Textbook of AI

Also known as: weak AI, applied AI

Narrow AI, also called weak AI or applied AI, refers to systems designed and trained for a particular, well-defined task. A spam filter, a facial recognition system, a chess engine, and a medical image classifier are all examples. Each can be extraordinarily effective within its domain—often surpassing human performance—but none has any understanding of anything outside that domain. A system trained to play Go cannot translate French, and a language model cannot drive a car.

All AI systems currently deployed in the world are narrow AI, including large language models. Although LLMs appear superficially general because they can produce fluent text on virtually any topic, their "intelligence" is bounded by the distribution of text they were trained on and the patterns they extracted from it. They do not have goals, persistent memory across interactions, or grounding in the physical world. The impressive capabilities of narrow systems should not be mistaken for general intelligence.

The distinction between narrow and general AI is central to debates about AI's future trajectory. Narrow systems pose concrete present-day risks and benefits—job displacement, bias, privacy, improved diagnostics—while general AI (AGI) raises more speculative questions about alignment, control, and the long-term future of humanity. Understanding which kind of system you are discussing is essential for reasoning clearly about AI policy and safety.