AGI, ASI and SI: what the words actually mean
Three abbreviations, often used as if they were the same thing. They are not, and the difference decides how forecasts about them resolve.
Published 6 October 2026 by Omar Nouha Sane
AGI, ASI and SI show up in headlines almost interchangeably. They describe very different things, and for anyone reading forecasts the difference matters: a market about one of them says nothing about the others.
AI as we use it today
The systems most people use, such as chat assistants, translation tools, image generators and coding helpers, are trained on huge amounts of data and can be impressively broad. They still fall short of skilled humans in many areas, especially in reliably carrying out long, open-ended tasks on their own. Researchers sometimes call this narrow or specialised AI, even though the line keeps moving.
AGI: artificial general intelligence
AGI describes a system that can handle most intellectual tasks at least as well as a typical human, and can learn new ones without being rebuilt for each. There is no single agreed definition. Some definitions focus on economically useful work, others on the ability to learn anything a person can learn. Because of that, forecasts about AGI usually depend on a specific test written into the question.
ASI: artificial superintelligence
ASI goes much further: a system that is far better than the best humans in practically every field, from science and engineering to strategy and persuasion. The term was popularised by the philosopher Nick Bostrom in his 2014 book Superintelligence. ASI is what the SafeASI tracker is about.
SI: the new US government term
Since 29 September 2026, “Super Intelligence” or SI is the official word that US federal agencies must use instead of AI. Despite the name, it does not mean ASI. Until a new federal definition is published, SI covers exactly what the existing legal definition of AI covers. A government document that says SI is talking about today’s AI. You can read more in AI is now “SI” in Washington.
- AI
- Today’s systems. Broad in some areas, limited in others.
- AGI
- Roughly human-level across most intellectual tasks.
- ASI
- Far beyond the best humans in almost every field.
- SI
- The US government’s new name for AI. Not the same as ASI.
From AGI to ASI: the intelligence explosion
In 1965 the mathematician I. J. Good argued that a machine able to improve its own design could set off a chain of ever faster improvements, which he called an intelligence explosion. If that happens, the step from AGI to ASI could be short. Others expect a slower path, with progress limited by computing power, energy and real-world testing.
Two markets on the tracker look at exactly this question: whether there will be less than a year between the first AGI and the first superintelligence, and whether an intelligence explosion will happen before 2035. You can find their current values in the market list.
Why definitions decide the numbers
Every prediction market resolves on its own wording, written by the person who created it. Two markets with similar titles can use different criteria and show different probabilities. That is why every number on SafeASI links to the original question, so you can check what exactly is being forecast.