In Which Areas of Technical AI Safety Could Geopolitical Rivals Cooperate?

Abstract: International cooperation is common in AI research, including between geopolitical rivals. While many experts advocate for greater international cooperation on AI safety to address shared global risks, some view cooperation on AI with suspicion, arguing that it can pose unacceptable risks to national security. However, the extent to which cooperation on AI safety poses such risks, as well as provides benefits, depends on the specific area of cooperation. In this paper, we consider technical factors that impact the risks of international cooperation on AI safety research, focusing on the degree to which such cooperation can advance dangerous capabilities, result in the sharing of sensitive information, or provide opportunities for harm. We begin by why nations historically cooperate on strategic technologies and analyse current US-China cooperation in AI as a case study. We further argue that existing frameworks for managing associated risks can be supplemented with consideration of key risks specific to cooperation on technical AI safety research. Through our analysis, we find that research into AI verification mechanisms and shared protocols may be suitable areas for such cooperation. Through this analysis we aim to help researchers and governments identify and mitigate the risks of international cooperation on AI safety research, so that the benefits of cooperation can be fully realised.

Technical AI Safety Cooperation Among Geopolitical Rivals

The paper "In Which Areas of Technical AI Safety Could Geopolitical Rivals Cooperate?" addresses the nuanced dynamics of international cooperation on AI safety among geopolitical rivals, primarily focusing on the United States and China. The authors investigate whether cooperation in specific AI safety domains might present low risks while offering potential mutual benefits. This discussion is particularly relevant in the context of managing shared global risks from AI technologies, where cooperation between geopolitical adversaries could be both beneficial and fraught with challenges.

Strategic Context and Historical Precedents

The paper begins by framing the persistence of international cooperation despite historical and current geopolitical rivalries. Using examples from the Cold War and contemporary AI collaborations, it highlights that cooperative efforts, even among rivals like the US and China, have not been unusual. The exploration underscores the potential for AI safety cooperation akin to the nuclear safety collaborations of the past, where mutual benefits outweighed the risks of competitive escalation.

Analytical Framework and Case Studies

Central to the paper is an analytical framework that considers the risks associated with advancing AI capabilities, sharing sensitive information, and enabling opportunities for malfeasance in the context of AI safety cooperation. A case study of US-China interactions demonstrates current and past collaborative efforts in academia, industry, and limited government-level exchanges, illuminating the potential spaces for further cooperation.

In particular, the analysis identifies four areas of technical AI safety research that could be suitable for cooperation, with varying risk profiles:

1. Verification Mechanisms: This includes techniques for confirming claims about AI systems, which could be less susceptible to risks of advancing a rival's capabilities or leaking sensitive information. The emphasis is on attesting to known properties rather than discovering new ones.
2. Protocols and Best Practices: The development of shared protocols can help standardize AI operations without advancing capabilities or requiring sensitive disclosures. This domain can largely sidestep risks that cooperation would inadvertently empower rivals or expose vulnerabilities.
3. Infrastructure for AI Safety: Shared infrastructure can aid interoperability and resource pooling but may conflate multi-use risks. While potentially beneficial, developing such infrastructure requires careful management of access to ensure it is not repurposed for harmful activities.
4. Evaluation Methodologies: The focus on evaluating AI system capabilities involves inherent risks, particularly regarding capability elicitation and strategic advances. However, evaluations on non-sensitive subjects might still be feasible areas for collaboration without significant risk exposure.

Practical and Theoretical Implications

The findings of this paper have implications for both policymakers and AI safety researchers. For practitioners, understanding the specific technical and geopolitical risks associated with cooperation can inform safer collaboration strategies. For theorists, the paper adds to the discourse on international relations in the context of emerging technologies, providing an empirical basis for exploring how shared governance can mitigate AI-related risks.

Future Directions

This essay highlights that while certain areas of AI safety, such as verification and protocols, appear to be fertile grounds for cooperation, practical implementation remains challenging. Future research could further delineate the boundaries of these cooperative areas, propose specific policy mechanisms, and explore additional domains of technical AI safety that could benefit from international cooperation.

In conclusion, the authors suggest that careful consideration of specific technical risks and robust mechanisms to mitigate them will be essential to unlocking the potential benefits of AI safety cooperation among geopolitical rivals. Such efforts could lead to a safer global AI landscape, leveraging the strengths of collaboration without compromising national security.