What Is AGI?

There is no universal definition — here's how each organization defines the finish line

OpenAI

— Sam Altman

Systems that outperform humans at most economically useful tasks.

Key Metric: Economic value — can AI replace human labor at scale?

Timeline: Late 2026 - early 2027

Source

Google DeepMind

— Demis Hassabis & Shane Legg

A system capable of exhibiting all the cognitive capabilities humans can, with versatility across domains.

Key Metric: Cognitive versatility — can it learn new skills with scarce data and surpass humans at useful work?

Timeline: 50% chance by 2028-2030

Source

Anthropic

— Dario Amodei

Avoids the term 'AGI' — calls it 'a marketing term.' Describes it as 'a country of geniuses in the data center' or systems smarter than a Nobel Prize winner in most subjects.

Key Metric: Expert-level performance — Nobel laureate capability across many fields

Timeline: Late 2026 - early 2027

Source

NVIDIA

— Jensen Huang

Already achieved — if AI can pass general tests like bar exams, medical exams, and CPA exams, that constitutes AGI.

Key Metric: Standardized testing — passing professional exams across domains

Timeline: Already here (March 2026)

Source

AMI Labs (Yann LeCun)

— Yann LeCun

Rejects 'AGI' entirely — 'There is no such thing as general intelligence.' Prefers 'human-level AI' and believes it requires world models, not LLMs.

Key Metric: World understanding — can it build internal models of reality?

Timeline: Decades away with current approaches

Source

Academic Consensus

— 2,778 AI Researchers (Survey)

High-Level Machine Intelligence (HLMI) — when unaided machines can accomplish every task better and more cheaply than human workers.

Key Metric: Full economic substitution across all tasks

Timeline: 50% probability by 2040

Source

Why Definitions Matter

When Jensen Huang says “we've achieved AGI” and Yann LeCun says “AGI makes no sense,” they're not necessarily disagreeing about AI capabilities — they're using different definitions. The lack of consensus means AGI milestones depend entirely on who you ask and what metric they use. This makes the “when will AGI arrive?” question fundamentally a question about definitions, not just technology.