Risks and Regulation Approaches of Generative AI Foundation Models: A Case Study of GPT-4

The development of generative artificial intelligence (AI) has posed profound and subversive challenges to human society. The technological update of GPT-4 causes problems and risks in underlying algorithms, training data, and intellectual property. In terms of the underlying algorithm, although there is a risk of algorithm discrimination in GPT-4, it is not necessary to make the algorithm publicly available. Instead, we should draw on brain-inspired research to promote GPT-4 to artificial general intelligence (AGI). As to training data, the massive data behind GPT-4 presents significant compliance risks; therefore, we should set up a data destruction system, maintain ideological security, and explore development options with Chinese characteristics. Regarding intellectual property, GPT-4 raises a series of infringement risks, triggering disputes over the attribution of authorship for the generated outputs. However, it has not yet constituted a challenge to the qualification of human subjects. To better deal with the risks in the process of technological development, we should formulate an appropriate regulatory scheme in time. Under the theory of meta-regulation, we should seek independent innovation in the digital age to facilitate the development of digital economy, learning from the institutional design of the Digital Services Act of the European Union and integrating the existing algorithm governance practices.