As AI systems have become increasingly capable, policymakers, the public, and the field of AI governance have begun to consider the potential impacts and risks from these systems—and the question of how best to govern such increasingly advanced AI. Call this field ‘Advanced AI Governance’. However, debates within and between these communities often lack clarity over key concepts and terms. In response, this report provides an overview, taxonomy, and preliminary analysis of many cornerstone ideas and concepts within Advanced AI Governance.
To do so, it first reviews three different purposes for seeking definitions (technological; sociotechnical; and regulatory), and discusses why and how terminology matters to both the study and practice of AI governance. Next, the report surveys key definitions in advanced AI governance. It reviews 101 definitions across 69 terms that have been coined for advanced AI systems, within four categories: (1) essence-based concepts that focus on the anticipated form of advanced AI; (2) development-based terms that emphasize the hypothesized pathways towards advanced AI; (3) sociotechnical-change-based terms that center the societal impacts of such AI, and (4) risk-based terms that highlight specific critical capabilities of advanced AI systems. The report then reviews distinct definitions of the tools of (AI) ‘policy’ and ‘governance’; different paradigms within the field of advanced AI governance, and different concepts around theories of change. By disentangling these terms and definitions, this report aims to facilitate more productive conversations between AI researchers, academics, policymakers, and the public on the key challenges of advanced AI.