There is no single official taxonomy, but most frameworks converge on three types and a set of levels. The types describe scope: narrow AI handles a defined task, general AI handles any intellectual task at human level, and superintelligence exceeds human performance everywhere.
The three types
- Narrow AI: one domain, often superhuman within it, brittle outside it.
- General AI: human-level breadth, capable of transferring across domains.
- Superintelligence: beyond the best humans in every domain, including AI research itself.
The levels
Levels usually grade performance within a type, from emerging to competent to expert to exceptional, and separately grade autonomy, from tool to assistant to agent that acts without supervision. A system can be expert-level and still require constant human direction, or modest and fully autonomous. Both axes matter.
Why the taxonomy is useful
Most public arguments about AGI are really arguments about which cell of the grid counts. Naming the axis first turns a shouting match into a testable claim.