AGI (Artificial General Intelligence)

A hypothetical AI system that matches or exceeds human capability across the full range of cognitive tasks — not just narrow domains. There's no agreed definition or test.

AGI — Artificial General Intelligence — refers to AI that matches or exceeds human cognitive ability across essentially the full range of tasks, not just one narrow domain. The contrast is with "narrow AI" like AlphaGo (superhuman at Go, useless at anything else) or current LLMs (impressive but uneven, can fail at things a child can do). It matters because AGI is the stated long-term goal of OpenAI, Anthropic, and DeepMind, and it's what investors, regulators, and AI labs talk about when they discuss "transformative AI" or existential risk. Whether AGI arrives in 5 years or 50 affects strategy, alignment research priority, and policy. The debate: there's no widely-accepted definition or test. Sam Altman has said GPT-4 is "clearly not" AGI; François Chollet's ARC challenge is one attempt at a quantitative AGI test. Some researchers think current LLM scaling will get there; others think a fundamentally different architecture is needed. A practical observation: current frontier models are highly capable in many domains and oddly weak in others (counting letters, novel reasoning, persistent agentic work). The gap between "impressive demo" and "reliable across all tasks like a human professional" is the AGI gap. Related: ASI, scaling laws, alignment, frontier model.