Technological Singularity

← Encyclopedia

The technological singularity is the forecast of a break in history: the claim that machine intelligence, once it exceeds the human kind, will improve itself in an accelerating loop, and that the world on the far side of that loop cannot be predicted from this one. The word is borrowed from mathematics, where a singularity is a point past which a function’s behavior stops yielding to analysis. Applied to the future, it names the point at which forecasting itself is held to fail.

The word entered the record in an obituary. In May 1958 Stanislaw Ulam published a memorial tribute to John von Neumann (1903–1957) in the Bulletin of the American Mathematical Society, and on page 5 recalled a conversation with him on “the ever accelerating progress of technology and changes in the mode of human life, which gives the appearance of approaching some essential singularity in the history of the race beyond which human affairs, as we know them, could not continue.” Popular accounts put the sentence in von Neumann’s mouth; it is Ulam’s prose, the paraphrase of a remembered conversation, and “essential singularity” there is a mathematician’s metaphor, not a coined doctrine. The idea then lay quiet for seven years.

The argument arrived in 1965, without the word. Irving John Good — of Trinity College, Oxford, and the Atlas Computer Laboratory at Chilton — published “Speculations Concerning the First Ultraintelligent Machine” in Advances in Computers, opening with the claim that the survival of man depends on building one early. His central passage supplies the engine every later version runs on: a machine that far surpasses all the intellectual activities of any man, however clever, could — since machine design is such an activity — design still better machines, and there would follow what Good named an “intelligence explosion.” Hence his most quoted sentence: “the first ultraintelligent machine is the last invention that man need ever make, provided that the machine is docile enough to tell us how to keep it under control.” The proviso, easy to read past, contains the next sixty years of the subject. Good never used “singularity”; later writers welded his recursion to Ulam’s word. By David Chalmers’s account, Good expected the machine by 2000.

The name and the thesis fused in 1993. Vernor Vinge — professor of mathematics at San Diego State University and a Hugo-winning novelist — delivered “The Coming Technological Singularity: How to Survive in the Post-Human Era” at the VISION-21 Symposium, sponsored by NASA Lewis Research Center and the Ohio Aerospace Institute, on March 30–31, 1993; the essay appeared in NASA’s conference proceedings that December and, slightly changed, in the Whole Earth Review that winter. Reference works sometimes credit Vinge with the concept outright; the 1958 print record says otherwise, but the modern thesis is his. His abstract runs two sentences: “Within thirty years, we will have the technological means to create superhuman intelligence. Shortly after, the human era will be ended.” He listed four roads — computers that wake, networks and their users waking together, interfaces intimate enough to make their users superhuman, biological improvement of human intellect — then did what futurists mostly avoid and committed to dates: he would be surprised, he wrote, if it came before 2005 or after 2030.

Ray Kurzweil converted the window into a timetable. His March 2001 essay “The Law of Accelerating Returns” generalized Moore’s Law into a claimed law of history: “we won’t experience 100 years of progress in the 21st century — it will be more like 20,000 years of progress (at today’s rate).” The Singularity Is Near (2005) carried the term to a mass audience, announcing an era in which intelligence becomes “increasingly nonbiological and trillions of times more powerful,” with dates attached: in the Stanford Encyclopedia of Philosophy’s summary, supercomputers at human computational capacity by 2010 and mind uploading by 2030; a machine passing the Turing test by 2029, on which he placed a $20,000 bet with Mitch Kapor in 2002; the singularity itself around 2045. The 2001 essay’s own nearer checkpoint — human-brain computing power, 2 × 10¹⁶ calculations per second, for $1,000 around 2023 — is now past its date and unadjudicated. Kurzweil also published a scorecard. His 2010 essay “How My Predictions Are Faring” grades 147 predictions he had made in the 1990s for the year 2009, finding 115 entirely correct, 12 essentially correct, 17 partially correct, and 3 wrong — 86 percent right by his own arithmetic. The grading is his, made in rebuttal to critics, and disputed; the predictions, at least, were public and datable enough to grade.

The thesis is narrower than its neighborhood: it claims nothing, in itself, about human enhancement or immortality — that program belongs to transhumanism, which absorbed the intelligence explosion among its far horizons — and nothing about the nature of reality. The load-bearing claim concerns the dynamics of intelligence growth: positive-feedback recursion (Good), a horizon of predictability (Vinge), a schedule (Kurzweil). The parts are separable, and the advocates themselves split along them. The economist Robin Hanson, writing in IEEE Spectrum in June 2008, accepted the discontinuity but stripped it of mystery: machine intelligence would be the next of the growth-mode transitions — hunting to farming to industry — after which a world economy that now doubles every fifteen years or so “would soon double in somewhere from a week to a month.” Vinge’s event horizon, Kurzweil’s scheduled merger, and Hanson’s growth spike are three theses sharing a word.

