC’s approach to AI with the first large-scale analysis of China’s AI strategy, using tools from natural language processing. We use topic modelling, part-of-speech tagging and diachronic word embeddings on a dataset of 10,700 Chinese legal and regulatory documents to (a) evaluate how the legislative and policy context for AI development has evolved over the last four decades, and (b) analyze large-scale semantic shifts around how the PRC uses conceptualizes and values AI and machine learning tools. We find clear periods in the PRC’s AI. Prior to 2015 very little mention is given to research and industrial development in this field of technology innovation. From 2015 onwards a rapid escalation occurs in mentions of AI in legislative documents, as China experiences a period of “catch-up growth” in which the government seeks to encourage, regulate, and capture the AI industry. Our analysis suggests that as public agencies caught up with technical innovations, the PRC’s legislative priorities shifted from promoting economic growth to using AI for social control and establishing better regulation over the industry. This paper offers a number of methodological innovations, demonstrating how natural language processing tools can provide detailed over-time analysis of large-scale policy shifts. We also advance research on the use of information systems, technology innovation, and technology policy in authoritarian settings.

Authors

• Hannah Bailey, University of Oxford

• Hannah Rose Kirk, University of Oxford

• Philip N. Howard, Oxford University