

AI Futures | Forecasting Transformative AI
A two-year and a hundred-year AGI timeline imply completely different strategies, so it’s worth having a view, even if it’s a broad distribution.
The case for the short end has recently become concrete: several frontier AI companies have announced they’re working towards recursive self-improvement — AI systems that accelerate the research producing the next generation of AI systems. If that loop closes, it could turn a decades-long timeline into a short one. So the question is who believes it will, and why. We will look at what different forecasts currently say and how they treat this mechanism: superforecasters and prediction markets, expert and public surveys over time, frontier-lab insiders, and the inside-view arguments built on scaling and task-length trends. We will also look at which signals — compute, data center investment, agents, AI R&D speedups — tell us whether a self-improvement loop is actually closing.
No technical or forecasting background needed.
📆 Thursday, Oct 15 – 18:30–20:00
📍 ETH Main Building, Rämistrasse 101, Room HG F 26.
Part 2 of the AI Futures series, presented by Zurich AI Safety.