New Study Explores International Institutions for Advanced AI Governance

The global impact of advanced artificial intelligence (AI) has generated discussions on the need for international governance structures to handle the opportunities and risks involved. Often drawing on analogies with other international institutions, such as the International Civil Aviation Organisation (ICAO), CERN (European Organisation for Nuclear Research), and the International Atomic Energy Agency (IAEA), these discussions serve as a starting point. However, it is crucial to recognize that the emerging AI technologies differ greatly from these analogies in aviation, particle physics, and nuclear technology.

In order to establish effective AI governance, it is essential to delve deeper into the specific benefits and risks that need international management, identify the governance functions required, and determine the organizations best suited to provide these functions. Our collaborative research paper, written in partnership with the University of Oxford, Université de Montréal, University of Toronto, Columbia University, Harvard University, Stanford University, and OpenAI, addresses these important questions and investigates the potential role of international institutions in managing the global impact of frontier AI development.

The Significance of International and Multilateral Institutions

Access to AI technology has the potential to greatly enhance prosperity and stability. However, the benefits of these technologies may not be evenly distributed, especially in underrepresented communities and the developing world. Limited access to internet services, computing power, machine learning training, and expertise can prevent certain groups from fully benefiting from AI advancements.

International collaborations can help address these challenges by encouraging the development of systems and applications that cater to the needs of underserved communities. They can also help overcome educational, infrastructural, and economic obstacles that hinder the adoption of AI technology in these communities.

Furthermore, international efforts are necessary to manage the risks posed by powerful AI capabilities. Without proper safeguards, these capabilities can be misused to cause harm. International and multi-stakeholder institutions can facilitate global consensus on the threats posed by different AI capabilities and establish international standards for identifying and addressing models with dangerous capabilities. Collaborations on safety research can also enhance the reliability and resilience of AI systems.

Lastly, international institutions can play a pivotal role in supporting best practices and compliance monitoring when economic competition among states might undermine regulatory commitments.

Potential Institutional Models for AI Governance

We propose four institutional models that can effectively support global coordination and governance functions:

• An intergovernmental Commission on Frontier AI could foster international consensus on the opportunities and risks associated with advanced AI, generate public awareness, contribute to a scientifically informed account of AI use and risk mitigation, and serve as a source of expertise for policymakers.
• An intergovernmental or multi-stakeholder Advanced AI Governance Organisation could align global efforts to address the risks posed by advanced AI systems by establishing governance norms, standards, and compliance monitoring.
• A Frontier AI Collaborative as an international public-private partnership could promote access to advanced AI technology, ensuring underserved societies benefit from cutting-edge AI advancements and facilitating international access to AI technology for safety and governance objectives.
• An AI Safety Project could bring together leading researchers and engineers, providing them with resources and advanced AI models to research technical mitigations of AI risks. This would promote AI safety research and development on a larger scale.

Operational Challenges

While these institutional models show promise, numerous challenges need to be addressed:

A Commission on Advanced AI will encounter scientific challenges due to the uncertainty surrounding AI trajectories and capabilities, as well as the limited research on advanced AI issues.

An Advanced AI Governance Organisation may struggle to keep up with the rapidly progressing AI landscape, given the limited capacity in the public sector. The effectiveness of its standards and monitoring also depends on international coordination and adoption.

A Frontier AI Collaborative may face obstacles in fully harnessing the benefits of advanced AI systems. Balancing the sharing of AI benefits with preventing the proliferation of dangerous systems presents a difficult tension to manage.

For the AI Safety Project, it is crucial to determine the best approach for safety research collaborations versus individual company efforts. Additionally, securing access to the most capable AI models from relevant developers for safety research poses a significant challenge.

Given the immense global opportunities and challenges presented by AI systems, further discussions among governments and stakeholders are necessary to determine the role of international institutions in AI governance and coordination. It is our hope that this research sparks valuable conversations and contributes to the development of advanced AI for the benefit of humanity.