The distinction between narrow AI and general AI is one of the field’s oldest framings. Narrow AI, sometimes called weak AI, refers to systems built to perform a specific task or a bounded set of tasks: playing chess, recognizing faces, recommending products, transcribing speech. Such systems can reach or exceed human performance within their domain while being completely unable to do anything outside it. A chess engine that crushes grandmasters cannot drive a car or hold a conversation. Essentially all AI in commercial use has historically been narrow AI.
General AI, or artificial general intelligence, refers to the original ambition of the field: a system with the broad, flexible competence of a human mind, able to learn and perform across a wide range of tasks rather than one. The 2023 DeepMind paper “Levels of AGI” makes the point that generality is best treated as a dimension, not a switch - systems vary in how broad their competence is as well as how strong it is at any given task. By that framing, a narrow system can be superhuman in depth while scoring near zero on breadth, whereas a general system spreads competence across many domains.
The line between the two has always been slippery, partly because of what is called the AI effect: once a task is automated, it tends to be reclassified as mere computation rather than real intelligence, so the goalposts for “general” keep moving. Large language models complicated the picture further, since a single model can now handle many loosely related tasks without being retrained for each - more general than classic narrow systems, but, by most accounts, still short of human-level generality.
Why business readers should care: most deployed AI is narrow and reliable only inside the task it was built for. Treating a narrow system as if it were generally capable - trusting it outside its domain - is a common and avoidable source of failure.