Definition
An AI can beat world champions at chess but cannot tie a shoelace. Another writes essays in seconds but cannot be trusted to count the letters in a word. What would it take for an AI to be good at everything a person can do?
Artificial general intelligence, or AGI, is a hypothetical AI that could understand, learn and perform a very wide range of intellectual tasks at least as well as a typical human, and adapt to new kinds of problems it was not specifically trained for. Today's AI systems are called narrow because each is strong in some areas and weak or useless in others. There is no agreed definition of AGI, and experts disagree about whether and when it will be built.
In simple words
Today's AIs are like specialists: a brilliant chess player, a fast translator, a good painter, each in its own room. AGI would be like a person who can walk into any room, learn what is needed and do the job well, including jobs nobody has seen before.
Why it matters
AGI is discussed in news, by companies and by governments, often with big claims. Understanding what it means, what today's AI can and cannot do, and why experts disagree helps you think clearly instead of being swept along by hype or fear.
Key terms
- Narrow AI
- AI that performs well on specific tasks but cannot generalise to very different ones.
- Artificial general intelligence (AGI)
- A hypothetical AI with broad, flexible abilities matching or exceeding a typical human across most intellectual tasks.
- Superintelligence
- A hypothetical AI far more capable than the best humans in almost every area.
- Generalisation
- Applying what was learned to new, different situations.
- Benchmark
- A standard set of tests used to compare AI systems.
- Hype
- Exaggerated excitement or claims, often for attention or money.
Explained in detail
Narrow AI today
Every AI system in use today is narrow in some sense. A chess engine cannot drive a car. A face-recognition system cannot write a poem. Large language models are much broader than earlier systems: one model can write, translate, summarise, explain science and help with maths. But they still make strange mistakes that a child would not, invent facts with confidence, struggle with some simple reasoning and physical common sense, and cannot act in the physical world without other systems. They also learn mainly during training, not continuously from experience as people do.
What AGI would need
| Ability | What it means | Where today's AI stands |
|---|---|---|
| Broad knowledge | Know about many fields | Strong in language models |
| Reliable reasoning | Think step by step without silly errors | Improving, but still inconsistent |
| Learning from little | Learn a new skill from a few examples, as people do | Limited |
| Continual learning | Keep learning from experience without forgetting | Mostly absent |
| Acting in the world | Plan and carry out long tasks, including physical ones | Early stages, with robots and software agents |
Why experts disagree
Some researchers and company leaders believe AGI could arrive within years, pointing to rapid progress in language models. Others think it is decades away or may need entirely new ideas, pointing to the persistent weaknesses above. Some argue the term is too vague to be useful, because intelligence has many dimensions. Part of the disagreement is about definitions: if AGI means passing many exams, some systems are close; if it means doing almost any job a person can do, reliably and safely, in the real world, it is much further. Be cautious of confident predictions in either direction, especially from those who stand to gain from them.
Why it matters
If AGI were built, it could speed up science and medicine enormously, but it would also raise serious questions: who controls it, how to make sure it follows human values, what happens to jobs, and how to prevent misuse. Many researchers work on AI safety, trying to make sure that more powerful systems remain honest, controllable and beneficial. Governments have begun holding international meetings on these risks. These questions do not need to be settled by experts alone; they affect everyone, and informed citizens should take part in the discussion.
How it works
- When you hear an AI described as "general" or "human-level", ask: at which tasks?
- Look for what it still gets wrong, not only what it gets right.
- Check who is making the claim and whether they benefit from it.
- Compare views from researchers who disagree.
- Think about both the possible benefits and the risks.
Try it yourself
Specialists and a generalist
Imagine a school fair with three AI helpers and one human volunteer.
Narrow AI or AGI?
Which description fits each system?
A spam filter
A chess engine
A system that recognises spoken commands for a smart speaker
A system that could learn any new job as well as a skilled person, from a little training
A machine that could do scientific research, run a business and care for patients equally well
Common mistakes
- Believing AGI already exists because a chatbot sounds human.
- Treating confident predictions about AGI as facts.
- Thinking AGI questions are only for experts.
Check yourself
Summary
- Today's AI is narrow: strong in some areas, weak in others.
- AGI is a hypothetical AI with broad, flexible, human-level ability; its definition and timing are debated.
- Treat claims carefully, and take part in discussing its benefits and risks.
Finished reading?
Sources: OECD AI Policy Observatory · IndiaAI (Government of India)