What is the AI singularity?
The AI singularity is a hypothetical moment when machine intelligence passes human intelligence and starts improving itself, so fast that people can't predict what follows. The idea has three dated roots.

| Year | Who | What they said |
|---|---|---|
| 1958 | Stanislaw Ulam | Reported that John von Neumann's talk "gives the appearance of approaching some essential singularity in the history of the race" (Wikipedia's account) |
| 1965 | I. J. Good | Described the first ultraintelligent machine as the last invention people need ever make, as long as it's docile enough to be controlled |
| 1993 | Vernor Vinge | Defined the singularity as "a point where our models must be discarded and a new reality rules" |
| 2005 | Ray Kurzweil | Predicted human level AI by 2029 and the singularity by 2045 in The Singularity Is Near (Wikipedia's account) |
Good's idea is called the intelligence explosion. A machine that can design better machines will design a better one still, and the loop runs away from us. Vinge's version, written for a NASA symposium in March 1993, said that within thirty years we'd have the means to create superhuman intelligence, and that shortly after, the human era would be ended.
Vinge first pushed the term in a 1983 Omni op-ed, per Wikipedia. His 1993 essay lists four routes to it: machines that wake up superhumanly intelligent, computer networks that become a superintelligent whole, interfaces so close that their users count as superhuman, and biology that enhances the human mind. He also notes that von Neumann seems to have meant ordinary progress by his remark, not superhuman minds.
Today the loop has a name in AI research, recursive self improvement. A model would write code or design chips that make the next model better, and each round would need less human help. Nobody has shown a system doing that end to end, and that's the gap the critics point at.
What is the difference between the AI singularity and the technological singularity?
"AI singularity" and "technological singularity" are close to the same phrase. The technological singularity is Vinge's wider term: any technology that produces greater than human minds, including brain computer links and biology. The AI singularity keeps only the machine route, because that's the one now in the news.
Some writers extend the idea to living forever. Per Wikipedia, Moravec predicted mind uploading in 1988 and Kurzweil argued in 2005 that medicine could make life expectancy effectively limitless. Those are separate claims from the self improving machine, and the dated forecasts below cover only the machine claim.
Who predicted the singularity, and when?
Eight named people or groups have put dates on it, and the dates stretch from 2000 to 2047. The first one has already passed.
| Who | Said | The forecast | Status as of November 2026 |
|---|---|---|---|
| I. J. Good | 1965 | More probable than not that an ultraintelligent machine would be built in the 20th century (Wikipedia's account) | The century ended in 2000 with none built |
| Vernor Vinge | 1993 | "I'll be surprised if this event occurs before 2005 or after 2030" | 2005 passed with no such event, and 2030 is still ahead |
| Hans Moravec | 1998 | Human level AI by 2040 and far beyond human by 2050 (Wikipedia's account) | Not yet due |
| Ray Kurzweil | 2005, repeated 2024 | Human level AI by 2029, the singularity by 2045 (Wikipedia's account) | Not yet due |
| Dario Amodei | Oct 2024 | Powerful AI "could come as early as 2026, though there are also ways it could take much longer" | The window runs to the end of 2026 |
| Sam Altman | Jun 2025 | "We are past the event horizon; the takeoff has started." | Disputed, see below |
| Sam Altman | 2025 | AI surpasses human intelligence across the board by 2030 (Al Jazeera's report) | Not yet due |
| Sam Altman | 25 Jul 2026 | "We're now, like, in the singularity" (Al Jazeera) | Disputed, see below |
| Demis Hassabis | May 2026 | Humanity is in the "foothills of the singularity" (Al Jazeera) | An opinion with no date |
| 2,778 AI researchers | Survey published 2024 | Unaided machines beat humans at every task: 10% by 2027, 50% by 2047 | Not yet due |
Every row names a speaker, a date and a source you can open. Rows marked "Wikipedia's account" come from the Wikipedia article and haven't been checked against the original book or paper. The earlier forecasts missed early and the later ones haven't come due, which is the only fair thing to say about a list like this.
For Altman's other dated calls, see Sam Altman's AGI predictions, which grades 19 of them.
Has the AI singularity already happened?
