Superintelligence in AI: what it means, explained

In plain words

Machines already beat us at narrow things: arithmetic, chess, finding a route. Superintelligence imagines a system that beats the best human experts at almost every mental task at once.

Philosopher Nick Bostrom described it in 1998 as an intellect "much smarter than the best human brains in practically every field, including scientific creativity, general wisdom and social skills."

Nothing like it exists, as of October 2026. It is an idea people use to think ahead. Some researchers argue that once AI can improve AI, abilities could rise quickly. Others think intelligence does not scale that simply.

The idea matters because of the stakes. If such a system were built, keeping it in line with human goals could be very hard. That concern drives much alignment and AI safety work, and plenty of debate about whether the worry is overblown.

Everyday example

A beginner against a grandmaster, in everything

Picture a club chess player facing a grandmaster: the beginner cannot even see why they are losing. Now imagine that gap not only in chess, but in medicine, engineering, law, negotiation and science, all at once. That is what people mean by superintelligence. It is a thought experiment, not a description of anything built.

Where the example breaks: Chess has a clear score; intelligence does not sit on one ladder. Whether one system could be far ahead in every field at once is exactly what people disagree about.

Go deeper

How it actually works

The idea is older than modern AI, and most of the debate is about speed and control.

A machine that builds better machines
In 1965 mathematician Irving John Good wrote that an "ultraintelligent machine" could design even better machines, and that the first one would be "the last invention that man need ever make, provided that the machine is docile enough to tell us how to keep it under control."
From AGI to superintelligence
Superintelligence is usually placed after AGI. One view is that AI doing AI research could make that step fast. Another is that progress will be held back by data, energy, real-world experiments and diminishing returns.
The control problem
A far more capable system pursuing even slightly wrong goals could be hard to correct. Bostrom's 2014 book Superintelligence made this argument widely known. Alignment research tries to make AI goals match human intentions.
Public statements
In May 2023 many AI researchers and lab leaders signed a one-line statement that "mitigating the risk of extinction from AI should be a global priority." In October 2025 a Future of Life Institute statement called for a prohibition on developing superintelligence until there is broad scientific consensus it can be done safely and strong public buy-in.

Also called: ASI, artificial superintelligence
Early version of the idea: Irving John Good's "ultraintelligent machine", 1965
Best-known book: Superintelligence: Paths, Dangers, Strategies, Nick Bostrom, Oxford University Press, 2014
Status: Hypothetical; no such system exists as of October 2026

Should we worry about superintelligence now?

One view

Yes, it deserves serious attention now

Some leading researchers and lab leaders argue AI is improving fast enough that systems beyond human level could arrive within years or decades, and that controlling them may be extremely hard. They say safety research and rules should come before such systems, not after.

The other view

No, it distracts from real problems

Other researchers argue superintelligence is speculative, that intelligence will not simply keep scaling, and that the focus pulls attention away from present harms such as bias, misinformation, surveillance and job disruption.

Most likely: Both concerns are held by credible experts, and many people work on both at once: today's harms and long-term safety. No one can say whether or when superintelligence will be built.

Questions people ask

Does superintelligence exist?

No. As of October 2026 no AI system is far beyond the best humans across nearly all fields. Some systems are superhuman at specific tasks, such as chess, which is a much narrower thing.

What is the difference between AGI and superintelligence?

AGI means roughly matching human ability across most thinking tasks. Superintelligence means greatly exceeding the best humans in practically every field. AGI is the human-level idea; superintelligence is the beyond-human one.

Who came up with the idea of superintelligence?

Mathematician Irving John Good described an "ultraintelligent machine" that could design better machines in 1965. Philosopher Nick Bostrom defined superintelligence in a 1998 paper and set out the risks in his 2014 book Superintelligence.

Why are people worried about superintelligence?

Because a system much smarter than us, with goals even slightly different from ours, might be very hard to correct or stop. This is called the control or alignment problem. Critics reply that the scenario is speculative and present-day harms deserve more attention.

What did the 2025 superintelligence statement say?

In October 2025 the Future of Life Institute published a short statement calling for a prohibition on developing superintelligence until there is broad scientific consensus it can be done safely and controllably, and strong public buy-in. Researchers, public figures and members of the public signed it.

Superintelligence