Plain-English definitions of the words that come up when people talk about superintelligence. Where a term has more than one definition in use, the entry says so. The guides explain the bigger ideas in full.

A

AI alignment
Research on making AI systems pursue the goals their developers and users intend. OpenAI's 2023 Superalignment team was set up to solve alignment for systems much smarter than humans. Guide: The superintelligence debate Source for AI alignment
AI++
David Chalmers' 2010 label for AI of far greater than human level, that is, superintelligence. He uses AI for human-level systems and AI+ for greater than human level. Guide: The intelligence explosion and the singularity Source for AI++
Artificial general intelligence (AGI)
AI that performs well across a wide range of tasks at roughly human level or better. There is no agreed definition; OpenAI's Charter uses "highly autonomous systems that outperform humans at most economically valuable work." Guide: AGI vs superintelligence Source for Artificial general intelligence (AGI)
Artificial superintelligence (ASI)
Superintelligence achieved by a machine. Google DeepMind's Levels of AGI framework uses the term for general AI that outperforms 100% of humans, and lists it as not yet achieved. Guide: AGI vs superintelligence Source for Artificial superintelligence (ASI)

B

Ban Artificial Superintelligence Act
A US bill introduced in September 2026 by Senator Bernie Sanders and Representative Greg Casar (S. 5493 and H.R. 10538) that would prohibit artificial superintelligence and create a Department of Artificial Intelligence. It had not passed as of October 2026. Guide: The superintelligence debate Source for Ban Artificial Superintelligence Act

C

Collective superintelligence
In Bostrom's 2014 taxonomy, a system made of many smaller intellects that, if they communicate and coordinate well enough, act together with far greater capability than any one of them. It is strongest on work that can be split up and done in parallel. Guide: Types of superintelligence Source for Collective superintelligence
Control problem
The problem of ensuring that a system more capable than its creators does what they intend and can be corrected or stopped. I. J. Good flagged it in 1965 when he made the machine's usefulness conditional on its being "docile enough to tell us how to keep it under control." Guide: The superintelligence debate Source for Control problem

E

Existential risk from AI
The risk that advanced AI causes human extinction or a similarly permanent and severe disempowerment of humanity. The 2023 Center for AI Safety statement called mitigating extinction risk from AI a global priority. Guide: The superintelligence debate Source for Existential risk from AI

F

Frontier AI
The definition the UK government used for its 2023 AI Safety Summit: highly capable general-purpose AI models that can perform a wide variety of tasks and match or exceed the most advanced models of the day. It excludes powerful narrow systems. Source for Frontier AI
Full automation of labor (FAOL)
The survey milestone at which every human occupation becomes fully automatable. Researchers in the 2023 AI Impacts survey gave it a 50% chance only by 2116, far later than HLMI. Guide: Superintelligence timelines Source for Full automation of labor (FAOL)

G

Genie AI
In Bostrom's 2014 book, an AI that carries out a high-level command and then waits for the next one. The design problem is getting it to follow intentions rather than exact words. Guide: Types of superintelligence Source for Genie AI

H

Hard takeoff (FOOM)
A fast takeoff in which one system rapidly becomes superintelligent. "FOOM" is informal shorthand for it, used in the 2008 online debate between Robin Hanson and Eliezer Yudkowsky. Guide: The intelligence explosion and the singularity Source for Hard takeoff (FOOM)
High-level machine intelligence (HLMI)
The milestone used in AI Impacts surveys: when "unaided machines can accomplish every task better and more cheaply than human workers." The 2023 survey's aggregate forecast gave it a 50% chance by 2047. Guide: Superintelligence timelines Source for High-level machine intelligence (HLMI)
Humanist superintelligence
Microsoft AI's term, introduced in November 2025, for advanced AI that is deliberately bounded and always works in service of people. It explicitly rejects unbounded, highly autonomous systems. Guide: How AI companies define superintelligence Source for Humanist superintelligence

I

Instrumental convergence
The claim that agents with many different final goals would pursue similar intermediate goals, such as self-preservation and acquiring resources, because those help with almost any goal. Guide: The superintelligence debate Source for Instrumental convergence
Intelligence explosion
The hypothesis that a machine able to design better machines would set off a rapid chain of ever more capable systems, leaving human intelligence far behind. I. J. Good proposed it in 1965. Guide: The intelligence explosion and the singularity Source for Intelligence explosion

