A Study of the Governance of Generative AI: Taking ChatGPT as an Example

生成式 AI 的治理研究:以 ChatGPT 为例

Abstract

Based on autoregressive generative models, pre-training, and human-feedback reinforcement learning, ChatGPT has achieved powerful natural language processing capabilities, overturning previous understandings of artificial intelligence. However, ChatGPT has also brought various risks in model training, generated content, and applications. Before governance begins, it should be clear that humanism is the value foundation of governance, inclusive and prudent agile governance is the governing philosophy, multi-party participation combining specific points and overall approaches is the main requirement of governance, and a systematic approach with multiple measures is the governance model. Therefore, the specific path of governance will unfold from the following five aspects: building a standard training dataset, improving the professional qualification and access system for AI trainers, strengthening algorithm supervision technology, implementing ethical governance solutions throughout the process, and optimizing the legal system and the allocation of legal responsibilities.