Tencent's AI chief says AI will master biology, and bioweapon risk is real

What Yao said in Hong Kong

As technews.tw reported, Shunyu Yao told a science seminar in Hong Kong that AI models will soon be strong enough to "master biology." He sees that as both a research breakthrough and a serious safety worry.

Yao is Tencent's chief AI scientist and used to be a researcher at OpenAI, the maker of ChatGPT. That background means this is a view from inside the industry.

Two sides of the same ability

Yao said frontier AI (the most advanced models) is increasingly being used in medicine and biology research. That could speed up the development of new drugs and treatments.

But bad actors could use the same ability to make biological weapons, which could also create public health risks. He added that AI's hacking ability and its biological ability could both soon reach a level where they cause real harm.

Open weights and the safety question

"Open-weight" means a company publishes the trained model's internal settings so anyone can download and run it. Yao said that for Chinese companies and any others taking this approach, safety has to be taken seriously. Once a model is out, assessing risk and adding protections becomes especially important.

He said the industry is still divided over whether top models should stay closed to prevent harm. He also argued that open models are useful for defense. His guess is that open and closed models will continue to exist side by side.

Tencent's own plans

Yao said Tencent is exploring AI-run scientific research. It is planning an "autonomous science lab" where AI could carry out complex experiments from start to finish, such as work on reverse protein folding.

He believes the technology itself can eventually be solved. The bigger challenge, in his view, is governance in society and inside organizations. Tencent also plans to release the weights and a full system card for its Hy4 model soon. The card is meant to explain its cybersecurity and biosecurity evaluations.