AGI

/eɪ dʒiː aɪ/ · Abbreviation · AI & Machine Learning

Definitions

  1. Artificial General Intelligence — a hypothetical AI system that matches or exceeds human cognitive abilities across all domains, not just narrow tasks. Whether we're 5 years or 50 years away (or if it's even possible) is the most heated debate in AI. Definition varies by who you ask and how much funding they need.

    In plain English: An AI that could think, learn, and solve problems as well as a human across any subject — the ultimate goal (or fear, depending on who you ask) of AI research.

  2. In corporate strategy and fundraising, AGI functions as a rhetorical device — a goalpost that justifies massive investment and talent acquisition. The definition conveniently shifts to stay just beyond whatever has been achieved.

    Example: 'Every AI lab claims they're building toward AGI. It's the perfect fundraising narrative because nobody agrees on what it means or how to measure it.'

    Source: industry critique

Etymology

1997
Mark Gubrud uses 'Artificial General Intelligence' in a discussion of military AI, distinguishing it from narrow task-specific AI
2007
Ben Goertzel and Cassio Pennachin publish 'Artificial General Intelligence,' establishing AGI as a research field
2023
OpenAI, DeepMind, and Anthropic all name AGI as their goal, making it the most debated and contested concept in tech

Origin Story

The holy grail of AI that nobody can quite define

**Artificial General Intelligence** was coined by Shane Legg and Ben Goertzel around 2002-2007 to distinguish human-level AI from the narrow AI that actually exists. Narrow AI beats humans at chess, Go, and image recognition; AGI would match human cognitive ability across all domains.

The concept existed long before the name. The field of AI was founded in 1956 at the Dartmouth Conference with the assumption that 'every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.' That turned out to be wildly optimistic.

The rise of large language models reignited AGI debates. Some researchers (like Ilya Sutskever) believe AGI is imminent; others (like Yann LeCun) argue current approaches can't get there. The definition remains contested -- which is convenient for anyone claiming to build it.

Coined by: Shane Legg, Ben Goertzel (term); Dartmouth Conference (concept)

Context: Early 2000s (term), 1956 (concept)

Fun fact: Ben Goertzel, who co-coined 'AGI,' also leads the SingularityNET project. He has predicted AGI by various dates, all of which have passed. The AI field has a long tradition of optimistic timelines -- Marvin Minsky predicted human-level AI within 'a generation' in 1967.

Related Terms