Jargon:Artificial General Intelligence (AGI)
Say it: AY-jee-eye · 1 min read
Say this at the party
“An AI that can learn and do almost any thinking job a person can, not just one narrow task. Nobody agrees exactly where the line is.”
In plain English
Today's AI can be superhuman at some things and oddly bad at others. AGI is the idea of an AI that is good across the board: it can pick up a new kind of job and get on with it, like a capable person can.
The catch is that there is no single test for it. Some people say we are nearly there. Others say we are decades away. Often they are using different definitions.
Where it came from
Mark Gubrud used the phrase in 1997. Shane Legg, who later co-founded DeepMind, and Ben Goertzel brought it back into use around 2002, to describe AI that is general rather than built for one task.
See it in action
OpenAI defines AGI as AI that outperforms humans at most economically valuable work. Google DeepMind proposed "levels of AGI", like the levels used for self-driving cars, so progress can be measured in steps.
What people get wrong
- AGI doesn't mean the AI is conscious or has feelings. Most researchers treat that as a separate question.
- There won't be one day when AGI is switched on. It is more likely to be argued about for years after it arrives.
Go deeper
- PaperMorris et al. (Google DeepMind), 2023Levels of AGI for Operationalizing Progress on the Path to AGIA sensible way to measure AGI in levels instead of yes or no.
- PaperBubeck et al. (Microsoft), 2023Sparks of Artificial General Intelligence: Early experiments with GPT-4The famous, and much argued about, claim that GPT-4 showed early signs.
- ArticleWikipediaArtificial general intelligenceA good map of the history and the arguments.
Related jargon
Drafted with AI, checked by the AI Council (Claude, ChatGPT and Gemini) and signed off by Phin.