← AI Terminology
AGI - Artificial General Intelligence
AGI stands for Artificial General Intelligence: a hypothetical AI system that can perform any intellectual task a human can, with comparable flexibility, reasoning, and generalisation across domains.
No system today is widely accepted as AGI — the term describes a threshold, not a product.
No system today is widely accepted as AGI — the term describes a threshold, not a product.
Why It Matters in AI
AGI is the north star that motivates much of frontier AI research — OpenAI, DeepMind, and Anthropic all name AGI (or transformative AI) as their explicit mission. The question of when (or whether) AGI arrives has enormous implications for economics, security, governance, and society. Defining it precisely is itself contested: different researchers draw the line at different capability thresholds.
Key Points
| Aspect | Description |
|---|---|
| Key debates | Is LLM scale a path to AGI? Does AGI require embodiment, world models, or genuine reasoning? |
| Turing Test | Early AGI proxy — can a human conversing with the system tell it's not human? Now considered insufficient |
| Company stances | OpenAI: "highly autonomous systems that outperform humans at most economically valuable work" |
| Safety relevance | An AGI powerful enough to act in the world without alignment would be an existential risk |
| Contrast with ANI | Artificial Narrow Intelligence (ANI) — today's AI, excels at specific tasks but can't generalise |
| Timeline estimates | Range from 2-5 years (optimists) to never (skeptics) — no consensus |
Simple Analogy
Current AI is like a set of world-class specialists — a chess grandmaster, a radiologist, a translator — each unbeatable in their domain but helpless outside it. AGI would be the equivalent of one person who is a competitive chess player, radiologist, translator, programmer, and lawyer — and can learn any new skill as needed.
Common Usage Examples
- OpenAI's mission statement: "ensure that AGI benefits all of humanity"
- DeepMind's "AlphaFold" and "Gemini" as steps toward general reasoning
- AGI benchmarks proposed: ARC prize (abstract reasoning corpus), GAIA, Humanity's Last Exam
- AI safety orgs (MIRI, ARC Evals, Anthropic's ASL policy) focus on AGI risk mitigation
- Sam Altman's 2024 essay "The Intelligence Age" framing AGI as imminent
Summary
In short: AGI is the hypothetical point where AI stops being a narrow tool and becomes a general-purpose mind — the most consequential milestone the field is working toward.