Pathways to Short Transformative AI Timelines

The previous chapters of this report laid out and examined the debates over two possible mechanisms of fast AI capabilities progress – compute scaling and recursive improvement – and their potential to produce TAI within the next ten years. Through this, several compelling arguments for short TAI timelines have already emerged, and have been shown to stand up reasonably well against some of the key sceptical arguments. I now go on to synthesise the core argumentative threads of these chapters to generate a set of scenarios which exhibit short TAI timelines. Each of these scenarios is underpinned by different assumptions about capabilities progress – and each seems, in light of previous reflections, to represent a plausible future for the development of AI. Through mapping out this space of scenarios, a more robust case for believing in short TAI timelines is brought into focus. This approach is inspired by Convergence Analysis’ research agenda for AI Clarity, which emphasises scenario planning as a tool for exploring and addressing AI risk under uncertainty. My specific choice of methodology here, which distils key ideas from earlier arguments under a set of scenarios, has two purposes: (a) to clarify how different assumptions about the world could support a short TAI timeline, and (b), to begin building a more concrete picture of what the next ten years of AI development might actually look like in a short timeline world.