What is superintelligence? A plain-English explainer
Last checked:
Superintelligence is a hypothetical kind of AI that would be far smarter than the best humans in nearly every field, not only in one skill. It doesn’t exist today. Experts disagree about whether today’s methods could ever get there, how quickly AI will keep improving, and how risky it would be. This page walks through those disagreements without picking a side.
You’ve probably seen the word in a headline and wondered whether you missed a memo. You didn’t. The term has been around for decades. It’s having a moment.
Where the word comes from
A well-known definition comes from the philosopher Nick Bostrom. In his paper “How Long Before Superintelligence?” (first written in 1997, published in 1998), he wrote:
“By a ‘superintelligence’ we mean an intellect that is much smarter than the best human brains in practically every field, including scientific creativity, general wisdom and social skills.”
Two parts of that do a lot of work. “The best human brains” means smarter than the top expert in each field, not smarter than the average person. And “practically every field” means across the board, not at one game or one test. A calculator beats you at long division. That doesn’t make it superintelligent, and it will not be offended to hear it.
Bostrom later expanded on the idea in his 2014 book Superintelligence: Paths, Dangers, Strategies (Oxford University Press).
Narrow AI, AGI, and superintelligence
These three terms get mixed up a lot, partly because they sound like levels in a video game. Here’s how they differ.
| Term | What it means | Does it exist? |
|---|---|---|
| Narrow AI | AI built for one specific job, like a spam filter | Yes |
| AGI (artificial general intelligence) | The Association for the Advancement of Artificial Intelligence (AAAI) says AGI “was initially defined as AI that could match or exceed human cognitive abilities across a broad range of tasks” | No agreed example |
| Superintelligence | AI “much smarter than the best human brains in practically every field” (Bostrom) | No |
Today’s AI chat tools don’t fit neatly in the first row. The International AI Safety Report 2026, written with guidance from more than 100 experts, calls them “general-purpose AI”: “AI models and systems that can perform a variety of tasks, rather than being specialised for one specific function or domain.”
They’re also uneven in a way that’s hard to picture until you see it. The Stanford HAI AI Index Report 2026 puts it this way: “AI models can win a gold medal at the International Mathematical Olympiad but cannot reliably tell time.” Researchers call that pattern the “jagged frontier.” It’s a useful reminder that being brilliant at one thing says very little about the next thing.
And AGI itself has no finish line everyone agrees on. The AAAI’s 2025 report on the future of AI research says: “AGI is not a formally defined concept, nor is there any agreed test for its achievement.” If AGI is fuzzy, superintelligence, which would go well beyond it, is fuzzier still.
Why the term is in the news
AI tools have improved quickly. The AI Index Report 2026 says plainly that “AI capability is not plateauing. It is accelerating and reaching more people than ever.” When the tools you use at work keep getting better, it’s natural to ask how far that goes.
That’s a fair question. It’s also one where the honest answer starts with “it depends who you ask.”
Where experts agree
The published reports we checked line up on a few points:
- Nobody has an agreed definition or test for AGI, which superintelligence would go beyond (AAAI 2025 report).
- Future progress is hard to predict. The International AI Safety Report 2026 says: “Whether capabilities will continue to improve as quickly as they recently have is hard to predict.”
- Today’s systems are uneven. They can be excellent at some tasks and unreliable at others (AI Index Report 2026).
Where experts disagree
This is the part the headlines tend to flatten. Here’s the range of views, attributed to the surveys and reports that describe them.
Whether it can be built this way. The AAAI surveyed 475 respondents from its research community for its 2025 report. The report says: “The majority of respondents (76%) assert that ‘scaling up current AI approaches’ to yield AGI is ‘unlikely’ or ‘very unlikely’ to succeed.” That’s a view about one path (making today’s systems bigger), not a verdict on whether AGI or superintelligence is possible at all. The same report describes “an ongoing debate about feasibility of achieving AGI.”
How fast things will move. The International AI Safety Report 2026 describes a wide range of possibilities: progress “could slow or plateau,” “continue at current rates, or accelerate dramatically.” It doesn’t pick one, and neither do we. You’ll see confident dates in some articles. The reports we checked treat the pace as uncertain.
How risky it would be. The International AI Safety Report 2026 describes the split on one of the most serious concerns, losing control of advanced AI systems:
“Experts disagree about the likelihood and potential severity of loss of control scenarios. Some believe that outcomes as extreme as the extinction of humanity are plausible. Others think that such catastrophic outcomes are implausible, arguing that AI systems will never develop the necessary capabilities or that monitoring mechanisms will identify and prevent dangerous behaviours.” (The report’s reference numbers are left out.)
The report sums it up as “a risk with uncertain likelihood but potentially extreme severity.”
What to do about it. Researchers also disagree on how to proceed. In the AAAI survey, “most respondents (70%) oppose the proposition that we should halt research aimed at AGI until full safety and control mechanisms are established.”
If that sounds unresolved, it is. When the people who study something for a living split this many ways, the honest takeaway is uncertainty, not doom or dismissal.
What this means for you
You don’t need to settle the superintelligence debate to use AI well, or to have an opinion about it. A good rule for any headline on this topic: look for who’s being quoted, whether it’s a survey or a single voice, and whether the article admits uncertainty. Learning to ask today’s AI good questions, and to check the answers, is the skill that pays off now. If you’re new, our Start here lessons cover the basics.
Try it yourself
The AI you can use today is impressive and uneven. Here’s how to get good results from it: How to write a good prompt with five habits that work for any AI task.
Want to see how an AI chat tool handles this exact topic? Paste this into any free one (our Start here lesson on picking a tool lists a few), then compare its answer with this page.
What you might get back (example; results vary)
Example output, trimmed. ChatGPT, logged out, October 17, 2026.
…
- Experts disagree on whether superintelligence is achievable, how difficult it would be to develop, and whether intelligence can keep improving without hitting fundamental limits.
- They also debate whether such an AI would follow human intentions, how reliably we could control it, and whether it would pose serious risks or bring enormous benefits.
- What remains uncertain is how superintelligence would actually work in practice: its capabilities, limitations, behavior, and consequences are not established facts, so confident claims about these are still debated.
Check its answer against the sources below. If it states a date as fact, or sounds more certain than the reports do, you’ve found a good example of why checking matters.
Sources
- Nick Bostrom, “How Long Before Superintelligence?”, International Journal of Futures Studies, vol. 2, 1998 (revised version with postscripts at nickbostrom.com). https://nickbostrom.com/superintelligence
- Nick Bostrom, Superintelligence: Paths, Dangers, Strategies, Oxford University Press, June 2014. https://www.oupcanada.com/catalog/9780199678112.html (named for background; the definition quoted above is from the paper)
- Association for the Advancement of Artificial Intelligence (AAAI), AAAI 2025 Presidential Panel on the Future of AI Research, March 2025. https://aaai.org/wp-content/uploads/2025/03/AAAI-2025-PresPanel-Report-FINAL.pdf
- International AI Safety Report 2026, February 3, 2026. https://internationalaisafetyreport.org/sites/default/files/2026-02/international-ai-safety-report-2026.pdf
- Stanford Institute for Human-Centered Artificial Intelligence (HAI), AI Index Report 2026, April 2026. https://hai.stanford.edu/ai-index/2026-ai-index-report
Filed under AI explained. Read more in our editorial policy and disclaimers.
Get one new AI how-to a week
Free, practical, and a little funny. One email a week, unsubscribe anytime.
We use Kit to send the newsletter, and Kit stores your email address. We only send the weekly how-to and never sell your address. After you sign up you'll land on Kit's page, then get an email to confirm. Privacy policy.