For years, “AGI” has been tech’s favourite moving goalpost. Every time AI gets better at something, we shift the definition. Can’t be AGI if a computer does it, right?
A new paper from researchers including Dan Hendrycks, Dawn Song, and Yoshua Bengio changes this. They’ve created the first systematic definition of Artificial General Intelligence, grounded in a century of cognitive science. And the results are uncomfortable.
The Framework
The researchers used Cattell-Horn-Carroll theory, the most empirically validated model of human intelligence, to break down cognition into 10 measurable domains:
General Knowledge
Reading and Writing
Mathematical Ability
On-the-Spot Reasoning
Working Memory
Long-Term Memory Storage
Long-Term Memory Retrieval
Visual Processing
Auditory Processing
Speed
Each domain gets 10% weighting. Hit 100%, you’ve matched a well-educated adult’s cognitive versatility and proficiency. That’s AGI.
