By the WahLiao desk · Last verified 1 October 2026
Artificial general intelligence (AGI) means a machine that can do most thinking tasks at least as well as a capable adult, across many fields rather than one. Artificial superintelligence (ASI) goes further: a hypothetical system that is far better than the best humans at almost everything. Neither has an agreed test, and as of 1 October 2026 there is no consensus that either exists. Super Intelligence, the kind of AI behind ChatGPT, Claude and Gemini, is very capable but uneven: Google DeepMind’s researchers placed chatbots of this type at the lowest rung of their own AGI ladder, “Emerging AGI”, and forecasts for the real thing run from “within a year or two” to “not with today’s methods at all”.
The most useful rule: when someone says AGI, ask which definition they mean and who benefits from the answer. The word is used by labs, investors and critics in quite different ways.
AGI and superintelligence: quick facts
| AGI | Human-level ability across a broad range of thinking tasks; no single agreed definition |
| ASI | Hypothetical system far beyond the best humans in virtually all domains |
| OpenAI’s wording | “Highly autonomous systems that outperform humans at most economically valuable work” (OpenAI Charter) |
| Google DeepMind’s paper | Six levels from “No AI” to “Superhuman”; 2023-era chatbots rated Level 1, “Emerging AGI” |
| Anthropic’s term | “Powerful AI”, a “country of geniuses in a datacenter” |
| Researcher survey | 2,778 AI researchers: 50% chance machines beat humans at every task by 2047 |
| Exists today? | No consensus that AGI does; nobody credible claims ASI does |
The WahLiao Verdict
| Is it AGI? | Depends on the definition. By most, not yet. |
| Is it ASI? | No. Nobody serious says so. |
| Timelines | Opinions, not facts. Wide range, strong incentives. |
| For your wallet | Buy tools for what they do today, not what AGI might do. |
| Our house name | A label for everyday tools, not a claim. |
Narrow AI, AGI and ASI in one table
| Type | What it means | Example | Exists today? |
|---|---|---|---|
| Narrow AI | Very good at one clearly scoped task | Spam filters, face unlock, chess engines, map route planning | Yes, everywhere |
| General-purpose chatbots | Broad but uneven: strong at language, code and analysis, weak at others, still makes confident mistakes | ChatGPT, Claude, Gemini | Yes. Whether this counts as early AGI is disputed |
| AGI | At least skilled-adult level across most thinking tasks, able to learn new ones | None agreed | No consensus that it does |
| ASI | Far beyond the best humans in virtually all domains | None; a hypothetical | No |
What the labs mean by AGI
OpenAI. Its Charter defines AGI as “highly autonomous systems that outperform humans at most economically valuable work”. That is an economic test: can the system do the jobs? The definition has had commercial weight. When OpenAI and Microsoft revised their partnership in October 2025, they agreed that any AGI declaration by OpenAI would be verified by an independent expert panel. A further revision in April 2026 set OpenAI’s revenue-share payments to Microsoft to continue through 2030 “independent of OpenAI’s technology progress”, taking the question out of the money.
Google DeepMind. A 2023 paper by DeepMind researchers, including co-founder Shane Legg, proposed “Levels of AGI”, modelled on the levels used for self-driving cars. It scores systems on two axes: how general they are, and how well they perform. The performance levels run from Emerging (as good as or slightly better than an unskilled person) through Competent (at least 50th percentile of skilled adults), Expert (90th), Virtuoso (99th) to Superhuman (better than all humans). The authors placed the chatbots of the day, including ChatGPT and Gemini, at Level 1, “Emerging AGI”, and defined ASI as Level 5 across a wide range of tasks.
Anthropic. Anthropic, which makes Claude, generally uses its own term, “powerful AI”. Its chief executive, Dario Amodei, describes a model smarter than a Nobel Prize winner across most fields, able to work on its own for days or weeks, and run as millions of copies at once: “a country of geniuses in a datacenter”. In its 2025 submission to the US government’s AI Action Plan, Anthropic said it expected such systems to emerge in late 2026 or early 2027. We are now inside that window, and whether it has arrived is exactly the kind of claim to check against evidence rather than announcements.
