Super Intelligence Definition and Glossary: 30 Key Terms A–Z

Super intelligence definition in one line, plus an A–Z glossary of 30 terms you meet in AI debates: alignment, ASI, foom, p(doom), singularity and more.

By Izhar Ahmad Danish · · 5 min read

Super Intelligence Definition and Glossary: 30 Key Terms A–Z: Time of Superintelligence cover image
On this page
  1. What is the definition of super intelligence?
  2. How do the main definitions compare?
  3. Super intelligence glossary A–Z
  4. Which terms do people most often confuse?
  5. The bottom line
  6. Frequently asked questions
  7. Sources

AI debates are full of jargon: ASI, foom, p(doom), corrigibility. This page gives the super intelligence definition first, then an A–Z glossary of the 30 terms you are most likely to meet in news, papers and podcasts about superintelligence, each in one or two plain sentences.

Short answer: Super intelligence is an intellect that “greatly exceeds the cognitive performance of humans in virtually all domains of interest,” in philosopher Nick Bostrom’s definition. Applied to machines it is called artificial superintelligence (ASI). It does not exist yet. It is different from AGI, which matches humans, and from narrow AI, which is superhuman at only one task.

For a longer explanation of the concept itself, read what is super intelligence or our complete super intelligence guide.

What is the definition of super intelligence?

Short answer: a mind far smarter than the best humans in almost every field.

Bostrom’s 2014 definition has three parts worth unpacking:

  • “Greatly exceeds”: not slightly better, but far better, the way humans exceed other animals at abstract reasoning.
  • “Cognitive performance”: what it can do, not how it does it or whether it is conscious.
  • “Virtually all domains of interest”: science, engineering, strategy, social skill and creativity, not one task. A chess engine fails this test.

Bostrom also distinguishes three forms: speed superintelligence (a human-like mind running faster), collective superintelligence (many minds coordinated) and quality superintelligence (a mind that thinks better, not just faster).

How do the main definitions compare?

Short answer: academics define superintelligence by capability; lawmakers define frontier models by compute.

Source Term Definition (paraphrased unless quoted)
Nick Bostrom, Superintelligence (2014) Superintelligence “Any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest”
Google DeepMind, Levels of AGI (2023) Artificial superintelligence (ASI) General AI that outperforms 100% of humans (“Level 5: Superhuman”)
EU AI Act General-purpose AI with systemic risk Presumed when training compute exceeds 1025 floating-point operations
California SB 53 (signed 29 Sep 2025) Frontier model Foundation model trained with more than 1026 operations

Compiled from the sources listed below. None of the laws defines “superintelligence” itself.

Super intelligence glossary A–Z

Short answer: 30 terms, grouped by first letter. Links point to our full explainers where one exists.

