B1 · Introduction to Artificial Intelligence

Narrow, General and Super AI

By capability, AI is divided into three levels. Narrow AI (ANI) performs specific tasks. General AI (AGI) would match people across nearly all thinking tasks. Super AI (ASI) would far exceed people in almost everything. Only narrow AI exists today.

10 min Beginner Jump to the activity

What you need: Nothing. This page starts from the beginning, with no programming and no formulas.

Definition

By capability, AI is divided into three levels. Narrow AI (ANI) performs specific tasks. General AI (AGI) would match people across nearly all thinking tasks. Super AI (ASI) would far exceed people in almost everything. Only narrow AI exists today.

In simple words

Narrow AI is a specialist doctor who knows only eyes. AGI would be a person who can learn any job. ASI would be something much smarter than any person at every job. ChatGPT can talk about nearly anything, but it's still counted as narrow by most experts, though some debate this.

Why it matters

News stories often blur these levels, so real products get mistaken for science fiction, and the other way round. Knowing the levels helps you judge claims, and it frames the safety debate.

Key terms

Narrow AI (ANI)
AI built to do one task or a small group of related tasks. All AI in use today is narrow.
Artificial general intelligence (AGI)
A hypothetical AI that could learn and perform almost any thinking task a person can, at a similar level.
Artificial superintelligence (ASI)
A hypothetical AI far more capable than the best humans in almost every area.
Capability
What a system is able to do, and how well.
Hypothetical
Imagined or proposed, but not yet shown to exist.

Explained in detail

Narrow AI: the AI we actually have

Every AI system in use today is narrow. A face-unlock model recognises faces and nothing else. A crop-disease model names diseases from photos of leaves but cannot tell you tomorrow's weather. Narrow does not mean weak: narrow systems can be superhuman at their task. AlphaFold, a narrow system, predicted the 3D shapes of more than 200 million proteins, work that won its creators a share of the 2024 Nobel Prize in Chemistry. But move a narrow system outside its task and it has nothing to offer.

Where do chatbots fit?

Modern chatbots can write, translate, summarise, explain and answer questions on almost any subject, so they are far more general than older AI. This has made the line between narrow and general AI harder to draw, and experts disagree about where today's systems stand. Most researchers still do not call them AGI, because they make basic reasoning errors, can state false things confidently, cannot learn continuously from experience the way people do, and have limited ability to act in the physical world. It is fair to say they are broader narrow AI, not yet general intelligence.

General AI: a goal and a debate

AGI would be a system that can learn any new intellectual task about as well as a person, transfer knowledge between areas, and handle unfamiliar problems. Several large companies say building it is their goal. Experts disagree strongly on whether it will arrive in a few years, in decades, or perhaps never, and even on exactly how to recognise it. Because predictions in AI have often been wrong in both directions, the honest answer to "when will AGI arrive?" is that nobody knows.

Superintelligence: a thought experiment

Artificial superintelligence describes AI that would far surpass the best human minds in science, strategy, creativity and social skill. It exists only as an idea, discussed by philosophers such as Nick Bostrom in his 2014 book "Superintelligence" and by researchers who study AI safety. The idea is taken seriously because, if it were ever built, making sure its goals matched human values would be extremely important. You will meet this topic again in the advanced level.

How it works

LevelCan doExists?Examples
Narrow (ANI)Specific tasks, sometimes better than peopleYes, everywhereFace unlock, chess engines, spam filters, LLMs
General (AGI)Learn and do nearly any thinking task a person canNo. Whether we're close is debated.None yet
Super (ASI)Far beyond people at almost everythingNo, only a hypothesisNone

Try it yourself

Three levels of AI, and where we are

Move along the line, and stop where the world is today.

Classify well-known AI systems

Put each system in narrow, general or super AI.

  • A chess engine that beats world champions

  • A chatbot that writes, translates and explains

  • A voice assistant

  • A robot that can learn any job a person can, as well as a person

  • A machine far smarter than all humans in every area

  • A self-driving car

Common mistakes

  • Assuming a system that's great at one thing is great at everything.
  • Calling every new chatbot AGI. The word has no agreed test yet, so ask how it's being used.

Check yourself

Question 1

1. Which level does all commercial AI belong to today, by the common view?

Question 2

2. What would make a system AGI?

Summary

  • Three levels: narrow, general, super.
  • Only narrow AI exists today.
  • AGI timelines are uncertain and debated.

Finished reading?

Sources: Russell & Norvig, Artificial Intelligence: A Modern Approach