What leading figures have forecast
These are attributed opinions, not facts.
| Who | What they have said | When |
|---|---|---|
| Sam Altman, OpenAI chief executive | “We are past the event horizon; the takeoff has started”, and “humanity is close to building digital superintelligence”; expected systems that can find novel insights in 2026 and robots doing real-world tasks by 2027 | Essay “The Gentle Singularity”, June 2025 |
| Dario Amodei, Anthropic chief executive | Powerful AI could be “as little as 1–2 years away”, though it could be “considerably further out” | Essay “The Adolescence of Technology”, January 2026 |
| Demis Hassabis, Google DeepMind chief executive | AGI could arrive around 2030, but large language models alone will not be enough; continual learning, long-term reasoning and memory are still missing | Y Combinator podcast interview, reported May 2026 |
| Yann LeCun, NYU professor and executive chairman of AMI Labs | The bet that language models will reach human-level intelligence is “complete BS”; at best, within five years we might be “on a good path” towards it, and the task is “almost certainly much harder than we think” | Brown University lecture, April 2026 |
| 2,778 published AI researchers (survey) | 10% chance that unaided machines outperform humans at every task by 2027, 50% by 2047 | Survey published 2024 |
Two things stand out. First, the heads of the biggest labs cluster at “a few years”, while the broad research community is far more cautious. Second, the sceptics are not saying progress is fake. LeCun, now executive chairman of the start-up AMI Labs, argues that systems trained mostly on text do not understand the physical world and that different designs, such as “world models”, are needed. For the history behind these camps, see A short history of Super Intelligence, and for the arguments about risk, Safety and risk: the debate, plainly.
Is today’s Super Intelligence AGI?
There is no consensus, and the disagreement is mostly about definitions. By DeepMind’s ladder, today’s chatbots already count as the lowest level of general AI. By OpenAI’s economic test, they plainly do not yet outperform humans at most paid work. By the everyday meaning, “can do what a capable colleague can do”, the honest answer is: in some tasks yes, in many tasks no.
The ability is jagged. A chatbot can draft a contract clause in seconds and then miscount the letters in a word, or state a wrong fact with total confidence. It does not remember and learn from last week’s conversation the way a colleague would, and it struggles with long, multi-week projects without a person steering. Those gaps, which the labs themselves name, are why most researchers do not call today’s tools AGI. For the mechanics behind confident mistakes, see How Super Intelligence works, and for the common misreadings, Myths vs facts.
Why WahLiao says “Super Intelligence”
A clear note on our name. On WahLiao, “Super Intelligence” (two words, capitalised) is our house name for the everyday AI tools you can use today: ChatGPT, Claude, Gemini, Copilot and the rest. It is not a claim that artificial superintelligence (ASI), the hypothetical system far beyond human ability discussed on this page, exists. When we mean that technical idea, we write it in lower case, “superintelligence”, or spell out ASI. The reasoning behind the name is on Why WahLiao says Super Intelligence.
What it means for you in Singapore
AGI talk should not move your spending. Pay for a tool because of what it does well for you this month: the plans and S$ prices are on Free vs paid plans, and eligible citizens can get six months of a premium tool free after a SkillsFuture AI course, now run by the Skills and Workforce Development Agency (SWDA), explained on the SkillsFuture offer page. If you are thinking about your job, Budget 2026 and your job is more useful than any AGI forecast. For what the next few years may bring, with the hype separated out, read The future to 2030.
AGI and superintelligence: FAQ
What is the difference between AI and AGI?
AI is the broad field, including narrow tools that do one job well. AGI is a specific goal within it: a system that matches a capable adult across most thinking tasks, not just one. There is no single agreed definition or test.
Has AGI been achieved yet?
There is no consensus that it has, as of 1 October 2026. Google DeepMind researchers call current chatbots “Emerging AGI”, the lowest level on their scale, while most researchers point to gaps such as continual learning and long-term planning.
What is artificial superintelligence (ASI)?
A hypothetical system that greatly exceeds the best humans in virtually all domains. Philosopher Nick Bostrom’s 2014 definition is “any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest”. It does not exist.
When will AGI arrive?
Nobody knows. Lab leaders have suggested anywhere from one or two years to around 2030; a survey of 2,778 AI researchers put a 50% chance on machines beating humans at every task by 2047; sceptics such as Yann LeCun say current methods will not get there.
Is ChatGPT AGI?
By most definitions, not yet. It is broad and often impressive but uneven, makes confident mistakes and does not learn continuously from experience. By DeepMind’s ladder it sits at the lowest level of general AI.
Read next
This page belongs to Super Intelligence. Next, read A short history of Super Intelligence.
Sources checked 1 October 2026: OpenAI, Charter; OpenAI, next chapter of the Microsoft partnership (October 2025); OpenAI, next phase of the Microsoft partnership (April 2026); Morris et al., Google DeepMind, Levels of AGI; Anthropic, recommendations for the US AI Action Plan; Dario Amodei, The Adolescence of Technology; Sam Altman, The Gentle Singularity; Futura Sciences, Hassabis on AGI around 2030; Brown University, Yann LeCun lecture; Grace et al., Thousands of AI Authors on the Future of AI; Benthall, citing Bostrom’s definition of superintelligence. General information. Last updated 1 October 2026.
