The governments of the United States and China are seeking to establish dialogues on AI security in a sensitive situation. While U.S. Treasury Secretary Scott Basset prepares to meet with Chinese Vice Premier He Lifeng, President Donald Trump has dismissed public warnings about the dangers of AI as 'imitation.' On the other hand, an article in the state-run China Daily accused leading American companies of using requests to slow down AI advancements.
The Need for Dialogue on AI
Neither the United States nor China is yet prepared to slow down their advancements, and there is no mutual trust. AI systems are rapidly advancing and cannot wait for diplomacy. However, these realities do not mean that Trump and Xi Jinping can ignore the increasing calls for dialogue on AI. Any dialogue must avoid two common pitfalls of U.S.-China diplomacy: setting goals that are so low that negotiations become meaningless or so high that agreement becomes impossible.
Read more: The House of Representatives approved a plan to reduce electricity costs for data centers
Achievable Goals for Negotiations
If the goals are too low, the dialogue will turn into a photo op and a general statement about promoting innovation and reducing risks. If the goals are too high and seek broad limitations on cutting-edge developments that neither side has the political trust or capacity to verify and enforce, the meeting will end in failure. A better short-term goal to pursue is for the U.S. and China to work in parallel on AI safety and demonstrate their advancements in practice. This effort should focus on technical exchanges that make each country's AI systems safer based on their own conditions.
The two countries should also establish a protocol for communication regarding emerging AI incidents to discuss them before they turn into international crises. These steps would represent meaningful progress and are more valuable than lofty statements or unverifiable commitments.
The Importance of Technical Exchanges
The Trump administration should seek technical exchanges on how to test models for dangerous capabilities and behaviors. These include AI systems that could assist amateurs in developing biological weapons or evade human oversight and control. American AI companies have studied these risks for years, and these risks have also begun to emerge in China's technical standards regarding safety testing. Sharing best practices and new insights on how to test and mitigate these risks could significantly enhance the safety of systems built in both countries.
Communication Protocol in Crisis
This diplomacy is less about negotiating a broad arms control treaty and more about establishing common safety practices and procedures for a technology that neither government fully understands or controls. The principle is that Washington must trust Beijing or vice versa; neither country will benefit if the other loses control of a powerful system.
For this approach to be effective, it must be accompanied by a communication protocol in times of crisis between the governments. What should be done if an American AI model goes out of control and hacks a Chinese company or government network? What should be done if a Chinese drone equipped with AI attacks a U.S. Navy ship and neither side can quickly determine whether the incident was accidental or intentional?
These are not existential risks that President Trump has dismissed, but they are close to incidents that are currently occurring in real-time. The two countries need to establish basic infrastructure to address these incidents. The success or failure of the communication channel will depend on its operational design.
Read more: Tension in the AI industry over how to regulate this technology · OpenAI identified six concerning behaviors of artificial intelligence
