AGI vs superintelligence: what's the difference?

Two rungs, not one

It helps to see the terms as a ladder.

AI

Software that does tasks we associate with intelligence. It exists, and has for decades.

AGI

Roughly any intellectual task a person can do. No current system is universally accepted as this; whether today's systems are close is disputed.

Superintelligence

Greatly exceeds the best human minds in virtually all domains. Hypothetical; no such system exists.

AGI is about breadth at human level. The test, informally, is whether a system could be given the range of tasks a capable adult handles (learn a new job from instructions, reason about an unfamiliar problem, plan over weeks, switch between fields) and cope. There is no agreed formal test, which is why researchers who look at the same system can disagree about how close it is.

Superintelligence is about margin above the best humans. Nick Bostrom’s definition, the one researchers use, is “any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest”. The stress is on greatly and on virtually all. A system that is superhuman at chess or protein folding but nothing else is narrow AI. A system that is a little better than a good professional across the board is AGI, or slightly past it. Superintelligence is the point at which the comparison stops being close.

Why the gap between them matters

Work on AI risk covers a wide field. Google DeepMind’s published approach to AGI safety, dated 2 April 2025, sorts the risks into four areas (misuse, misalignment, accidents and structural risks), and most of that work concerns systems that exist now or could exist before anything reaches human level. But one particular argument, and the one most often behind the word “superintelligence” in a headline, is specifically about what might happen after AGI.

The worry, first set out by I. J. Good in 1965 and developed by Bostrom, is that designing better AI is itself an intellectual task. A system that could do that task as well as its makers could improve itself; the improved version could improve itself faster; and the process might not stop at human level. Good called it an “intelligence explosion”. On this view AGI is unstable: a brief stop on the way to superintelligence.

The counter-view is that there is no reason to expect the step from “human-level” to “far beyond human” to be quick or automatic. Intelligence may hit diminishing returns; physical bottlenecks (compute, data, experiments in the real world) may slow everything down; and “human-level” may turn out to be a wide plateau rather than a threshold.

Both camps include serious researchers, and the disagreement is not only about mechanism but about magnitude. Reuters reported that Evan Hubinger, Anthropic’s alignment lead, estimated a greater than 10 per cent chance of AI causing human extinction within a decade. That is an estimate of catastrophic risk, not a claim that AGI leads automatically or quickly to superintelligence; other researchers regard the figure as far too high. This page only needs the reader to see that the two rungs are different, and that the distance between them is disputed.

Why the words got harder to use in September 2026

Until this month the two terms had a stable, if specialist, meaning. Then the US administration introduced a third.

Over four days, “Super Intelligence” became the administration’s preferred name for artificial intelligence in general: a Truth Social poll on 19 September proposing “Superior”, “Extreme” or “Supreme” Intelligence; a second poll on 21 September between “Super” and “Superior”; and the President’s UN General Assembly address on 22 September. At the UN, the President said US documents “will be changed to use the much more accurate term Super as opposed to Artificial”, and, in the same breath, “Let’s see if I have any power. Maybe I do and maybe I don’t.” The White House’s own press release on the speech, published the same day, refers to artificial intelligence as “hereinafter officially called ‘Super Intelligence.’”

The practical effect for a reader is that the phrase now points at three different things depending on who is speaking:

Who says it	What they mean	Does it exist?
The US administration, since 22 Sept 2026	All AI, renamed	Yes: it is the technology in use today
AI researchers, since Bostrom (2014)	A system greatly beyond the best humans	No: hypothetical
Some AI companies (Safe Superintelligence Inc.; Meta, through Meta Superintelligence Labs)	A stated goal, using the same word; neither has published a definition matching Bostrom’s	No: a target, not a product

Three cautions follow. First, the administration’s usage does not imply that anyone believes superintelligence, in the research sense, has arrived; it is a naming decision about existing technology. Second, the third row records that two companies use the word for what they are aiming at; it does not establish that they mean precisely what researchers mean by it. Third, the change so far consists of two social-media polls (19 and 21 September 2026), a statement in a speech (22 September 2026) and a White House press release using the words “officially called” (22 September 2026). Those are three different things, and none of them is an executive order: as of 23 September 2026, we found no executive order, directive or federal style guide implementing the change among the presidential actions published on whitehouse.gov, and no evidence of the term in use across federal documents. This is a dated finding that this site re-checks on each review. The dated rename timeline records each step and distinguishes what was said, what was polled, what was announced and what has become policy.

How to read a sentence that uses either word

A short habit that resolves most confusion: when you meet “AGI” or “superintelligence” (or “super intelligence”) in a headline, ask which of the three rows above the writer is on.

• “OpenAI says it is building AGI”: the second rung, a company’s stated goal.
• “Researchers warn superintelligence could be dangerous”: the third rung, hypothetical, in Bostrom’s sense.
• “US documents will now say Super Intelligence”: the first row of the table, all AI, renamed, no change to the technology.

If the sentence does not make clear which is meant, the ambiguity is the writer’s, not yours.

How this site uses the words

• AI for the technology in general use.
• AGI for the hypothetical human-level-general system.
• Superintelligence only in the research sense.
• “Super Intelligence”, in quotation marks, when reporting the US administration’s usage.