The concept originates with mathematician I.J. Good's 1965 essay "Speculations Concerning the First Ultraintelligent Machine," which argued that a machine able to improve its own design would trigger successive, ever-faster rounds of self-improvement until it far exceeded human intellect — the "last invention" humanity would need to make. Nick Bostrom's 2014 book Superintelligence later popularized the idea and distinguished a fast "hard takeoff" from a slower "soft takeoff." The term is often conflated with the technological singularity, but names a narrower, specific mechanism (recursive self-improvement) rather than the broader claim about post-explosion unpredictability. Researchers disagree on whether current AI progress shows early signs of this feedback loop — such as AI systems accelerating AI research — or whether compute, data, and algorithmic bottlenecks would slow any such loop well short of a true explosion.