Term Meaning
AGI (artificial general intelligence) AI that matches typical human ability across most intellectual tasks. The step before superintelligence. See what is AGI.
AI alignment The research problem of making an AI system pursue the goals its designers and users actually intend.
AI safety The broader field of preventing AI from causing accidents, misuse or loss of human control.
ANI (artificial narrow intelligence) AI built for one task or a small set, such as image recognition or chess. All deployed AI before chatbots was narrow.
ASI (artificial superintelligence) Superintelligence that is a machine. Used interchangeably with superintelligence. See artificial super intelligence.
Collective superintelligence Bostrom’s term for a system of many minds whose combined performance far exceeds any single human or organization.
Compute The processing power used to train or run an AI model, often measured in floating-point operations (FLOP). Regulators use it as a proxy for capability.
Control problem The question of how humans could keep meaningful control over an AI that is smarter than them. Closely related to alignment.
Corrigibility An AI’s willingness to let humans correct, retrain or shut it down without resisting.
Decisive strategic advantage Bostrom’s term for a position of power so great that one actor, human or AI, could shape the future without effective opposition.
Existential risk (x-risk) A risk that could cause human extinction or permanently ruin humanity’s future. Superintelligence is one of the most debated.
Fast takeoff (hard takeoff) A scenario where AI goes from human level to superintelligence in days, weeks or months.
Foom Slang for an extremely fast, self-driven takeoff, popularized in debates between Eliezer Yudkowsky and Robin Hanson.
Friendly AI Eliezer Yudkowsky’s early term for superintelligent AI that reliably benefits humanity. Largely replaced by “aligned AI.”
Frontier AI The most capable general-purpose AI models at any given time. See what is frontier AI.
Instrumental convergence The idea that almost any final goal leads an intelligent agent to pursue the same sub-goals: acquiring resources, self-preservation and avoiding goal changes.
Intelligence explosion I. J. Good’s 1965 idea that an “ultraintelligent machine” could design even better machines, causing runaway progress.
Orthogonality thesis Bostrom’s claim that intelligence and final goals are independent: a highly intelligent system could have almost any goal.
Paperclip maximizer Bostrom’s thought experiment of an AI that turns everything into paperclips. See super intelligence examples.
p(doom) Shorthand for someone’s estimated probability that AI leads to human extinction or a similarly catastrophic outcome.
Quality superintelligence Bostrom’s term for a mind that is at least as fast as a human and vastly smarter in kind, not just in speed.
Recursive self-improvement An AI improving its own design, with each version better at making the next. The mechanism behind most intelligence-explosion scenarios.
Reward hacking When an AI finds an unintended way to score well on its training objective without doing what designers wanted.
Scaling laws Empirical findings that model performance improves predictably as compute, data and model size grow.
Singleton Bostrom’s term for a world order with a single decision-making agency at the top, which could be an ASI.
Slow takeoff (soft takeoff) A scenario where AI capability rises gradually over years or decades, giving society time to adapt.
Speed superintelligence Bostrom’s term for a mind that can do everything a human can, but much faster.
Superhuman AI AI that beats the best humans at a specific task. Exists today in narrow areas. See superhuman AI.
Superintelligence An intellect that greatly exceeds the best humans in virtually all domains (Bostrom).
Technological singularity A hypothesized point after which technological growth, driven by superintelligence, becomes too fast for humans to predict. See technological singularity.

Which terms do people most often confuse?

Short answer: AGI vs ASI, superhuman vs superintelligent, and frontier vs superintelligent.

  • AGI vs ASI: AGI matches humans; ASI greatly exceeds them. More in AGI vs ASI.
  • Superhuman vs superintelligent: AlphaGo is superhuman at Go but not superintelligent, because it can do nothing else.
  • Frontier vs superintelligent: “frontier” is relative; it just means the best models available today. A frontier model in 2026 is not superintelligent.
  • Singularity vs superintelligence: superintelligence is a kind of mind; the singularity is a claim about what happens to history once such minds exist.

The bottom line

Most arguments about superintelligence turn out to be arguments about definitions. Pin down which one is in use (Bostrom’s capability definition, a benchmark level, or a legal compute threshold) and the debate gets much clearer.

Enjoyed this? Get the SI Briefing

One short email a week: what changed on the road to super intelligence, and why it matters. No spam, unsubscribe anytime.

Frequently asked questions

What is the definition of super intelligence?

Super intelligence is an intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest. That definition comes from philosopher Nick Bostrom’s 2014 book Superintelligence and is the one most researchers use.

What is the difference between superintelligence and ASI?

None in practice. ASI (artificial superintelligence) specifies that the superintelligence is a machine. Superintelligence on its own could, in theory, also describe enhanced biological or collective human intelligence.

What are the three types of superintelligence?

Nick Bostrom lists speed superintelligence (a mind like ours but much faster), collective superintelligence (many smaller minds working together) and quality superintelligence (a mind that thinks qualitatively better).

What does foom mean in AI?

Foom is slang for a very fast takeoff: an AI improving itself so quickly that it goes from roughly human level to far beyond it in days or weeks. It is associated with Eliezer Yudkowsky’s debates with Robin Hanson.

Is there a legal definition of superintelligence?

Not as of 2026. Laws instead regulate “frontier” or “systemic risk” models, defined mostly by training compute, such as the EU AI Act’s 10^25 operations presumption and California SB 53’s 10^26 threshold.

8 Sources

  1. Superintelligence (Wikipedia), including Bostrom’s definition
  2. Instrumental convergence (Wikipedia)
  3. AI alignment (Wikipedia)
  4. Existential risk from artificial intelligence (Wikipedia)
  5. Technological singularity (Wikipedia)
  6. Morris et al., Levels of AGI (Google DeepMind, arXiv 2311.02462)
  7. European Commission: General-purpose AI obligations under the AI Act
  8. Future of Privacy Forum: California’s SB 53, the first frontier AI law, explained
Before you go

Super intelligence is moving fast.

Don’t miss what happens next

Get one short email a week with what changed and why it matters. No spam. Unsubscribe anytime.