What Is Super Intelligence? A Plain-English Definition

Super intelligence is any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest. That is philosopher Nick Bostrom’s definition, and it’s the one most researchers use. In practice, people usually mean a future artificial intelligence that is smarter than the smartest humans at almost everything: science, engineering, strategy, persuasion and creativity.

It does not exist yet. But since some of the largest companies on Earth now say building it is their goal, it’s worth knowing exactly what the word means.

The definition, unpacked

Bostrom’s definition has three parts, and each one matters:

1. “Greatly exceeds”: being slightly better than a human isn’t enough. A super intelligence would be to us roughly what we are to other animals in problem-solving power.
2. “Humans”: the bar is the best human minds, not the average person. Beating a typical person at a quiz doesn’t qualify.
3. “Virtually all domains”: this rules out narrow systems. A chess engine is superhuman at chess and useless at everything else, so it isn’t a super intelligence.

That last point is the one people miss most often. Calculators have outperformed humans at arithmetic for decades. Super intelligence means being superhuman at general thinking.

Where the idea came from

• 1965, I. J. Good: the British mathematician, who worked with Alan Turing at Bletchley Park, described an “ultraintelligent machine” that could design even better machines. The result, he wrote, would be an “intelligence explosion”, and such a machine would be “the last invention that man need ever make.”
• 1993, Vernor Vinge: the science fiction author and mathematician popularized the technological singularity, the point after which superhuman intelligence makes the future impossible to predict.
• 1998 and 2014, Nick Bostrom: the Oxford philosopher formalized the term in a 1998 paper and in his 2014 bestseller Superintelligence: Paths, Dangers, Strategies, which brought the topic to Bill Gates, Elon Musk and the AI labs.
• 2023–2026: rapid progress in large language models moved the topic from philosophy seminars to company strategy. OpenAI, Meta, Safe Superintelligence Inc. and others now openly talk about building it. In September 2026 the US government even began calling all AI “Super Intelligence” (see why AI is now called SI).

The three types of super intelligence

In Superintelligence, Bostrom describes three forms. Any of them would be extraordinarily powerful.

Speed and collective super intelligence are, in principle, things we could get just by scaling up human-level AI. Quality super intelligence is the most uncertain, and the most alien.

How could super intelligence actually arise?

Researchers discuss several possible paths:

• Scaling AI: the path most labs are betting on today. You train larger models on more data and compute, add reasoning, tools and memory, and see if general ability keeps rising.
• Recursive self-improvement: once an AI is good enough at AI research, it helps build its successor, which builds a better one, and so on. This is Good’s intelligence explosion, and it’s the scenario behind forecasts like AI 2027.
• Whole brain emulation: scanning and simulating a human brain in software, then running it faster or copying it. This is far off and rarely considered the leading path now.
• Biological enhancement: genetic selection or brain–computer interfaces that raise human intelligence. Most experts think this would be too slow to compete with machines.

What super intelligence is not

• Not today’s chatbots. ChatGPT, Claude and Gemini are impressive and general-purpose, but they still make basic errors and don’t beat top experts across the board. For the full comparison, see super intelligence vs AI.
• Not the same as AGI. AGI (artificial general intelligence) means roughly human-level ability. Super intelligence is the level after that. We explain the gap in AGI vs ASI.
• Not necessarily conscious or evil. Super intelligence is about capability, not feelings or intentions. The worry experts raise is not malice but misalignment: a highly capable system pursuing goals that don’t match ours.
• Not a robot. It could exist purely as software in a data center.

Why it matters

If super intelligence arrives, it may be the most consequential event in human history, for better or worse.

The case for hope: a system smarter than every scientist could speed up medicine, energy, materials and climate research by decades. Many advocates imagine cures for diseases, abundant clean energy and a big jump in living standards.

The case for caution: a system smarter than us would be hard to control or correct. In 2023 hundreds of AI researchers signed a one-sentence statement that “mitigating the risk of extinction from AI should be a global priority,” and in 2025 a broad coalition called for a ban on developing superintelligence until it can be done safely (see the superintelligence statement).

Both camps agree on one thing: it is worth getting right before it happens, not after.

How close are we?

Nobody knows. Forecasts range from “within a few years” (some lab leaders and the AI 2027 scenario) to “decades away” or “never” (many skeptics). What has changed is that serious people now argue about when, not if. A fuller look at predictions lives on our super intelligence pillar page.

The bottom line

Super intelligence is a hypothetical intelligence, almost certainly an AI, that would vastly outperform the best humans in nearly every area of thought. The idea is 60 years old, it was formalized by Nick Bostrom, and it’s now the explicit goal of several of the world’s most powerful companies. Understanding the definition is the first step to following one of the most important stories of this decade.