What is the technological singularity?

The technological singularity is a hypothesized point in the future after which technological change becomes so fast, usually because machines have become smarter than people, that the world beyond it cannot be predicted from before it. The science fiction writer and mathematician Vernor Vinge gave the idea its modern form in 1993. The inventor Ray Kurzweil dates it to 2045.

What does “technological singularity” mean?#

The technological singularity is a moment, or a short period, after which human affairs change so much and so fast that people living before it cannot meaningfully describe what comes after.

The term is a metaphor. In mathematics, a singularity is a point where a function is undefined or becomes infinite. In physics, the best-known example is the center of a black hole, hidden behind an event horizon from which no information escapes. Vinge used that image directly: in his telling, the arrival of greater-than-human intelligence is a horizon beyond which our models of the future stop working.

Most uses of the word today tie the singularity to superintelligence, meaning AI that far exceeds the best human minds across virtually every field. But the two are not the same thing. Superintelligence is a kind of mind. The singularity is what some people expect to happen to the world once such minds exist.

Who coined the term?#

The term took shape in three stages. The technological sense is usually traced to Stanislaw Ulam’s account of a conversation with John von Neumann. Vernor Vinge, often credited with coining the modern term, tied it to superhuman intelligence. Ray Kurzweil put a date on it.

Stanislaw Ulam and John von Neumann, 1958

The account appears in a 1958 tribute to von Neumann by the mathematician Stanislaw Ulam. Ulam recalled a conversation about “the ever accelerating progress of technology and changes in the mode of human life,” which gave the appearance of approaching “some essential singularity in the history of the race beyond which human affairs, as we know them, could not continue.” The passage does not connect the singularity to machines that think, and Vinge later read it as describing ordinary progress, not the creation of superhuman intellect.

Vernor Vinge, 1983 and 1993

Vinge, who died in March 2024, made the link to intelligence. In a January 1983 op-ed for Omni magazine, he wrote that when we create intelligences greater than our own, “human history will have reached a kind of singularity.” Ten years later he expanded the argument in “The Coming Technological Singularity,” a paper for a symposium sponsored by NASA Lewis Research Center. Its abstract opens:

“Within thirty years, we will have the technological means to create superhuman intelligence. Shortly after, the human era will be ended.”

— Vernor Vinge, 1993

In the same paper Vinge wrote that he would be surprised if the event occurred before 2005 or after 2030. He listed four routes to it: computers that are “awake” and superhumanly intelligent, large computer networks that “wake up” as a superhumanly intelligent entity, computer-human interfaces so intimate that users may reasonably be considered superhumanly intelligent, and biological science that improves on the natural human intellect.

Ray Kurzweil, 2005

Kurzweil turned the singularity into a dated forecast. The Singularity Is Near (2005) set it at 2045, and his 2024 sequel, The Singularity Is Nearer, kept that date along with his prediction of human-level AI by 2029. His account also has people merging with the machines, extending their own intelligence through technology rather than only being overtaken by it.

What are the three versions of the singularity?#

The word now covers three separate claims. In a 2007 essay, the AI researcher Eliezer Yudkowsky sorted them into what he called the three major singularity schools, and the split is still a useful way to read anyone who uses the word.

Yudkowsky’s point was that the three are logically distinct. Their core claims support one another, but their strong versions tend to contradict each other. The strong accelerating-change claim says the future can be charted from smooth curves. The strong event-horizon claim says it cannot be charted at all. Someone can accept one claim and reject the other two.

The intelligence explosion is the mechanism most people mean when they connect the singularity to AI. It is covered in detail in the guide to Good’s runaway-AI argument.

How is the singularity different from AGI and superintelligence?#

AGI and superintelligence describe what an AI system can do. The singularity describes what happens to the world afterward.

A world could have AGI with no singularity, if progress after it stays gradual. Under Kurzweil’s version, a singularity could also come through human enhancement rather than a single superintelligent system.

When will the singularity happen?#

No one knows, and the published dates depend on which version of the singularity the forecaster means.

• Vinge (1993): the technological means to create superhuman intelligence within thirty years, and he would be surprised if the event came before 2005 or after 2030.
• Kurzweil (2005, restated 2024): the singularity in 2045, after human-level AI in 2029. Kurzweil’s 2045 date sits beside the other major forecasts in the timelines guide.
• Sam Altman (June 2025): the OpenAI chief executive wrote that “we are past the event horizon; the takeoff has started,” describing a gradual rather than sudden process.

The 2023 Expert Survey on Progress in AI, run by AI Impacts, does not ask about “the singularity” by that name. Its closest question is on the intelligence explosion argument: that if AI systems did nearly all research and development, technological progress could become more than ten times faster within five years. Of 299 respondents, 29% rated that argument likely or quite likely to be broadly correct, 24% gave it about even odds, and 47% rated it unlikely or quite unlikely.

Has the singularity already started?#

By the classic definition, no. Sam Altman says it has, but he is using the word in a different sense, so the disagreement is mostly about the definition rather than the evidence.

In “The Gentle Singularity” (June 2025), Altman described AI, researchers, capital and infrastructure reinforcing one another, so that progress compounds without any single dramatic moment. He wrote that, at least so far, it is “much less weird than it seems like it should be,” and that living through it will probably “feel impressive but manageable.” That turns Vinge’s idea around: for Vinge, the defining feature of the singularity was that people on this side of it could not make sense of what came next. Nieman Lab’s response asked when the singularity had become “gentle.”

The older versions each supply a test. Vinge’s test is whether forecasting has broken down. Forecasters, Altman included, still publish dated predictions for the years ahead. Kurzweil’s is whether the computing trend he charts stays on track. He says it does, but his date is still nineteen years away.

What are the main criticisms?#

The objections fall into four groups. Understanding intelligence may get harder the further research goes. Being able to imagine a singularity is not evidence for one. Physical work moves at its own pace. And intelligence may not be the kind of quantity that compounds when a mind improves itself.

Progress gets harder

In 2011, Microsoft co-founder Paul Allen and the computer scientist Mark Greaves argued in MIT Technology Review that “the singularity isn’t near.” They proposed a “complexity brake”: the deeper scientists go into understanding intelligence, the more complicated it turns out to be, so progress slows rather than speeds up.

Imagining it is not evidence

The psychologist Steven Pinker told IEEE Spectrum in 2008 that there is “not the slightest reason to believe in a coming singularity,” adding that the ability to picture a future is no sign that it is likely or even possible.

Physical work sets the pace

A singularity needs the world to change quickly, and faster thinking is only part of that. Experiments, factories and approvals take physical time, and a mind that thinks faster still waits for the results. Arvind Narayanan and Sayash Kapoor make a version of this case in “AI as Normal Technology” (2025): AI’s effects, they argue, will spread over decades, slowed by safety requirements and the pace at which institutions adopt new tools.

Intelligence may not scale that way

François Chollet argued in 2017 that intelligence is situational, bound up with a body, an environment and a culture, so improving a mind on its own does not produce runaway gains. The full exchange between Chollet and his critics is in the intelligence explosion guide.

For the safety concerns that follow from a fast transition, and the public statements calling for limits on superintelligence, see the debate guide.