The United States and China are engaged in a conflict over AI development, with the Department of Commerce implementing export controls to limit China’s access to technology. These controls could extend beyond computer chips to include general-purpose AI programs. Of particular concern are frontier models, which are advanced AI systems with flexible and wide-ranging uses that could also have dangerous capabilities. A white paper co-authored by tech companies suggests the need for regulation to prevent the potential harms of frontier AI. However, such controls could have unintended consequences, and the focus on frontier models may distract from addressing present-day AI-related issues.
| Signal | Change | 10y horizon | Driving force |
|---|---|---|---|
| Conflict over AI development between US and China | From cooperation to tension and competition | Increased regulation, restricted access to AI models | National security concerns and economic warfare |
| Export controls on AI technology | Limiting China’s access to AI development | More friction with China, weaker AI innovation in the US | National security and economic competition |
| Concerns over frontier models of AI | Regulation and licensing of advanced AI models | Restricted deployment and development of frontier AI | Potential for unforeseen abuses and threats to public safety |
| Collaboration between tech companies on frontier models | Industry group focused on safe and responsible development | Research and recommendations on responsible frontier-model development | Ensuring safety and accountability in AI development |
| US-China collaboration in AI development | Collaborative advancements in AI research and applications | Impact on global technology advancement and AI leadership | Strong collaboration and mutual advancement |
| Technical feasibility of export controls on AI models | Uncertainty and potential challenges in enforcing controls | Circumvention of controls, unintended consequences | Hypothetical threats and technical limitations |
| Distraction from addressing present-day harms of AI models | Shifting regulatory attention away from existing model harms | Neglected focus on privacy, copyright, and job automation | Fear-mongering and anti-China framing |
| Concentration of power and erosion of policy ideas | Decline in diversity of AI safety discussions | Neglect of worker impacts and environmental concerns | Concentration of resources and influence in a few companies |