What is superintelligence?
The established meaning of the word in AI research, where it comes from, why it is not a description of today's AI, and why it matters more now that the word is in the news.
In short: In AI research, superintelligence means an artificial intelligence that greatly exceeds the cognitive performance of humans in virtually all areas that matter — science, planning, persuasion, engineering, and the rest. It is a hypothetical future system, not a product you can buy. No AI that exists today meets the definition. The word has been in the news since September 2026 because the US administration adopted “Super Intelligence” as its name for AI in general; that is a different use of the word, and this site keeps the two apart.
The definition
The standard definition comes from the philosopher Nick Bostrom, whose 2014 book Superintelligence: Paths, Dangers, Strategies did more than anything else to put the word into general use:
“any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest.”
Three parts of that sentence do the work.
“Greatly exceeds.” Not slightly better, and not merely faster. A pocket calculator beats every human at long division; nobody calls it superintelligent. The bar is a large margin over the best humans, not the average.
“Virtually all domains.” Being superhuman at one thing does not count. A chess program beat the reigning world champion in 1997, and today’s protein-structure prediction systems solve, in minutes, problems that used to take laboratories years. They are narrow. Superintelligence, as the word is used in research, is general: it would be superhuman at nearly everything a mind can be asked to do.
“Intellect.” The definition is about capability, not about consciousness, feelings or intentions. Whether a superintelligent system would want anything is a separate question, and one the research community argues about. The definition itself takes no position.
Where the idea comes from
The concept is older than the word.
In 1965 the mathematician I. J. Good, who had worked with Alan Turing at Bletchley Park, described what he called an “ultraintelligent machine” — one that “can far surpass all the intellectual activities of any man however clever”. Good pointed out that designing better machines is itself an intellectual activity, so such a machine could design better machines still, producing what he called an “intelligence explosion”. He added, in a sentence that is quoted more often now than at any time since he wrote it, that “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.”
In 1993 the computer scientist and novelist Vernor Vinge used the term “the technological singularity” for the point at which such a runaway becomes possible, and predicted it “within thirty years”. That prediction has not come true.
Bostrom’s 1998 essay “How Long Before Superintelligence?” and then the 2014 book fixed the word and the definition that researchers use today. The book also set out why such a system, if it were built, might be very hard to control — an argument that shaped the field now called AI safety or AI alignment.
Three kinds, according to Bostrom
The book distinguishes three forms a superintelligence might take. They are worth knowing because people often mean different ones without saying so.
| Form | What it means | Everyday analogy |
|---|---|---|
| Speed superintelligence | A mind like a human’s, but running much faster — thinking in an hour what a person thinks in a year. | A researcher who never sleeps and reads the whole library before lunch. |
| Collective superintelligence | Many ordinary-level minds working together so well that the whole outperforms any individual by a large margin. | A civilisation, or a company, but with far tighter coordination than any real one. |
| Quality superintelligence | A mind that is not merely faster or more numerous but better — able to grasp things that humans cannot, in the way humans grasp things chimpanzees cannot. | There is no everyday analogy; that is the point. |
Most public discussion is about the third kind, or about a mixture of all three.
How it relates to AI and AGI
The three terms sit on a ladder, and the ladder matters more than any single rung.
- AI (artificial intelligence) is any software that does a task we associate with intelligence. It exists, and has for decades. Today’s chatbots, image generators and coding assistants are AI.
- AGI (artificial general intelligence) is a hypothetical AI that could learn and do roughly any intellectual task a person can. Whether current systems are close to this is disputed among researchers.
- Superintelligence is the rung above: not merely matching humans across the board, but greatly exceeding them.
The International AI Safety Report 2026, chaired by the Turing Award winner Yoshua Bengio and written by more than 100 experts with the backing of over 30 countries, is the closest thing to a consensus account of what current “general-purpose AI” systems can and cannot do. It describes rapid progress on many tasks and persistent weaknesses on others. It does not describe any existing system as superintelligent.
The difference between the three terms, in more detail.
Why the word matters now
Until September 2026, “superintelligence” was a word used mainly by researchers, philosophers and a handful of company executives. Then the US administration adopted “Super Intelligence” as a replacement name for artificial intelligence generally, and the word arrived in newspaper headlines attached to ordinary present-day software. The dated timeline of that is here.
That creates a practical problem for readers. The same word now has two meanings: the administration’s (“all AI”) and the research community’s (“hypothetical AI far beyond human ability”). A sentence like “superintelligence could arrive within a decade” is a serious forecast under the second meaning and a nonsense under the first, since AI is already here. When you meet the word, the first question is which meaning the writer intends.
This site uses “AI” for the technology in general use, “superintelligence” only in the research sense, and “Super Intelligence” in quotation marks when reporting the administration’s usage.
What this page does not settle
Whether superintelligence in the research sense can be built at all, when, and what would happen if it were, are all open questions on which serious people disagree. Pages on timelines, on the safety debate and on who is building it are in preparation and will be linked from here when published. In the meantime, the glossary covers the terms those debates use.
Sources
- Nick Bostrom, Superintelligence: Paths, Dangers, Strategies (Oxford University Press, 2014). The definition is in chapter 2; the three forms are in chapter 3.
- Nick Bostrom, “How Long Before Superintelligence?”, International Journal of Futures Studies, vol. 2, 1998.
- I. J. Good, “Speculations Concerning the First Ultraintelligent Machine”, Advances in Computers, vol. 6, 1965, pp. 31–88.
- Vernor Vinge, “The Coming Technological Singularity: How to Survive in the Post-Human Era”, NASA VISION-21 Symposium, 1993.
- International AI Safety Report 2026, chaired by Yoshua Bengio, published 3 February 2026 — internationalaisafetyreport.org.
- Reuters, “Trump says US will henceforth call AI ‘super intelligence’”, 22 September 2026 — via 933thedrive.com.
- IBM, “Deep Blue” — the 1997 match against Garry Kasparov. Google DeepMind, “AlphaFold” — protein-structure prediction.
Common questions
Does superintelligence exist?
No. In the research sense — an AI that greatly exceeds the best humans at nearly all intellectual tasks — no such system exists. Today's AI systems beat humans at some narrow tasks and fall short at many others.
Who coined the word superintelligence?
The idea goes back at least to the mathematician I. J. Good in 1965, who wrote about an "ultraintelligent machine". The word "superintelligence" was given its now-standard definition by the philosopher Nick Bostrom, in a 1998 essay and then in his 2014 book Superintelligence: Paths, Dangers, Strategies.
Is superintelligence the same as the singularity?
They are related but not identical. "The singularity" is an older, looser term — used by Vernor Vinge in 1993 — for a point at which technological change driven by machine intelligence becomes too fast for humans to follow. Superintelligence is a description of a kind of system; the singularity is a claim about what might happen once such a system exists.