Topics / What is AGI
What is an AGI (artificial general intelligence)?
The problem as a graph
The graph contrasts narrow AI with an AGI. On the left a cluster whose edges all stay inside one domain — many nodes, but no bridge outward. On the right a central AGI node with active edges to several domain clusters: language, images, logic, everyday life. It is exactly these cross-level edges that make the difference visible.
Graph as text
- Narrow AI (one domain) → Relation within the domain (passive)
- Relation within the domain → Language domain (passive)
- AGI (connects all levels) → Language domain (active)
- AGI (connects all levels) → Image domain (active)
- AGI (connects all levels) → Logic domain (active)
- AGI (connects all levels) → Everyday domain (empty)
Step by step
- First sort the domains as their own network levels: language, images, logic, everyday life. Each level groups entities of one kind — this reveals where an AI is strong and where it is not.
- For a current system, check in which single level its relations are dense. A language system links words to words, an image system pixels to patterns — both stay inside their own field.
- Look for bridge relations that lead out of that level. If they are missing, the system cannot carry an insight from one domain into another — that is the limit of narrow AI.
- Now picture the same network whose relations reach every level: an entity from language can connect directly with one from images or logic. These cross-level relations are exactly what marks an AGI.
- Measure generality not by the number of entities but by the reach of the relations. Many nodes in one field do not make an AGI — what counts is that the relations cross the domain boundary.
- Note which new connections a system can make across domains, and you have a practical gauge for how close it sits to general rather than narrow intelligence.
Seen through the model
Picture a current language system. In the level of language its network is dense: it links words, sentences and meanings with remarkable confidence. But try to show it an image or have it act in the world, and the network ends at the domain boundary. The relations do not reach across. That is exactly what 'narrow' means: many active relations, but all within the same single level.
An AGI would be the same principle, just without that boundary. An entity from language — say the concept 'staircase' — could connect directly with an entity from images, from logic and from everyday life: recognise it, describe it, plan with it, handle it. The generality lies not in suddenly having more knowledge, but in the relations reaching every network level. The network no longer stops at the edge of a domain.
This also makes clear why 'general' does not mean 'bigger'. You could feed a language system endless extra words and still have narrow AI — many nodes in a single field. An AGI only emerges when a new kind of relation is added: the cross-level one, able to connect any entity with any other, no matter which domain it comes from.
This is one way to see AGI — not a finished truth but a lens: not a 'mind' awakening, but a relational network losing a structural boundary. How such a network grows ever more capable, and whether it is bound to arrive, are separate questions — they lead onward to superintelligence and the question of time.
Frequently asked
What is the difference between narrow AI and an AGI?
Narrow AI is a relational network that is strong in just one domain, say in language alone or images alone. Its relations stay within that single network level and do not reach beyond it. An AGI is a network whose relations reach across all levels and can connect any entities from any domains. The difference lies not in the amount of knowledge, but in the reach of the relations. It is exactly these cross-level relations that turn narrow intelligence into general intelligence.
Does an AGI already exist?
No, an AGI in the full sense does not yet exist; it is a theoretical concept. All systems in common use today are narrow AI — networks densely linked within one or a few domains, but whose relations end at the domain boundary. Some newer systems already connect several domains such as language and images, feeling their way toward more generality. As long as their relations cannot freely reach every network level, however, they stay short of a true AGI. The term describes a goal, not a state already reached.
Does AGI mean the machine has consciousness?
Seen through this model, AGI first means only a structural property: relations that reach across all network levels. That is a statement about the reach of the connections, not about a mind or a will. General capability and consciousness are therefore separate questions that should be kept cleanly apart. Whether and how something like self-reference arises from such a network can be considered separately, but it does not change the definition. AGI here means cross-level generality, not necessarily an inner life of its own.
Keep thinking
Related terms: Entity, Relation, Network level, Zoom in / zoom out, The three states: empty, active, passive