AI that could handle almost any thinking task a person can, not just one or two.
AGI stands for artificial general intelligence. Instead of an AI that only plays chess or only translates, it means one that could learn new tasks by itself and handle nearly anything a person can. Several big AI companies say building it is their goal.
The trouble is that there's no agreed definition. One researcher suggests measuring intelligence by how quickly a system picks up new skills; another group has proposed scoring AI across many areas of human thinking. So some people think AGI is close, others think it's far off, and some say the term is too vague to be useful.
Think of it like this
Think of an all-round athlete: not someone who's only great at football, but someone who can pick up almost any sport quickly. AGI means an all-rounder like that for thinking tasks. But how many sports you need to master to count as all-round is exactly what people argue about.
For example
In 2025 one research team proposed scoring AGI across 10 areas of human thinking, and rated GPT-4 at 27% and GPT-5 at 57%.
Common mix-ups
Mix-upThere's an agreed definition of AGI.
ActuallyResearchers don't agree on what counts as AGI. For example, in 2019 François Chollet treated intelligence as how efficiently you pick up new skills and released ARC, a test where you work out the rule of an unfamiliar coloured-grid puzzle from just a few examples.
Mix-upA figure like "57% of AGI" tells you exactly how far away AGI is.
ActuallyThat score combines ten areas, so one area being almost empty can be hidden inside a single number. In the same study, the biggest gap was long-term memory storage — keeping things remembered over time.
Try it in class
For teachers at the start of a lesson — or for students working on their own.
- Question
- If your class designed an "AGI test", what events would you include?
- Activity
- Borrowing the all-round athlete comparison, each group lists five things an AGI should be able to do and how each would be measured. Pool the lists and mark what overlaps and what doesn't — you'll see why no single yardstick has won. Put the definitions in records E0097 and E0728 side by side and compare them in a line each.
Records about this
Source-checked moments from AI history.
The variance explained by the first principal component across benchmarks fell from 92% in 2023–24 to 64% after reasoning models arrived — an argument for specialization beneath apparent generality
Moves the definition of AGI from rhetoric to a psychometric scale; the biggest deficit is long-term memory storage
Provided the yardstick for the later AGI debate