By the 2000s the idea had institutions. In 2000 Eliezer Yudkowsky, with Brian and Sabine Atkins, founded the Singularity Institute for Artificial Intelligence, which organized the annual Singularity Summit; Singularity University followed in 2008, founded by Peter Diamandis and Kurzweil and explicitly inspired by the 2005 book. Academic philosophy arrived in 2010 with Chalmers’s “The Singularity: A Philosophical Analysis” in the Journal of Consciousness Studies, a formal reconstruction that deliberately stipulates “within centuries” where advocates said decades. In 2013 the Singularity Institute renamed itself the Machine Intelligence Research Institute, turned for five years to the foundational mathematics of AI alignment, and now states its purpose as research and outreach “intended to help prevent human extinction from the development of artificial superintelligence.” The community that organized around the singularity retired the word. The awaited event had become, in its own vocabulary, a risk to be prevented.

The criticisms are as named and dated as the forecasts. Steven Pinker, polled for IEEE Spectrum’s June 2008 singularity report: “There is not the slightest reason to believe in a coming singularity. The fact that you can visualize a future in your imagination is not evidence that it is likely or even possible” — pointing to domed cities and jet-pack commuting. The physicist Theodore Modis went after the curves themselves in 2006: “Nothing in nature follows a pure exponential” — natural growth follows the logistic S-curve, which mimics an exponential early and then bends; even explosions, he argued, are logistics. He added a provenance problem: the fourteen sets of evolutionary “milestones” beneath the canonical acceleration chart were recycled from his own work and not all independent. By 2012 his verdict had hardened into a chapter title, “Why the Singularity Cannot Happen.” Paul Allen and Mark Greaves answered Kurzweil directly in MIT Technology Review in October 2011 with what they called the complexity brake: the deeper science goes into natural systems, the more specialized knowledge it takes to advance, and brain science moves by irregular, unscheduled conceptual shifts that do not compound on a Moore’s Law schedule. The 2045 calculation, they judged, “seems to us quite far-fetched.” Kurzweil replied on October 20: the Law of Accelerating Returns is not a physical law but an emergent statistical regularity, like the laws of thermodynamics.

The most repeated criticism is a joke. In Ken MacLeod’s 1998 novel The Cassini Division the Singularity is derided as “the Rapture for nerds,” and the phrase escaped the fiction; by June 2008 IEEE Spectrum’s special report carried the headline “Waiting for the Rapture.” The charge underneath the joke is serious: that the singularity is a secular eschatology — salvation, immortality, and a date — and belongs with religious expectation rather than forecasting.

This encyclopedia reads the charge as half right. The anatomy is apocalyptic: a coming event, a transfigured world beyond it, a community that prepares, dates set and reset as they pass. Good’s machine was due by 2000; Vinge’s window ran 2005 to 2030; Nick Bostrom argued in 1998 for superintelligence within the first third of the new century; Kurzweil holds to 2045; Modis says never. At most one of these can be right, and the earliest is already wrong — expectation surviving its failed dates is an old pattern in prophecy. What the rapture charge misses is the conduct. Apocalyptic movements do not, as a rule, publish falsifiable windows, wager $20,000 against named skeptics, grade their own predictions in public, or convert their central institution from awaiting the event to preventing it. Whatever else it is, the singularity is an eschatology that published its deadlines — and left them standing where anyone can read them.

The scoring is underway. Vinge’s thirty years came due in 2023, and the record of that year is mixed. Microsoft researchers argued that GPT-4 “could reasonably be viewed as an early (yet still incomplete) version of an artificial general intelligence (AGI) system” — the strongest such claim on record from mainstream research, hedged twice in one sentence and contested since. The same year, a survey of 2,778 researchers who had published in top AI venues put the chance of machines outperforming humans in every possible task at 10 percent by 2027 and 50 percent by 2047 — thirteen years earlier than the previous year’s survey — with between 38 and 51 percent of respondents giving at least one chance in ten to outcomes as bad as human extinction. Expert opinion moved sharply toward the singularity camp without declaring the event arrived; no consensus holds that superhuman intelligence exists. Vinge died on March 20, 2024, at seventy-nine — a year past his horizon, six years short of the far edge of his window. The dates that remain stand where their authors set them: 2029, 2030, 2045. They will be graded the way the others were — by arriving.

→ Related: Transhumanism · Kardashev Scale · Eschatology · Prophecy