What is Artificial general intelligence?

How it works (as a definition)

There is no agreed technical definition of AGI, and no test that everyone accepts. That is the most important thing to know about the term. Common definitions include:

• Human-level across tasks: an AI that can do most cognitive work a person can, and learn new tasks without being rebuilt.
• Economic: OpenAI's charter describes AGI as "highly autonomous systems that outperform humans at most economically valuable work".
• Levels rather than a line: a 2023 Google DeepMind paper (Morris et al.) proposes levels of AGI based on both performance (from "emerging" to "superhuman") and generality (narrow to general), and argues for judging capabilities rather than how a system works internally. It treats autonomy as a separate dimension from capability.

Today's systems are narrow AI in the strict sense, though large language models and agents are much more general than earlier AI: one model can write, code, translate and reason across many subjects. Whether scaling these systems leads to AGI, or whether new ideas are needed, is an open and actively argued question.

Why it matters

• It drives strategy and investment. Several leading labs state AGI as their goal, and their claims about how close it is shape funding, hiring and policy.
• It shapes safety and regulation debates. Concerns about very capable systems acting autonomously are framed around AGI and superintelligence.
• It is easy to misuse. Because the definition is loose, "AGI" in a headline can mean anything from a strong benchmark result to a marketing claim. Ask which definition is meant and what evidence supports it.

For practical work today, the more useful questions are what a specific model can do reliably, at what cost, and with what oversight.

Example

A model that scores well on a maths olympiad benchmark but cannot reliably book a trip end to end, or learn a new job from a week of on-the-job feedback, is impressive narrow-to-general AI. Under the levels framework it might rank high on performance in some areas and lower on generality, which is a more informative description than "AGI" or "not AGI".