China, AI, and the US Race

The AI Competition Between the U.S. and China

Many agree that the United States must be victorious in the AI competition with China. The Biden administration warned of losing its lead if AI deployment isn't accelerated for national security. The Trump administration also aimed to maintain American AI dominance.

Washington's strategy involves restricting China's access to key technology while boosting domestic AI innovation. This includes light regulation, investments in semiconductors and energy, and AI adoption across federal agencies. These policies have helped U.S. firms lead in market share and model performance.

However, this lead may not last. Chinese AI companies are making breakthroughs, narrowing the gap between U.S. and Chinese AI capabilities. Washington needs to prepare for a future where it might lose the AI competition or where Chinese AI models are globally popular.

Planning for Second Place

Preparing for potentially finishing second doesn't mean repeating the 5G competition failures. Instead, the U.S. can promote frameworks that highlight AI's appeal to emerging markets. It can also ease model migration, create systems for comparing AI models, and securely share data.

Until mid-2024, the United States seemed to have a winning formula for AI dominance, driven by research, private sector investment, and minimal regulation. U.S. models showed advancements, reducing errors and enhancing capabilities. American tech companies developed larger models, increasing their global market share.

China has been closing the gap through government initiatives, AI education, research investment, industry coordination, and infrastructure investment. By late 2024, the performance gap narrowed. Some Chinese models have matched U.S. performance, and China leads in integrating AI into manufacturing.

U.S. export controls have been less effective than expected. China has found ways to acquire advanced chips and develop its own. Chinese firms have also optimized software to improve hardware performance and lower costs. The days of absolute American AI dominance are ending.

Alternative Strategies for the U.S.

American AI firms may still lead in some areas. Major companies continue to invest in AI infrastructure, startups, and talent. But the pace of U.S. innovation could slow due to challenges in AI discovery, talent competition, and research funding cuts. Policymakers should plan for a world with competing AI ecosystems.

The U.S. should find ways to showcase its models' merits. New evaluation frameworks could focus on transparency, accountability, accessibility, cost, and adaptability. Minimizing migration costs between models will also be important.

Standardizing application programming interfaces could lower the costs of switching between models. Giving users confidence in accessing multiple models is a good strategy if Chinese models take the lead.

As Chinese models gain power, the U.S. must do more than highlight risks. U.S. companies should build systems that mitigate the risks of relying on any single model. An intermediate abstraction layer can isolate systems from the foundational model, increasing resilience.

Adopting Chinese models carries risks, including incorrect outputs, data exposure, and service disruptions. U.S. companies should ensure their applications are resilient against these threats. Adjudication systems can compare responses from different models, warn users, and prevent the use of incorrect responses.

The United States should regulate data sharing with foreign model builders without blanket bans. In some cases, the benefits of fine-tuning a Chinese model with U.S. data may outweigh the risks. Data anonymization and masking can reduce privacy and security risks.

Washington needs to standardize evaluation metrics and develop guidelines for data sharing. It should also assist partners in migrating between models and building adjudication systems. While striving to maintain its lead, the U.S. needs realistic policies. Preparing for the possibility of not achieving outright dominance is essential.

Failing to adapt and compete would be worse than finishing second.