Suppose that nations became so concerned about the risks from advanced artificial intelligence (AI) that they were willing to consider bold and ambitious international coordination proposals. One such proposal involves the establishment of a new international AI authority responsible for the safe and secure development of highly advanced AI systems. This paper expands on this idea by exploring design considerations for a joint international AI lab. First, it describes motivations: why nations may come together to participate in an international joint lab. The joint lab proposal is also compared to a proposal for a national “AGI Manhattan Project,” discussing some of the merits and drawbacks of each approach. Second, the paper discusses the purposes of a joint lab and its main objectives. Third, the paper outlines the governance of the joint lab. Drawing on precedents from Biosafety Level 4 (BSL-4) labs, the paper explains how the lab would operate and make critical decisions. Research and information security is discussed (techniques that the joint lab would use to prevent key parameters or insights from being stolen or leaked), as well as emergency protocols (techniques the lab would use to detect and prevent potential global security emergencies). Finally, the paper highlights the limitations of this proposal and open questions for future research.
