AI Governance: Data and Ethics

Data is likely to become the bottleneck in AI development. The current wave of AI development has been driven primarily by advances in data, compute power, and algorithms, with data playing a critical role in all three areas. Research shows that high-quality data for training foundation models is close to exhaustion, and synthetic data might not be able to address this challenge. We believe that limited circulation is the main cause of the shortage of high-quality data. While data processing is indeed important, the costs of data processing (e.g., data cleaning and labeling costs) are not the main constraint. Rather, this data shortage challenge stems from insufficient circulation and underutilization of existing datasets. Transaction costs related to data protection and intellectual property (IP) protection represent the main barriers to data circulation. While China’s data governance model has traditionally emphasized confirmation of data rights, greater attention has been paid recently to improving the efficiency of data circulation. For example, the “three rights separation” framework has been introduced as a streamlined form of data rights confirmation to encourage circulation, though its effectiveness remains uncertain. This uncertainty is manifested in the current “fragmentation” of data exchanges and the lack of enthusiasm among enterprises to recognize data as an asset on their balance sheets. The key to facilitating data circulation is reducing data transaction costs. Data rights confirmation cannot effectively reduce transaction costs, and the intrinsic characteristics of data make ownership difficult to define. More effective approaches involve appropriate definition of the boundaries of data openness, of the scope of privacy protection, and of level of IP protection. Data rights confirmation can hardly resolve these issues. The value of data lies in its scale: An individual piece of data holds little value. Thus, data rights confirmation may in fact increase transaction costs and undermine economies of scale. Fairness and justice are especially important in AI ethics. Artificial intelligence may further amplify biases and discrimination, widen the digital divide, and potentially impact the labor market. The personified nature and interactivity of AI bring its consciousness and rights into the new frontiers of current ethical debate, making it another key challenge to understand and address the potential threats to human dignity posed by AI. China’s discussions on AI ethics are still progressing; there is a need to deepen research and advance governance concepts. AI ethics principles form the groundwork for global AI governance, and at the international level there are already some universally applicable principles in place.