Artificial intelligence, as the term is used in research, covers any system that performs tasks associated with human thinking, while superintelligence is a hypothetical level far beyond the best human minds. The philosopher David Chalmers draws the line in three steps: AI at human level, AI+ above the most intelligent human, and AI++ so far beyond that the gap resembles the one between a mouse and the smartest person. No system has been shown to reach even the second step across the board, so superintelligence remains an argument and a policy debate, not a product.
Key takeaways
- Existing AI systems are built for specific tasks or sets of tasks; superintelligence names a level of general ability that nothing has demonstrated.
- The classic argument for it, from the statistician Good in 1965, is that a machine that can design better machines could start a chain of improvements.
- Chalmers identifies the weak point of that argument: it assumes that more intelligence reliably produces proportionally better design ability.
- On October 22, 2025, the Future of Life Institute published a statement calling for a prohibition on developing superintelligence until broad scientific consensus and public support exist.
Three levels, defined
The terms below follow the definitions Chalmers gave in a 2010 paper in the Journal of Consciousness Studies. Other authors define the words differently, which is a frequent source of confusion in debates about whether a given system "counts."
| Label | Meaning in Chalmers' paper | Status |
|---|---|---|
| AI | Artificial intelligence at least as capable as an average human | Contested; depends on the test used |
| AI+ | More intelligent than the most intelligent human | Not demonstrated across the board |
| AI++ (superintelligence) | As far beyond the most intelligent human as that human is beyond a mouse | Hypothetical |
The table is a map of definitions, not a ranking of products. A chess engine beats every human player and is still a narrow tool, which shows why superiority in one task does not place a system on this scale.
Good's argument from 1965
In the article "Speculations Concerning the First Ultraintelligent Machine," published in the book series Advances in Computers in 1965, Good defined an ultraintelligent machine as one that can far surpass every intellectual activity of any person. His reasoning, as quoted in Chalmers' analysis, runs in a line: designing machines is itself an intellectual activity, so such a machine could design better machines, producing an "intelligence explosion" that leaves human intelligence behind.
The argument contains no claim about timing. It says what would follow if a machine of that kind existed, which is why the same text can be cited by people who expect it soon and by people who expect it never.
Where the argument is weakest
Chalmers sets the argument out as two premises: that AI+ will exist, and that if it does, AI++ will follow. He calls the second premise the key claim and says it rests on a proportionality thesis, the assumption that more intelligence always yields proportionally more ability to design intelligent systems. He lists ways it could fail: there may be upper limits to intelligence, returns may diminish past some point, or intelligence may not correlate well with design skill. The argument also assumes that intelligence is a single quantity that can be compared between systems. These are open philosophical questions, not settled results, and they are the reason sober sources treat superintelligence as a possibility to study rather than a forecast.
Why the topic reached policy
The Statement on Superintelligence, hosted by the Future of Life Institute, calls for "a prohibition on the development of superintelligence, not lifted before there is broad scientific consensus that it will be done safely and controllably, and strong public buy-in." The institute's October 2025 newsletter dates its release to October 22, 2025 and lists signers such as the AI researchers Yoshua Bengio, Stuart Russell, and Geoffrey Hinton, along with business figures, artists, and former members of Congress. When the signature page was checked on October 24, 2026, it showed 77,854 signatures, including 5,000 from a petition by another organization.
The statement is a call to action by an advocacy group, not a law or a measurement. It does not say that superintelligence exists or is near, and it does not define a legal threshold.
FAQ
Is ChatGPT or any current chatbot superintelligent?
No. Systems of this kind can exceed most people on some tests and fail simple tasks on others, so they do not meet a definition that requires far-above-human ability across the board.
Is superintelligence the same as AGI?
No. Artificial general intelligence usually means human-level ability across many tasks, which corresponds to the first step in Chalmers' scheme. Superintelligence sits well above it.
Can an AI become superintelligent by improving itself?
That is the intelligence-explosion argument. It is a hypothesis with a stated weak premise, not a demonstrated process.
Do experts agree it will happen?
No. The Future of Life statement shows that a large group considers the risk serious enough to call for a prohibition, while the philosophical literature still debates whether the underlying assumptions hold.
Related reading
- Documented harms of AI to society
- Whether AI can feel anything
- AI and robotics in society
- How AI systems are developed
More in AI Basics.
Last verified: October 2026
Sources
- Chalmers D., The Singularity: A Philosophical Analysis, Journal of Consciousness Studies 17, 2010
- Good IJ, Speculations Concerning the First Ultraintelligent Machine, Advances in Computers, vol. 6, 1965 (as quoted by Chalmers)
- Statement on Superintelligence, Future of Life Institute, 2025
- FLI Newsletter, October 2025, Future of Life Institute