What is Superintelligence? — AI Glossary

What it is: Superintelligence (also called ASI — Artificial Superintelligence) is a hypothetical AI system that surpasses human intelligence across nearly every domain. It would be smarter than the smartest human at math, science, strategy, creativity, and persuasion. Whether it’s achievable, and how soon, is heavily contested.
Who it is for: Anyone following AI policy, safety research, or the public debates about AI’s long-term risks.
Best if: You want to understand the “what if AI keeps getting smarter” conversation that drives much of AI safety research.
Skip if: You only care about AI tools that exist today — superintelligence is a forward-looking concept and may never arrive. Want one practical AI workflow every morning? Subscribe to our free daily newsletter.

What is superintelligence?

Superintelligence (also called ASI — Artificial Superintelligence) is a hypothetical AI system that decisively surpasses the best human performance in essentially all cognitive tasks: scientific research, mathematics, strategy, social skills, creative work, software engineering, and beyond. It’s the level beyond AGI, where AI doesn’t just match humans but exceeds them.

The term was popularized by Oxford philosopher Nick Bostrom in his 2014 book Superintelligence: Paths, Dangers, Strategies, which laid out the safety challenges of building a superhuman AI. The concept has since become central to AI safety research and a focal point for the public debate about AI risks.

Why does superintelligence matter?

If superintelligence is possible and we build it, the consequences depend entirely on whether its goals are aligned with human values. A superintelligent system pursuing goals well-aligned with human flourishing could accelerate solutions to nearly every hard problem: disease, climate change, scientific discovery. A superintelligent system pursuing goals subtly misaligned could be catastrophic — not from malice but from competently optimizing for the wrong objective at superhuman scale.

That asymmetry — enormous upside if it goes right, existential risk if it goes wrong — is why AI labs like Anthropic and OpenAI exist with the stated mission of building safer AI. The work on alignment is fundamentally about preparing for the superintelligence scenario, even if it’s decades away.

Will we actually build superintelligence?

Honest answer: nobody knows. Three positions you’ll hear:

• Soon and serious — some researchers (Sam Altman, Dario Amodei, Geoffrey Hinton) say superintelligence is plausibly within decades and the risks are real. They advocate for accelerated safety research and careful deployment.
• Far away or never — others (Yann LeCun, many academic researchers) argue the current AI paradigm has fundamental limits that won’t reach superintelligence without major new ideas. They view concern as premature.
• Different problem — some thinkers argue we should focus less on hypothetical superintelligence and more on the near-term harms of AI systems we’re building now (bias, misinformation, job displacement).

Reasonable people disagree. What’s clear: the frontier of AI capability has moved faster in the last 5 years than most predicted, which is why even skeptics have updated their timelines toward shorter rather than longer.

Related terms

• AGI
• AI Safety
• AI Alignment
• P(doom)

Learn more on Beginners in AI

• AI Glossary

Sources and further reading

• Bostrom — Superintelligence (book)
• Anthropic — Core Views on AI Safety

Last reviewed: May 2026. AI terminology evolves quickly — verify specifics on the official source pages above.

You may also like

• AI Glossary
• AGI
• AI Safety
• AI Alignment
• P(doom)