AI will soon master biology, unleashing new safety risks, Tencent AI chief warns

Artificial intelligence expert Vinces Yao from Tencent AI warned that AI systems will soon reach a level where they can master biology, bringing both opportunities and significant security risks. During a science symposium in Hong Kong, Yao noted that as frontier AI models are increasingly applied to medicine and biology, they could accelerate drug discovery but also pose dangers such as the creation of bioweapons or triggering future pandemics.

He emphasized the need for responsibility and seriousness in addressing these security and safety concerns. Tencent, which recently unveiled its top-performing Chinese AI model Hy4, plans to use AI for autonomous scientific research, including running experiments end-to-end like protein folding. However, Yao cautioned that while technical challenges can be overcome, the bigger issue lies in societal and organizational aspects.

He stressed the importance of careful evaluation and collaboration with security labs when releasing AI models, emphasizing that both Chinese and US companies are recognizing the potential harms of frontier AI. Despite differing views on whether open-source models are better for security, Yao suggested a mix of closed and open-source models could be the optimal approach.

China and the US are converging on shared concerns about AI risks, with Chinese tech giants contributing to a report on AI safety at the World Internet Conference. The country is also tightening controls on frontier AI systems, and AI is expected to be a key topic of discussion at upcoming high-level meetings between Chinese and US leaders.