L

Law of Accelerating Returns
Ray Kurzweil's 2001 argument that technological progress grows exponentially because each generation of tools is used to build the next. It underlies his forecast of a singularity in 2045. Guide: The intelligence explosion and the singularity Source for Law of Accelerating Returns
Levels of AGI
A 2023 Google DeepMind framework that grades AI by performance (Emerging, Competent, Expert, Exceptional, Superhuman) and by breadth (narrow or general). General AI at the Superhuman level is labeled artificial superintelligence. Guide: AGI vs superintelligence Source for Levels of AGI

N

Narrow AI
AI built to perform one task or a narrow set of tasks, such as playing chess or predicting protein structures. A narrow system can be superhuman at its task without being general. Guide: AGI vs superintelligence Source for Narrow AI

O

Oracle AI
In Bostrom's 2014 book, an AI that only answers questions rather than acting in the world. It is often proposed as a safer design, though it can still be misused. Guide: Types of superintelligence Source for Oracle AI
Orthogonality thesis
Bostrom's claim that intelligence and final goals are independent, so almost any level of intelligence could be combined with almost any goal. Being highly capable does not by itself imply sharing human values. Guide: The superintelligence debate Source for Orthogonality thesis

P

Personal superintelligence
Meta's stated goal, set out by Mark Zuckerberg in July 2025: superintelligent AI that helps each person pursue their own goals, delivered through everyday devices. Guide: How AI companies define superintelligence Source for Personal superintelligence
Powerful AI
Anthropic chief executive Dario Amodei's preferred term for AI smarter than a Nobel Prize winner across most relevant fields, able to work autonomously and run as many copies. He summarizes it as a "country of geniuses in a datacenter." Guide: AGI vs superintelligence Source for Powerful AI

Q

Quality superintelligence
In Bostrom's 2014 taxonomy, a mind that can carry out intellectual tasks humans cannot do in practice, not merely do human tasks faster. The gap is often compared to the one between humans and other animals. Guide: Types of superintelligence Source for Quality superintelligence

R

Recursive self-improvement
A process in which an AI system improves its own design, and each improved version is better at making further improvements. It is the mechanism behind the intelligence explosion argument. Guide: The intelligence explosion and the singularity Source for Recursive self-improvement

S

Singleton
Bostrom's term for "a world order in which there is a single decision-making agency at the highest level." A superintelligence that gained decisive control could become one. Guide: Types of superintelligence Source for Singleton
Speed superintelligence
In Bostrom's 2014 taxonomy, a system that can do what a human mind does, but much faster. An emulated human brain on faster hardware is the usual example. Guide: Types of superintelligence Source for Speed superintelligence
Statement on Superintelligence
A one-sentence statement released by the Future of Life Institute on October 22, 2025, calling for a prohibition on developing superintelligence, not lifted before there is broad scientific consensus that it will be done safely and controllably, and strong public buy-in. Guide: The superintelligence debate Source for Statement on Superintelligence
Superalignment
OpenAI's name for the problem of steering and controlling AI systems much smarter than humans, and for the team it formed in July 2023 to work on it within four years. Guide: How AI companies define superintelligence Source for Superalignment
Superintelligence
An intellect that far exceeds the best human minds across virtually all fields, not in one narrow skill. Nick Bostrom's 2014 definition is "any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest." Guide: What is superintelligence? Source for Superintelligence

T

Takeoff
The transition from roughly human-level AI to superintelligence. Bostrom distinguishes slow (decades or centuries), moderate (months or years) and fast (minutes to days) takeoff. Guide: The intelligence explosion and the singularity Source for Takeoff
Technological singularity
A hypothesized point after which technological change, usually driven by superhuman intelligence, becomes too fast to predict from before it. Vernor Vinge tied the idea to superhuman intelligence in 1993. Guide: The intelligence explosion and the singularity Source for Technological singularity
Transformative AI
A term introduced in 2016 by Open Philanthropy, now called Coefficient Giving, in posts by Holden Karnofsky. It refers to AI that precipitates a transition comparable to or more significant than the agricultural or industrial revolution, so it defines AI by its effects, not its abilities. Guide: AGI vs superintelligence Source for Transformative AI

U

Ultraintelligent machine
I. J. Good's 1965 term for "a machine that can far surpass all the intellectual activities of any man however clever." Guide: What is superintelligence? Source for Ultraintelligent machine

W

Weak superintelligence
An intellect with about the same abilities as a human brain but much faster. Bostrom's 1998 paper notes that such a mind "has been called" a weak superintelligence. It anticipates the later category of speed superintelligence. Guide: Types of superintelligence Source for Weak superintelligence
Whole brain emulation
Scanning a brain in fine detail and running a software model that behaves essentially like the original. Anders Sandberg and Nick Bostrom set out a technical roadmap in 2008. Guide: Types of superintelligence Source for Whole brain emulation