Nobody has shown that it has, and the people closest to the question disagree. Altman made the strongest claim on the Relentless podcast on 25 July 2026, and Hassabis said in a Google I/O keynote in May 2026 that we are in the "foothills of the singularity", which is a softer claim (both per Al Jazeera).
Kai Riemer and Sandra Peter, professors at the University of Sydney, answered in The Conversation on 30 July 2026. They say a singularity needs two things, and today's AI has neither. Models stay frozen after training, so they can't improve themselves while running. And "beyond human intelligence" assumes human and machine minds sit on one scale, which they reject.
Sean O hEigeartaigh of the University of Cambridge told Al Jazeera "We aren't there yet", and said he agrees more with Hassabis's foothills wording. Altman's own 2025 post is gentler than the headlines. It calls the change a gentle singularity and expects systems that "figure out novel insights" in 2026, which is a forecast about capability and not a claim about self improvement.
Altman and Hassabis also gave different dates for the same feeling. Altman's 2025 post said the takeoff had started, and his 2026 remark put us inside it. Hassabis kept a distance with the word foothills, which implies the climb is still ahead. Both speak for labs that build the models, so read their words as a view from inside the industry and not as an outside audit.
The term is slippery for a reason. Vinge defined it by what happens after, a point where predictions fail. A person who says we're in it is claiming the predictions already fail, and that's a judgment call, not a measurement.
How does the singularity relate to AGI and ASI?
AGI is a level of ability, ASI is a higher level, and the singularity is an event those levels might trigger. They aren't synonyms.
| Term | What it names | Where it comes from |
|---|---|---|
| AGI | AI that matches people across most tasks | Defined differently by each lab, see AGI vs AI |
| Artificial superintelligence (ASI) | AI that beats every human, not only at one thing | Level 5 Superhuman in DeepMind's Levels of AGI paper, listed there as not yet achieved |
| Singularity | The runaway moment after machines pass us and improve themselves | Good 1965 and Vinge 1993 |
You can have AGI with no singularity. A system as good as an average person at most jobs, running on a fixed model, isn't improving itself. You can also argue for a singularity with no AGI, if narrow tools speed up AI research until progress outruns people. That's the version Amodei weighs in his October 2024 essay, and he rejects the fast form: "Intelligence may be very powerful, but it isn't magic fairy dust."
Researchers split on speed too. A hard takeoff means the jump from human level to far beyond takes days or weeks. A soft takeoff means years. Altman's "gentle" wording and Amodei's physical limits are both soft takeoff arguments.
Why do critics say the singularity may never come?
Critics say progress hits limits, and that intelligence isn't one number you can multiply. Wikipedia's article collects the main objections, and the ones below are the ones a reader can check.
Steven Pinker wrote in 2008 that there is "not the slightest reason to believe in a coming singularity". Paul Allen proposed a complexity brake, where each advance makes the next one harder to find. Robin Hanson doubted that the easy gains in intelligence would last. Jeff Hawkins argued that physical limits on computing power mean there would be no singularity, and Theodore Modis says it can't happen because people confuse logistic curves with exponential ones. Daniel Dennett called singularity talk "preposterous" in 2017 and said it distracts from more pressing problems, and Jaron Lanier rejects the idea that technology creates itself. Wikipedia lists these as critic positions and doesn't rank them.
Experts split on whether the core argument holds. Wikipedia reports a 2017 email survey of authors at the 2015 NeurIPS and ICML conferences: 12% found the intelligence explosion argument quite likely, 17% likely, 21% about even, 24% unlikely and 26% quite unlikely. Half said unlikely, and 29% said likely.
The newest objection is Riemer and Peter's. They say the premise that humans and machines sit on one scale is the weak link, because AI is far above people at some tasks and far below at others. For the safety side of the argument, read AI alignment and will AI take over the world.
What do three real singularity claims look like when you check them?
Three claims from this page show the method. Each one gets its date, its test and a verdict that stops at what the sources support.
- Good's 20th century call. Wikipedia reports he judged it more probable than not that an ultraintelligent machine would be built in the 20th century. The century ended in 2000. A check on the date alone settles it, and it shows how early forecasts of this kind ran ahead of events.
