Characteristics of Artificial General Intelligence (AGI)
Artificial General Intelligence, or AGI, refers to a hypothetical AI system that can understand, learn, and apply knowledge across a wide range of tasks at a level comparable to, or beyond, human ability, rather than excelling at one narrow task. No system today is broadly agreed to have achieved this. The characteristics below describe what researchers generally mean when they define the concept, in contrast with the narrow AI systems that exist today.
Last updated: 8 September 2026
Key characteristics
Generality across domains
AGI would be able to apply its intelligence to a wide, open-ended range of tasks and domains, rather than being built and trained for one specific job such as image recognition or translation.
Transfer learning
It would be able to take knowledge or a skill learned in one context and apply it effectively to a new, previously unseen context, much as a human who learns to ride a bicycle can adapt that balance and coordination to riding a motorcycle.
Autonomous reasoning
AGI is expected to reason through novel problems it was not explicitly trained on, forming and testing its own hypotheses rather than simply pattern-matching against training data.
Self-directed learning
Rather than requiring a human to curate every training dataset, AGI would be able to identify what it needs to learn next and acquire that knowledge with far less direct supervision.
Adaptability to novel situations
It would perform reasonably well in situations that differ meaningfully from anything in its prior experience, rather than failing unpredictably outside its original training distribution.
Common-sense understanding
AGI is expected to grasp everyday practical knowledge about how the physical and social world works, the kind of implicit understanding humans rarely need to state explicitly but that current narrow AI systems often lack.
Long-term planning and goal pursuit
It would be able to set and work toward goals over an extended sequence of steps, adjusting its plan as circumstances change, rather than only producing a single-step response to a single prompt.
Frequently asked questions
What is the difference between AGI and the AI systems available today?
Today's widely used AI systems, including large language models, are generally described as narrow or specialised AI: extremely capable within the tasks they were trained for, but without the reliable, human-level generality, autonomous goal-setting, and transfer learning that define AGI. Whether current systems represent meaningful steps toward AGI, or a fundamentally different kind of capability, is actively debated among researchers.
Has AGI already been achieved?
As of this writing, there is no broad consensus among AI researchers that AGI has been achieved. Definitions of AGI vary, and some argue certain modern systems already show signs of general capability, while others maintain that key characteristics, such as robust autonomous reasoning and true transfer learning, have not yet been convincingly demonstrated.
Why does the definition of AGI matter?
Because AGI is not a single agreed technical benchmark, how it is defined significantly affects claims about timelines, safety risks, and regulation. Different research organisations and governments use somewhat different working definitions of the term.