FOOM

FOOM is informal shorthand for a fast, local AI takeoff. In this scenario an AI able to improve its own design enters a runaway feedback loop and rises, in weeks or hours rather than decades, from roughly human level to far beyond any rival. Its engine is recursive self-improvement – each gain in capability makes the next gain easier to find.

Where the idea comes from

In 1965 I. J. Good, then at Trinity College, Oxford, published “Speculations Concerning the First Ultraintelligent Machine” in Advances in Computers, volume 6, based on talks from 1962 and 1963. He defined an ultraintelligent machine as one that could far surpass all the intellectual activities of any human. Because designing machines is one of those activities, such a machine could design better ones, and there would be an “intelligence explosion”. Good attached a condition that is often dropped: the first such machine would be the last invention humanity needs only if it were docile enough to tell us how to keep it under control.

The word and the debate

The word is associated above all with the AI theorist Eliezer Yudkowsky and with the debate he held in late 2008 with the economist Robin Hanson on Overcoming Bias, the blog they then shared. Hanson titled his summary of Yudkowsky’s position “AI Go Foom” (November 10, 2008). On December 1, 2008, in “Recursive Self-Improvement”, Yudkowsky stated the claim: a fast, local increase in capability, fast meaning weeks or hours, producing a mind vastly smarter than anything else around. In “Hard Takeoff” the next day he argued that a self-improving AI would most likely either fizzle out or go FOOM. The decisive break, he thought, would come once an AI could do AI research itself.

Hanson agreed that machine intelligence would bring almost unprecedented impact and risk. He disputed the speed and, above all, the locality. In “Setting the Stage” (November 18, 2008) he put the chance that a single hand-coded AI would suddenly turn from nearly powerless to overwhelmingly powerful at under 1 percent, and Yudkowsky’s estimate at over 10 percent. Drawing on the economics of growth, Hanson expected progress to come from many small, shared improvements across a whole economy, and expected emulated human brains to arrive before hand-coded AI.

The two debated in person at Jane Street Capital on June 29, 2011. In 2013 the Machine Intelligence Research Institute (MIRI) published the posts, the 2011 transcript, a summary by Kaj Sotala and Yudkowsky’s report “Intelligence Explosion Microeconomics” as the free ebook The Hanson-Yudkowsky AI-Foom Debate.

Fast and slow takeoff

Later discussion treated FOOM as one end of a spectrum of takeoff speeds. In “Takeoff speeds” (February 24, 2018) Paul Christiano proposed an operational test for a slow takeoff: world economic output completes a doubling over four years before it first doubles within a single year. His central argument was that AI which is mediocre at self-improvement will come before AI that is excellent at it, so acceleration should be visible in advance. Slow does not mean gentle: powerful AI transforms the economy earlier, which he summed up as “slower takeoff means faster progress”. He put the probability of a fast takeoff at about 30 percent; many in AI safety, he thought, put it at 70 to 90.

Why it matters now

The International AI Safety Report 2026, published on February 3, 2026, records that AI tools are now routinely used in AI development to write training code, design hardware and generate synthetic training data. It does not use the word FOOM and reports that experts disagree on whether this automation will dramatically accelerate progress. It cites a 2025 pilot study by METR with the Forecasting Research Institute. Asked whether any two-year period before 2029 would bring as much growth in effective compute as the six years from 2018 to 2024, eight AI forecasting experts gave a median of 20 percent and ten superforecasters 8 percent. If AI systems were shown to outperform human researchers on month-long projects, the superforecasters’ median rose to 18 percent. The report warns that if each advance that speeds up AI research also eases the next, decades of progress could happen in years.

Related

Singularity · Sharp Left Turn · Scaling Laws · p(doom) · Recursion Without Explosion (essay) · Nothing Went Foom! (article on the song) · A Field Guide to Nothing Went Foom! · What is AIciety? · Accelerationism

Sources: I. J. Good, “Speculations Concerning the First Ultraintelligent Machine” (1965), Virginia Tech repository · MIRI: The Hanson-Yudkowsky AI-Foom Debate (2013) · LessWrong: The Hanson-Yudkowsky AI-Foom Debate · LessWrong: AI Takeoff · Eliezer Yudkowsky: Hard Takeoff · Paul Christiano: Takeoff speeds · International AI Safety Report 2026 · METR: Forecasting the Impacts of AI R&D Acceleration