- Vinge's window. He wrote that he'd be surprised if the event came before 2005 or after 2030. The first edge passed years ago. The second hasn't. A fair reading is that the window is still open and the event is disputed, and nothing in the 1993 essay says what counts as the event happening beyond "a new reality rules".
- Altman's July 2026 claim. The quote is on record, so the claim is checkable against Vinge's two tests, self improvement and passing humans across the board. Riemer and Peter say it fails both. Hassabis's softer wording passes without a test because it has no date. The way to settle a case like this is to name the test before reading the answer.
What can you ask an AI about the singularity?
A model is a good partner for checking singularity claims if you make it show its sources. These five prompts work in Claude, ChatGPT, Gemini or Whizi, and each asks for a link you can open.
- Find the original: "Find the original source for this claim and quote the exact sentence with its date: [claim]. If you can't find the primary source, say so."
- Compare definitions: "Compare how Vernor Vinge, Ray Kurzweil and I. J. Good define the singularity. Give a table with the year, the definition and a link for each."
- Check a forecast date: "Here is a forecast: [person, year, claim]. What was the target date, has it passed, and what happened? Use web search and mark anything you couldn't source."
- Steelman the critics: "Give the strongest argument against the technological singularity, with the named person and year behind each point, and tell me which point is weakest."
- Separate the terms: "Explain the difference between AGI, artificial superintelligence and the singularity in one sentence each, then give one example that fits one term but not the others."
Treat the answers as leads. Models can misdate a quote or merge two speakers, so open the link behind every year and name before you repeat it.
Where does Whizi fit when you read about the singularity?
Whizi is one app where you can ask Claude, GPT and Gemini the same question, which suits a topic with this many disputed quotes. Web search is on by default in a new chat, and side by side Compare puts two models on one prompt on Pro.
The free plan gives you 10 messages a day on Whizi AI, Auto and Claude Haiku, with no card, and Pro's first week is $1.99. A dedicated reference site is quicker for a single date, and Wikipedia's article is the better place to browse the history. See Whizi pricing for what each plan holds.
What can an AI not tell you about the singularity?
It can't tell you whether the singularity will happen, because nobody knows, and a model that sounds sure is guessing. The dated forecasts above are all human opinions with a record of missing.
A model also trains on text that already repeats the famous quotes, so it can blur who said what. Altman's July 2026 line appears with slightly different wording in different outlets, which is the kind of gap a model smooths over. Check the original recording or transcript before you quote it. A quote that moves between outlets, gets shortened in a headline and then lands in a model's training data can end up attributed to the wrong year and the wrong speaker. Don't trust it.
For the history behind the idea, read when was AI invented. Treat any chatbot's date for the singularity as an echo of the forecasts in the table, not as a prediction of its own.
- Find the speaker's exact words and the date before repeating a singularity quote
- Name which definition is in use: beat humans, improve itself, or merge with machines
- Check the forecast date against today before calling it right or wrong
- Keep AGI, superintelligence and the singularity as three separate terms
- Open the source behind every year and name a chatbot gives you
Frequently asked questions
Has the AI singularity happened yet?
No one has shown that it has. Sam Altman said on 25 July 2026 that we're now in it, and Demis Hassabis said in May 2026 that we're in its foothills. Two University of Sydney professors argued on 30 July 2026 that today's models can't improve themselves while running, so a key test isn't met.
When will the singularity happen?
No one knows, and named forecasts range from Vinge's 2005 to 2030 window to Kurzweil's 2045. A survey of 2,778 AI researchers gave machines beating humans at every task a 10% chance by 2027 and 50% by 2047, which is about human level AI, not the singularity itself.
Is the singularity the same as AGI?
No. AGI is AI that matches people across most tasks, while the singularity is a runaway event in which machines pass humans and improve themselves. AGI could arrive without a singularity, which is why AGI vs AI treats them as separate questions.
What is artificial superintelligence?
Artificial superintelligence, or ASI, is AI that outperforms every human, not only at one task. DeepMind's Levels of AGI paper calls this Level 5 Superhuman and lists it as not yet achieved, and many singularity forecasts treat ASI as the thing that triggers the event.