Foundations
Narrow AI
Also known as: weak AI, artificial narrow intelligence
AI that is built for specific tasks and has no general understanding; this describes essentially all AI in use today.
Narrow AI is AI that is good at specific things but has no broad, general understanding, and that is all the AI that exists today. A chess engine, a spam filter, even a powerful chatbot: each is impressive within its scope but does not truly understand the world the way a person does. "Narrow" is the honest label for today's AI, in contrast to the hypothetical, general AGI that does not yet exist.
Narrow AI (also "weak AI") refers to systems designed and trained for particular tasks or domains, without general intelligence. It is contrasted with artificial general intelligence, which would be broadly capable across any task.
A nuance about modern models
Today's LLMs can do a wide range of tasks, which makes them feel general. But they are still narrow in the important sense: they lack robust general understanding, reliable reasoning across all domains, and genuine autonomy. Their breadth comes from scale and training, not from human-like general intelligence.
Why the label matters
Calling current AI "narrow" keeps expectations honest and separates real, deployed capability from speculation about AGI and superintelligence.
Narrow AI denotes task- or domain-bounded systems lacking general competence, encompassing essentially all deployed AI. The interesting modern wrinkle is that foundation models exhibit striking breadth, general-purpose across many tasks via in-context learning, blurring the intuitive narrow/general line while remaining narrow in the technical senses that matter: uneven, non-robust competence (the "jagged frontier"), limited reliable reasoning and world-modelling, no durable autonomy, and brittleness under distribution shift.
The distinction is therefore better framed as a spectrum of generality than a binary. Rigorous discussion avoids conflating breadth of application with the robust, transferable, self-directed generality implied by AGI. Maintaining the narrow-AI framing is important for calibrated claims, deployment risk assessment, and separating demonstrated capability from the contested trajectory toward more general systems.