US Must Compete and Cooperate with China on AI, Former Commerce Secretary Argues

The United States must simultaneously compete with and cooperate with China on artificial intelligence (AI), according to Gina Raimondo, former Secretary of Commerce. In an opinion piece, Raimondo argued that a dual approach is necessary to navigate the complex landscape of AI development and its global implications, particularly concerning national security.

Raimondo highlighted recent high-level meetings between U.S. and Chinese leaders, noting the increasing stakes surrounding AI. She pointed to widespread security incidents reported by major AI companies globally and the ongoing debate among leading AI labs about pacing the development of frontier models. The U.S. government is actively exploring its role in these decisions.

Having previously focused on U.S. economic, technological, and security competition with China, Raimondo emphasized actions taken, such as implementing export controls on chip technology and establishing the AI Safety Institute. She stated that while U.S. companies are right to consider pacing AI development, it is crucial for the U.S. to maintain its lead over China, as Chinese dominance in AI would threaten American interests.

Pursuing Cooperation on Safety Standards

Raimondo suggested that China also has reasons to prioritize AI safety, citing remarks from its Minister of State Security about the potential risks advanced AI poses to the stability of the Chinese Communist Party. This shared concern creates an opportunity for alignment on certain safety standards and a mutual understanding of AI risks. The possibility of an AI dialogue between U.S. Treasury Secretary Scott Bessent and Chinese Vice Premier He Lifeng indicates a potential space for cooperation.

Raimondo advocated for convening experts from both countries to establish and maintain this dialogue. She acknowledged the need for caution and the development of verification mechanisms, drawing parallels to the U.S. approach to nuclear arms control dialogues with the Soviet Union during the Cold War, which involved zero-trust methods to verify agreements.

Concurrently, Raimondo stressed the importance of continuously updating export controls on advanced chips and addressing China's access to U.S. compute power, including remote access and technology diversion. Collaboration with international partners like the Netherlands, Japan, and Germany is essential to tighten controls on semiconductor manufacturing equipment.

Addressing AI Distillation and Theft

Raimondo also called for AI companies to take adversarial distillation more seriously. Distillation, a process where AI labs train weaker models using the outputs of more advanced models, has become a significant concern. She cited instances where Chinese AI labs, including Moonshot AI, have reportedly engaged in industrial-scale distillation campaigns, training their models on U.S. AI outputs. This practice raises concerns about the theft of American AI technology, despite China's efforts to indigenize its AI capabilities.

Raimondo urged American AI labs to allocate more resources towards identifying and exposing such activities to inform policymakers and the public. She concluded that managing the U.S.-China relationship effectively, through both technological competition and collaboration on risks, is essential for the safe deployment of AI and for ensuring the U.S. remains resilient to the transformative changes AI will bring.