Narrow AI
In short
Narrow AI is artificial intelligence built to do one task, or a small set of tasks, such as playing chess, predicting protein shapes or spotting spam. It can be very good at that job but cannot turn its hand to anything else.
- Also called
- Weak AI, Artificial narrow intelligence, ANI, Task-specific AI
- Topic
- The basics
- Level
- Beginner
- First described
- John Searle introduced the "weak" versus "strong" AI distinction in his 1980 paper "Minds, Brains, and Programs"2. Who first said "narrow AI" is unclear.
The UK government's 2023 paper on frontier AI defines narrow AI as "an AI system that performs well on a single task or narrow set of tasks, like sentiment analysis or playing Chess"1.
Narrow does not mean weak in ability. The same paper names AlphaGo, AlphaFold and the image generator DALL·E 3 as "powerful narrow AI systems", while noting they cannot perform as wide a variety of tasks as frontier chatbots1.
The opposite is general-purpose AI. The EU AI Act defines a general-purpose AI model as one that "displays significant generality" and can competently perform "a wide range of distinct tasks"3. Chatbots built on large language models fall on that side of the line.
Everyday examples of the idea include a photo app that groups faces, a bank's fraud check or a sat nav's route planner. Each is trained or programmed for one job.
"Weak AI" is often used as a synonym, but it started out meaning something else. The philosopher John Searle coined "weak" or "cautious" AI in 1980 for the view that computers are a useful tool for studying the mind, as opposed to "strong" AI, the claim that a suitably programmed computer really has a mind2.
Example
AlphaFold predicts the 3D shape of proteins far better than older methods, but it cannot write an email or play chess. The UK government cites it as a powerful narrow AI system1.
Why it matters
Most rules and risks people worry about depend on whether a system does one thing or many. Narrow AI can still cause harm, but its harms are easier to predict than those of general-purpose models.
Questions people ask
Is ChatGPT narrow AI?
Usually not. Chatbots like ChatGPT can do a wide variety of tasks, which is the mark of general-purpose AI. The UK government contrasts them with narrow systems such as AlphaGo and AlphaFold1.
What is the difference between narrow AI and general AI?
Narrow AI performs well on a single task or narrow set of tasks1. General-purpose AI can competently perform a wide range of distinct tasks3.
Is narrow AI the same as weak AI?
People often use the terms interchangeably today. Originally, John Searle's "weak AI" meant using computers as a tool to study the mind, not a system limited to one task2.
Related terms
Sources
- Department for Science, Innovation and Technology (2023), Frontier AI: capabilities and risks, discussion paper, introduction and footnote 7
- Searle, J. R. (1980), Minds, Brains, and Programs, Behavioral and Brain Sciences 3(3)
- Regulation (EU) 2024/1689 (AI Act), Article 3(63), definition of 'general-purpose AI model', via artificialintelligenceact.eu (EUR-Lex blocks automated access)