Defining AGI: Why the Term Resists a Single Meaning
Artificial General Intelligence is the most consequential idea in modern technology — and the most contested. Researchers disagree not only on when AGI will arrive, but on what it actually is.
Key facts
- There is no single agreed scientific definition of AGI.
- OpenAI's charter defines AGI as 'highly autonomous systems that outperform humans at most economically valuable work.'
- DeepMind's 2023 framework distinguishes capability levels from breadth.
- Most public AGI forecasts assume task-generality, not strong AI or consciousness.
Origins of the Term
The phrase 'Artificial General Intelligence' was popularized in the early 2000s by researchers including Ben Goertzel and Shane Legg to distinguish broad, human-level cognition from the narrow systems that dominated AI at the time. Earlier traditions — from Alan Turing's 1950 imitation game to John McCarthy's 1956 Dartmouth proposal — assumed general intelligence as the implicit goal of the field.
As deep learning produced powerful but narrow systems in the 2010s, 'AGI' became shorthand for the original ambition: a system that could learn anything a human can learn, reason about novel problems, and operate competently across virtually any cognitive domain.
Competing Definitions
There is no consensus definition. Different researchers anchor AGI to different properties — task generality, economic value, autonomy, self-improvement, or human cognitive parity. Each definition implies a different research agenda and a different risk profile.
- Task generality: a system that can perform any cognitive task a human can.
- Economic AGI: a system that can perform most economically valuable work (OpenAI's working definition).
- Self-directed learning: a system that can autonomously acquire new skills without retraining.
- Cognitive parity: matching human performance across standard cognitive benchmarks.
- Strong AI: a system with genuine understanding and, by some accounts, subjective experience.
Levels of AGI
A 2023 DeepMind framework proposed six performance levels (no AI, emerging, competent, expert, virtuoso, superhuman) crossed with two breadth axes (narrow vs general). Under this taxonomy, today's frontier LLMs sit at 'emerging AGI' — broadly competent but unreliable.
The framework matters because it separates capability from autonomy. A system can be highly capable without being agentic, and dangerous autonomy can emerge well below 'expert' performance.
Why the Definition Matters
Definitions shape policy. Funding agencies, safety regulators, and corporate boards rely on capability thresholds to trigger oversight. A vague target produces vague accountability.
Definitions also shape research priorities. If AGI is defined by economic output, the field optimizes for deployment. If it is defined by autonomy or self-improvement, alignment and control take precedence.
Frequently asked
Is AGI the same as 'strong AI'?
+
Not quite. 'Strong AI' (John Searle) implies genuine understanding or consciousness; AGI is usually defined operationally by capability and generality, leaving consciousness aside.
Are current LLMs AGI?
+
Under most definitions, no — they lack consistent reasoning, long-horizon planning, and reliable self-correction. Under DeepMind's framework they qualify as 'emerging AGI'.
Sources & further reading
Continue in this series
Forecasting
AGI Timelines: What Top Researchers Actually Predict
Beyond AGI
Superintelligence: What Comes After Human-Level
Philosophy of Mind
Could AGI Be Conscious — and Would It Matter?
Safety
AI Alignment: The Core Technical Challenge
Risk Analysis
Existential Risks from Advanced AI
Economics
The Economic Impact of AGI