FOOM is an onomatopoeia, the sound of something igniting all at once, raised to a thesis: once an AI exists that can improve its own design, it will go from human level to vastly superhuman in days or weeks, not decades. It is also called a hard takeoff, as opposed to a gradual slow takeoff. The parent idea is I. J. Good's "intelligence explosion": an ultraintelligent machine would design better machines, and "the intelligence of man would be left far behind" (Good, 1966, p. 33).
The word was enshrined in the debate between Eliezer Yudkowsky and the economist Robin Hanson, fought on the blog Overcoming Bias in late 2008, resumed face to face in 2011, and collected in a book by MIRI (Hanson & Yudkowsky, 2013). Yudkowsky argued for a local, fast takeoff: a single system breaking away from the pack through recursive self-improvement. Hanson bet on distributed, economic growth, with many small improvements spread across many actors, as in any industrial revolution. The song takes sides without nuance: "'cause the future goes FOOM" (osmarks, n.d.).
Who won? Retrospectives disagree. Recent progress has come more from scaling compute and data (scaling hypothesis) than from a program rewriting itself, which looks like Hanson's world; but it is concentrated in a few labs and produces jumps that surprise their own makers (emergent abilities), which feeds Yudkowsky's. Skeptics add that intelligence is situated: it depends on data, experiments, hardware and organizations, and no bottleneck in the physical world is removed merely by thinking faster. The underlying question is at once empirical and conceptual: does intelligence applied to itself have increasing or diminishing returns?