Narrow AI does one job, or a small set of jobs, and it is the AI everyone uses today. AGI, artificial general intelligence, would be about as capable as people across most thinking work. Superintelligence would be far beyond the best humans at almost everything. The first exists; the other two are still ideas, and nobody agrees exactly where the lines fall.
- Narrow AI: one job, done well. Face recognition, chess programs, protein-shape predictors. It exists everywhere.
- AGI: human-level across most mental work. Debated, with no agreed definition or test.
- Superintelligence (ASI): beyond the best humans at nearly everything. Hypothetical.
- Superhuman at one task is still narrow. Range and skill are two different questions.
Think of three kinds of helper. Narrow AI is a pocket calculator: faster than any person at sums, useless at writing a birthday card. AGI would be a capable colleague who can turn their hand to most office jobs. Superintelligence would be a colleague so far ahead that you could not follow their work, the way a cat cannot follow what happens in a laboratory.
| Term | What it means | Examples | Does it exist? |
|---|---|---|---|
| Narrow AI (weak AI) | Means: AI built for one task or a small set of tasks. | Examples: face recognition, chess programs, AlphaFold. | Status: yes, everywhere. |
| AGI (general AI) | Means: AI about as capable as people across most thinking tasks, including learning new ones. | Examples: none agreed. Where today’s chatbots sit is disputed. | Status: debated; no agreed test. |
| Superintelligence (ASI) | Means: AI far smarter than the best humans in practically every field. | Examples: none. | Status: hypothetical; “not yet achieved”. |
What is narrow AI?
The plain point: “narrow” describes how wide the AI’s skills reach, not how good it is.
Stanford’s Institute for Human-Centered AI puts it simply: “Narrow AI is intelligent systems for one particular thing, e.g., speech or facial recognition.”1 IBM says narrow AI “is the only type of AI that exists today. Any other form of AI is theoretical.”2
Narrow AI can still beat every person at its one job. A Google DeepMind paper rates AlphaFold, which predicts the 3D shape of proteins, as “Superhuman Narrow AI”, because it does a single task better than the world’s top scientists.3 More on narrow AI meaning.
What is AGI?
The plain point: AI that could handle most thinking tasks a person can. The trouble is that people mean different things by it.
The same DeepMind paper describes “general” as a “wide range of non-physical tasks, including metacognitive tasks like learning new skills”.4 Others define it by work rather than skills: OpenAI’s charter frames AGI as systems that can do most economically valuable work.5 AGI meaning sets out four competing definitions. There is no agreed test, which is why one headline can announce AGI while another calls it decades away.
What is superintelligence?
The plain point: beyond human, across the board, and so far only on paper.
The philosopher Nick Bostrom gave a widely used definition: an intellect “much smarter than the best human brains in practically every field, including scientific creativity, general wisdom and social skills.”6 IBM calls it “a hypothetical software-based artificial intelligence (AI) system with an intellectual scope beyond human intelligence.”7 The DeepMind paper puts it at the top of its ladder, as general AI that “outperforms 100% of humans”, and marks it “not yet achieved”.8 More on superintelligence meaning.
How do the three fit together?
The plain point: they are steps on a ladder of range. Skill is a separate ladder.
The DeepMind paper sorts AI by two things at once: how well it performs, and how many kinds of task it can do. That is why AlphaFold can be superhuman and narrow at the same time, and why a chatbot can be broad but only middling. Read the three words as a ladder of range:
- Narrow: one job, sometimes done better than any person.
- General: most thinking jobs, about as well as people.
- Super: nearly all thinking jobs, better than anyone.
Ask two questions about any AI: how many kinds of task can it do, and how well? “Superhuman at chess” answers the second and says nothing about the first.
Where do today’s chatbots sit?
The plain point: in the grey zone. Respected sources put them on different sides of the line.
The DeepMind authors placed the chatbots of 2023, such as ChatGPT, at the lowest general level, “Emerging AGI”: “Competent” at some tasks such as short essays and simple coding, “Emerging” at most.9 IBM, by contrast, counts everything that exists today as narrow (source above). Both views fit the facts: chatbots are broad, but not reliable across everything a person can do. This site has found no agreed placement for the models of 2026.
When will AGI or superintelligence arrive?
The plain point: nobody knows. Every date is a forecast, and some forecasters sell AI.
In a large survey of published AI researchers, released in 2024, the combined estimate gave a 10% chance of machines outperforming humans at every task by 2027, and 50% by 2047.10 OpenAI wrote in 2023 that superintelligence “could arrive this decade”.11 Meta said in 2025 that “Developing superintelligence is now in sight.”12 On the other side, a statement hosted by the Future of Life Institute calls for “a prohibition on the development of superintelligence” until it can be shown to be safe.13 These are each organisation’s own statements, not measurements. Is AI dangerous? sets out who worries about what.
Questions people ask
Is ChatGPT narrow AI or AGI?
It depends on whose definition you use. The DeepMind levels paper placed 2023 chatbots at “Emerging AGI”, while IBM counts all existing AI as narrow (sources above).
Does AGI exist yet?
Not by any agreed test, because there is none. AGI meaning explains why the Turing test and benchmarks have not settled it.
Does superintelligence exist?
No. IBM calls it hypothetical, and the DeepMind levels paper lists it as “not yet achieved” (sources above).
Is narrow AI the same as weak AI?
Today, yes: IBM says “Weak AI, also known as narrow AI, focuses on performing a specific task”.14 “Weak” refers to range, not quality.
Where to go next
- “Tell me more about each word.” Read narrow AI meaning, AGI meaning and superintelligence meaning.
- “What other ways are there to sort AI?” See types of AI, explained.
