artificial superintelligence (ASI)
A hypothetical artificial intelligence that would vastly surpass the best humans in virtually every intellectual task; it does not exist yet.
In plain terms
Artificial superintelligence is currently a concept, not a purchasable product. It refers to a machine that is better than us not in just one area, like a chess program in chess, but in almost everything: scientific research, design, persuasion, and strategic thinking.
By way of analogy: today's AI is like a very well-read, fast intern who sometimes makes mistakes. Human-level AI would be comparable to a skilled professional. Superintelligence, however, would relate to us as an adult scientist relates to a preschooler: we often could not follow its line of reasoning.
They talk about it so much because the stakes are high. Supporters say it could solve problems humanity has struggled with for decades, such as curing diseases or energy supply. Critics argue that the very danger is that a system smarter than us is difficult to monitor and control.
Experts disagree on when it will arrive, or if it will arrive at all; estimates range from a few years to "never".
How it works
- The starting point for today's AI: language models usable for various tasks and the agents built on them. These already perform some multi-step tasks independently, but in many areas still lag behind a knowledgeable human.
- The next assumed step is human-level general AI, which can perform any intellectual work that a human can.
- According to theory, such a system could also participate in AI research, meaning it could help design its own, better successor.
- If this cycle accelerates, capabilities could surpass human ones in a short time. This is called an intelligence explosion.
- This is an assumption: we do not know whether more computing capacity and data are sufficient, or if a breakthrough unknown to anyone is needed.
An everyday example
You read in the news that a tech company leader says 'superintelligence is coming in a few years,' and the company is building huge data centers based on this. In such cases, you are not hearing about a finished product, but about a goal and a promise, which is also addressed to investors.
Who is affected
A common misconception
Limits and risks
- There is no unified, measurable definition, making it difficult to decide when someone has achieved it.
- Predictions on timelines vary widely and often come from those with a vested business interest.
- The main risk is controllability: it remains unresolved how to ensure that a system smarter than us follows human goals.
- Debate about the distant future can distract from today's real AI problems, such as errors, data privacy, or workplace impacts.
Similar, but not the same
- general AI – Artificial general intelligence means a level roughly on par with humans, while superintelligence would far exceed it.
- Technological singularity – The singularity is a hypothetical turning point where development accelerates unpredictably, and superintelligence is the system that could trigger it.
- Narrow artificial intelligence – Narrow AI can be superhuman in specific tasks, such as chess or image recognition, while superintelligence would be so in almost every field.
Related concepts
–
general AI, Technological singularity, intelligence explosion, AI safety, goal alignment
A brief history
The basic idea was proposed by British mathematician I. J. Good in 1965, when he wrote about the self-improving 'ultra-intelligent machine' and the intelligence explosion. The concept was made widely known by philosopher Nick Bostrom's 2014 book, Superintelligence.
Where things stand
At the end of 2026, no superintelligent system exists. Several major AI companies openly declare it a goal and spend huge sums on it, but experts dispute when it can be achieved, and whether it can be achieved at all.
In the news
- Democrats promise AI regulation as they prepare to take control of US CongressNov 4, 2026 · Jog
- Microsoft diagnostic AI outperforms generalist doctors on complex casesOct 22, 2026 · Agent Age
- Microsoft's MAI-DxO in Science: twice as accurate as doctors on hard casesOct 22, 2026 · Health
- GitHub outlines "ambient intelligence" direction with physical Copilot controlsOct 19, 2026 · Code
- Hinton, Bengio and 400 scientists call for international body to oversee advanced AIOct 13, 2026 · Agent Age
- Trump administration has leading tech firms sign voluntary AI safety accordSep 28, 2026 · Agent Age
- Meta unveils Muse agent and Charm device as Amazon blocks itSep 21, 2026 · Agent Age
- Sanders and Casar bill would ban development of superintelligenceSep 21, 2026 · Agent Age
Also known as
ASI · szuperintelligencia · artificial superintelligence · szuperintelligens MI · szuperintelligens AI · artificial superintelligence · superintelligence · superintelligent AI
This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.