The art of not making an AI safety deal with China

Despite the popular image of artificial intelligence leading to apocalyptic outcomes, Silicon Valley companies continue to race to develop and sell it. They often cite China as the driving force behind their lack of alternatives, even though post-ChatGPT developments indicate the main competition arises among themselves. However, the prevailing belief is that cooperation with China on advanced AI safety is unattainable.

This perception is further reinforced by the common view of China itself as a significant AI safety risk. Consequently, this positions the world in a challenging position. Yet, there exists a limited opportunity for improvement.

After a two-year break in official efforts to address AI safety, the United States and China have agreed to resume talks. According to Reuters, these discussions may take place around the expected summit involving U.S. President Donald Trump and Chinese President Xi Jinping. The success criteria should not involve trusting Beijing or engaging in a mutually beneficial agreement.

Instead, the objective should be to establish sufficient transparency between the two sides regarding their concerns about AI capabilities and the measures they are taking to mitigate these risks. This transparency aims to prevent suspicion from escalating into an excuse for inaction.