- “What is the singularity?” See singularity meaning.
- “How would anyone keep it safe?” Read AI alignment meaning and is AI dangerous?.
- “Does AI understand anything?” See does AI understand anything?.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. Which of the three exists today?
- Narrow AI
- AGI
- Superintelligence
Show the answer
Answer: Narrow AI. The page says narrow AI is everywhere, while AGI is debated and superintelligence is hypothetical.
2. A program beats every human at predicting protein shapes but does nothing else. What is it?
- Superintelligence, because it is superhuman
- AGI, because it beats scientists
- Narrow AI, because it does one task
Show the answer
Answer: Narrow AI, because it does one task. The page explains that being superhuman at one task is still narrow: range and skill are separate questions.
3. Why do people disagree about whether today’s chatbots are AGI?
- Chatbots have never been tested at all
- There is no agreed definition or test, and chatbots are broad but not reliable at everything
- Every expert agrees they are superintelligent
Show the answer
Answer: There is no agreed definition or test, and chatbots are broad but not reliable at everything. The page says AGI has no agreed test, and sources place chatbots on different sides of the line.
- 1Stanford HAI, Artificial Intelligence Definitions, Sep 2020, read at source 5 Oct 2026 (as cited on this site’s narrow AI page): “Narrow AI is intelligent systems for one particular thing, e.g., speech or facial recognition.”↑
- 2IBM, Types of Artificial Intelligence, undated page, read at source 5 Oct 2026 (as cited on this site’s types of AI page): Artificial Narrow AI “is the only type of AI that exists today. Any other form of AI is theoretical.”↑
- 3Morris et al. (Google DeepMind), Levels of AGI for Operationalizing Progress on the Path to AGI, arXiv:2311.02462, revised 24 Sep 2025, read at source 5 Oct 2026 (as cited on this site’s narrow AI page): “AlphaFold … is a Level 5 Narrow AI (‘Superhuman Narrow AI’) since it performs a single task (predicting a protein’s 3D structure from an amino acid sequence) above the level of the world’s top scientists.”↑
- 4Morris et al., Levels of AGI, arXiv:2311.02462, 2023, read at source 5 Oct 2026 (as cited on this site’s AGI meaning page): the “General” column covers a “wide range of non-physical tasks, including metacognitive tasks like learning new skills”.↑
- 5OpenAI, OpenAI Charter, 2018, as cited on this site’s AGI meaning page (link checked 16 Sep 2026) as the source of the economic definition. Not re-read on 19 Oct 2026the page returned an error to an automated request. ↑
- 6Nick Bostrom, How Long Before Superintelligence?, 1997, revised 1998, read at source 5 Oct 2026 (as cited on this site’s superintelligence page): “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.”↑
- 7IBM, What is artificial superintelligence?, published 14 Dec 2023, read at source 5 Oct 2026 (as cited on this site’s superintelligence page): “Artificial superintelligence (ASI) is a hypothetical software-based artificial intelligence (AI) system with an intellectual scope beyond human intelligence.”↑
- 8Morris et al. (Google DeepMind), Levels of AGI for Operationalizing Progress on the Path to AGI, arXiv:2311.02462, revised 24 Sep 2025, read at source 5 Oct 2026 (as cited on this site’s superintelligence page), Table 1: “Level 5: Superhuman outperforms 100% of humans” … “Artificial Superintelligence (ASI) not yet achieved”.↑
- 9Morris et al., Levels of AGI, arXiv:2311.02462, 2023, read at source 5 Oct 2026 (as cited on this site’s AGI meaning page): “as of this writing in September 2023, frontier language models (e.g., ChatGPT, Bard, Llama2, etc.) exhibit ‘Competent’ performance levels for some tasks (e.g., short essay writing, simple coding), but are still at ‘Emerging’ performance levels for most tasks”.↑
- 10Grace et al., Thousands of AI Authors on the Future of AI, Jan 2024 (revised Oct 2025), read at source 5 Oct 2026 (as cited on this site’s AGI meaning page): “If science continues undisrupted, the chance of unaided machines outperforming humans in every possible task was estimated at 10% by 2027, and 50% by 2047.”↑
- 11OpenAI, Introducing Superalignment, 5 Jul 2023, read at source 5 Oct 2026 (as cited on this site’s superintelligence page): “While superintelligence seems far off now, we believe it could arrive this decade.”↑
- 12Meta, Personal Superintelligence, 30 Jul 2025, read at source 5 Oct 2026 (as cited on this site’s superintelligence page): “Developing superintelligence is now in sight.”↑
- 13Future of Life Institute, Statement on Superintelligence, undated page, read at source 5 Oct 2026 (as cited on this site’s superintelligence page): “We call 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.”↑
- 14IBM, What is strong AI?, published 13 Oct 2021, read at source 5 Oct 2026 (as cited on this site’s narrow AI page): “Weak AI, also known as narrow AI, focuses on performing a specific task, such as answering questions based on user input or playing chess.”↑
Every source on this page is reused from this site’s pages, with the read dates recorded there: Stanford HAI’s AI definitions and IBM’s strong AI page (5 Oct 2026), on narrow AI meaning; IBM’s types of AI page (5 Oct 2026), on types of AI; Morris et al., Levels of AGI, Grace et al., Thousands of AI Authors on the Future of AI (5 Oct 2026) and the OpenAI Charter (link checked 16 Sep 2026), on AGI meaning; Bostrom, IBM’s superintelligence page, OpenAI’s Superalignment post, Meta’s Personal Superintelligence and the Statement on Superintelligence (5 Oct 2026), on superintelligence meaning.
The calculator and colleague comparisons and the ladder of range are this site’s own ways of explaining, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK