What is superintelligence?

Superintelligence means an intellect that far surpasses the best human minds in nearly every field, not in one narrow skill. In a machine it is called ASI. The standard definition is from Nick Bostrom’s 2014 book Superintelligence. The idea is older: I. J. Good described an “ultraintelligent machine” in 1965. No existing system meets the definition as of October 2026.

What does “superintelligence” mean?

Superintelligence means an intellect that far surpasses the best human minds in nearly every field. The most cited wording is Nick Bostrom’s, from his 2014 book Superintelligence: Paths, Dangers, Strategies.

“Any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest.”

— Nick Bostrom, Superintelligence (Oxford University Press, 2014)

Bostrom had given an earlier version in a 1998 paper, “How Long Before Superintelligence?” There he described an intellect much smarter than the best human brains in practically every field, “including scientific creativity, general wisdom and social skills.”

Three parts of the definition do most of the work.

• “Greatly exceeds.” Matching a strong human is not enough. The gap has to be large.
• “Virtually all domains.” Breadth is the test. A system that is superhuman at one task and ordinary at the rest does not qualify.
• “Any intellect.” The definition does not say the intellect must be a computer. Bostrom’s 1998 paper leaves open how it is built and whether it is conscious.

The breadth test is why chess engines are not superintelligences, however strong they are. Google DeepMind’s 2023 “Levels of AGI” paper lists Stockfish, AlphaZero and AlphaFold as superhuman, but only as narrow systems. Each one outperforms every human at a single kind of task. None of them can do the rest of what a human mind does.

What is superintelligence not?

Superintelligence is not another word for AGI, and it says nothing about consciousness. Most confusion about the term comes from treating it as a synonym for something nearby.

• It is not artificial general intelligence. AGI usually means roughly human-level ability across many tasks. Superintelligence means far beyond human level. Some definitions of AGI already sit close to that line, which is why the terms blur. The AGI vs superintelligence guide covers the difference.
• It is not a claim about consciousness. The standard definitions describe performance. They say nothing about whether the system has experiences.
• It is not a company or a scientific community. Bostrom’s 1998 paper excludes them on the ground that they are not intellects. His 2014 book does discuss “collective” superintelligence, a system built from many smaller minds, so the boundary is less clean than it first looks. The types of superintelligence guide explains the forms.
• It is not a product. When a company names a lab or a team after superintelligence, that names a goal. It does not report that the goal has been reached.

Who came up with the idea?

No single person. The idea is older than the word: I. J. Good described a machine that could outthink any person in 1965, and Nick Bostrom gave “superintelligence” its working definition in 1998 and made it mainstream in 2014.

1. 1958. The mathematician Stanislaw Ulam, in a tribute to John von Neumann published in the Bulletin of the American Mathematical Society, recalled a conversation about accelerating technology that seemed to approach an “essential singularity” in human history. This is usually cited as the first use of “singularity” in this sense.
2. 1965. The mathematician I. J. Good published “Speculations Concerning the First Ultraintelligent Machine.” He defined an ultraintelligent machine as one that could “far surpass all the intellectual activities of any man however clever.” He argued that such a machine could design better machines, producing an “intelligence explosion.”
3. 1993. Vernor Vinge, a mathematician and science-fiction author, presented “The Coming Technological Singularity” at a NASA-sponsored symposium. He predicted the technical means to create superhuman intelligence within thirty years.
4. 1998. Bostrom’s paper “How Long Before Superintelligence?” gave the word a working definition and asked when it might arrive.
5. 2014. Bostrom’s book made the term mainstream. It reached the New York Times science best-seller list in August 2014, and Bill Gates publicly recommended it.

Good’s most quoted line is usually cut short. He wrote that the first ultraintelligent machine is the last invention humanity need ever make, “provided that the machine is docile enough to tell us how to keep it under control.” The condition was part of the idea from the start.

The history of the explosion argument, and the case against it, is in the intelligence explosion guide.

How do AI companies use the word now?

AI companies mostly use “superintelligence” as the name of a goal, and they do not all mean the same thing by it.

• OpenAI wrote in May 2023 that it was conceivable that within ten years AI systems would exceed expert skill level in most domains. In January 2025, chief executive Sam Altman wrote that the company was beginning to turn its aim to “superintelligence in the true sense of the word.”
• Meta formed Meta Superintelligence Labs in June 2025. Mark Zuckerberg describes the goal as “personal superintelligence” for everyone.
• Safe Superintelligence Inc., founded in June 2024 by Ilya Sutskever, Daniel Gross and Daniel Levy, says it has one goal and one product: a safe superintelligence.
• Microsoft AI announced a superintelligence team in November 2025 and calls its goal “humanist superintelligence,” which it describes as bounded and in service of people rather than unlimited.
• Anthropic’s chief executive, Dario Amodei, says he dislikes the term “AGI” and uses “powerful AI” instead.

Some of these uses are much narrower than Bostrom’s. Microsoft’s version is deliberately limited. A cybersecurity firm launched a “Cyber Superintelligence Lab” in September 2026 without defining the term at all. When a company says superintelligence, the first question is whether it means all domains or one. The company definitions guide quotes each lab directly.

Why is the word back in the news?

The word moved from philosophy into business and political coverage between 2023 and 2026. For most of its history it belonged to philosophers and a small research community. Three developments changed that.

• Companies adopted it as a goal. Between 2023 and 2025, OpenAI, Meta, Safe Superintelligence Inc. and Microsoft each named superintelligence publicly as something they are building toward.
• Public statements targeted it by name. On October 22, 2025, the Future of Life Institute released a one-sentence Statement on Superintelligence calling for a prohibition on its development, not lifted before there is broad scientific consensus that it will be done safely and controllably, and strong public buy-in. Signers included AI researchers Geoffrey Hinton and Yoshua Bengio, Apple co-founder Steve Wozniak, and Prince Harry and Meghan, Duchess of Sussex.
• Lawmakers wrote bills about it. On September 8, 2026, UK Labour MP Alex Sobel introduced a bill in the House of Commons to prohibit artificial superintelligence. Later that month, Senator Bernie Sanders and Representative Greg Casar introduced the Ban Artificial Superintelligence Act of 2026 in the US Congress. Neither bill had passed as of October 2026.

The arguments on each side are laid out in the superintelligence debate guide.

Does superintelligence exist yet?

No. As of October 2026, no AI system meets the standard definition.

The DeepMind “Levels of AGI” framework, in its September 2025 revision, lists general artificial superintelligence as “not yet achieved.” Its top level, Superhuman, means a system that outperforms 100% of humans. Today’s systems reach that level only on narrow tasks.

Whether general superintelligence is possible, and when it might arrive, is disputed. A 2023 survey of 2,778 AI researchers put a 50% chance on machines outperforming humans at every task by 2047. Some lab leaders have named dates in the late 2020s. Some researchers, including Princeton’s Arvind Narayanan and Sayash Kapoor, argue that superintelligent AI is “incoherent as usually conceptualized.” The timelines guide lists the forecasts with their dates and sources.

Why does the exact definition matter?

The exact definition matters because proposed laws now turn on it, in the way some AI contracts already turn on the definition of AGI.

The US bill, S. 5493, defines artificial superintelligence as a system that has, or can easily be modified to have, either of two capabilities: it “exceeds human cognitive performance and capabilities across most domains or tasks,” or it has “sufficient capabilities to plan and execute the destruction or disempowerment of humanity.” The UK bill’s text has not been published. TIME reported that it defines superintelligent AI as a system that could disempower state authorities. Neither uses Bostrom’s words. The US test asks for “most domains or tasks” where Bostrom asks for “virtually all domains,” and it does not require the gap to be large.

A definition written for a philosophy book can stay loose. A definition written into law decides what is banned.

The practical habit is to keep a definition, its source and its date together, and to ask which one a speaker is using. The glossary collects the related terms with their sources.