In late 2008, economist Robin Hanson and AI-safety writer Eliezer Yudkowsky held an extended written debate on the blog Overcoming Bias about how advanced AI would arrive. The exchange became known as the AI-Foom debate, after “foom,” Yudkowsky’s onomatopoeia for an extremely fast intelligence takeoff. It is one of the most cited early arguments about takeoff dynamics and remains a touchstone in discussions of AI risk.
Yudkowsky argued for a hard takeoff: that an AI capable of improving its own design could enter a loop of recursive self-improvement, each gain making the next gain easier, and shoot from roughly human-level to vastly superhuman quickly - perhaps too quickly for anyone to react. On this view a single project could pull decisively ahead, which is why he stressed the need to get an AI’s goals right before that point. Hanson was skeptical of so sudden and so local a jump. Drawing on economic history, he expected progress to be more gradual and broadly distributed across many actors and systems, more like past technological and economic transitions than a lone system erupting. His own preferred scenario, later developed in “The Age of Em,” ran through brain emulation rather than a self-rewriting AI.
The debate ranged over recursive self-improvement, whether a theory of “friendliness” was needed before building powerful AI, and the relative odds of engineered AI versus whole brain emulation arriving first. It never produced a winner - the two largely talked past each other on how much a single system could outpace the rest of the world. MIRI later compiled the posts and follow-ups, including a 2011 live debate, into a 2013 ebook. The hard-takeoff-versus-slow-takeoff question it crystallized is still unresolved and still shapes how people argue about AI timelines today.