Narrow AI (ANI)

What it means

Narrow AI is AI built for a defined job. IBM describes it as AI that "can be trained to perform a single or narrow task, often far faster and better than a human mind can," and adds: "However, it can't perform outside of its defined task."¹ For example, a system that spots fraud in card payments may catch patterns no clerk would see, but it cannot draft a contract or schedule a meeting. You would need a different system for each of those jobs.

The word matters because it is half of a pair. Its opposite is general AI, or [AGI](/definitions/agi): a system that could learn and do almost any intellectual task a person can. IBM and the tech publisher TechTarget both say that everything in use today falls on the narrow side. IBM calls narrow AI "the only type of AI that exists today,"¹ and TechTarget writes: "All forms of modern AI systems can be classified as narrow AI."²

Where the word comes from

We could not find a reliable source that shows who first said "narrow AI," so we do not name one. We can document that the term was in use by 2002. That year, an essay published on Ray Kurzweil's website, KurzweilAI.net, said: "There are today hundreds of examples of narrow AI deeply integrated into our information-based economy."³

The other name, "weak AI," has a different history. In 1980 the philosopher John Searle contrasted "strong AI" with "weak AI." The Stanford Encyclopedia of Philosophy summarizes the strong claim as the idea that "suitably programmed computers (or the programs themselves) can understand natural language." It describes weak AI as the far more modest claim that computers are useful tools for studying the mind.⁴ Searle was asking whether machines truly understand, not how many tasks they can do. Over time, "weak AI" came to be used as a synonym for narrow AI, and IBM and TechTarget now treat the two as the same.¹²

How others define it

• IBM: "It can be trained to perform a single or narrow task, often far faster and better than a human mind can."¹
• TechTarget: "Narrow AI systems can only do what they are designed to do and can only make decisions based on their training data."²
• Kurzweil's site (2002), using the word for everyday systems already at work: "hundreds of examples of narrow AI deeply integrated into our information-based economy."³

We searched NIST, the EU AI Act, and OECD materials and found no official definition of "narrow AI." These bodies define "AI system" in general terms instead. Please read that as "we did not find one," not as proof that none exists.

Where they agree and differ

They agree on the core idea: a narrow system is strong inside its job and cannot work outside it, and that describes all AI today.¹²

They differ in two ways. First, "weak AI" originally made a claim about understanding, not scope, so treating it as a synonym loses Searle's point.⁴ Second, it is not obvious where the line falls. IBM says that "even OpenAI's ChatGPT is considered a form of Narrow AI because it's limited to the single task of text-based chat."¹ Many readers will find that odd, since the same chatbot writes, translates, and summarizes. The strongest objection comes from a May 2026 paper by Boris Kriuk, which has not yet been peer reviewed. It argues that the narrow-versus-general split leaves out a middle stage: "Between the narrow systems we deploy and the general intelligence we speculate about lies an entire regime" that "has never received its own name."⁵ We think the objection is fair. Today's chat and agent systems do not fit neatly at either end. For leaders, though, the narrow label is still a useful default, because every system you can buy today still has limits on what it can do.

In the news

In a paper published in Nature, researchers from Carnegie Mellon, NYU, Stanford, and MIT presented Ataraxos, an AI for the board game Stratego.⁶⁷ According to the authors, "Ataraxos defeated the most decorated human Stratego player of all time by a large margin."⁶ The Decoder reports that it won 15 games, lost one, and drew four, and that training reportedly cost less than $8,000.⁷

The coverage does not use the words "narrow AI." Calling Ataraxos narrow is our judgment, and we are confident in it. It is a clear case: superhuman at Stratego and useless for writing an email. There is a twist, though. The authors say it was built on "general techniques" that also produced strong players for other games.⁶ They also suggest the methods could carry over to negotiations and markets, but only if someone builds an accurate simulator for each one.⁷ That is often how narrow AI gets wider: the method is reused, but each new job needs its own careful setup.

What it means for your organization

Every AI tool or [AI agent](/definitions/ai-agent) you deploy is narrow, even when it seems broad in a demo. That means three things. Give each system a defined job, as the [knowledge core](/definitions/knowledge-core) idea suggests. Watch for the moments when people ask it to work outside that job. And keep [a person in the middle](/definitions/a-person-in-the-middle) where a mistake would be costly. A narrow system will rarely tell you it has left its area. It will just get things wrong with confidence.

Ask your team: "For each AI tool we use, what job was it built for, and where are we already asking it to do something else?"