Narrow artificial intelligence (ANI)

in English: narrow AI (artificial narrow intelligence)

basicwidespread

An artificial intelligence that performs a single task or narrow scope of tasks well but is unusable outside that scope.

In plain terms

A good example of narrow artificial intelligence is a chess program that can defeat the world champion but is capable of nothing else: it cannot write a recipe or read a map. Within its own domain, it can be better than any human, but outside this circle, it is useless.

This includes spam filters, facial recognition, navigation route planners, bank fraud detectors, and programs that analyze X-ray images. Each was built for a well-defined task, and most were trained on examples.

The counterpart to this concept is artificial general intelligence (AGI). It would be able to adapt to new situations and learn any intellectual task like a human. Narrow AI already surrounds us, while AGI remains a vision for most experts.

Today's chatbots and AI agents stand between the two. They solve many tasks, so they are no longer 'narrow' in the old sense, but there is no human-level, all-encompassing general intelligence. The industry debates how to classify them.

How it works

Developers assign a specific task, such as having the system identify mole abnormalities from a photograph.
Thousands or millions of examples are collected for this single task, and the system learns characteristic patterns from them. (Some older narrow AI, such as expert systems, do not learn from examples but follow hand-written rules.)
The finished model is evaluated using a metric suitable for the task, such as how many cases it identifies correctly and how many it misses.
In production, the system only gives sensible answers to the inputs it was trained on: a dermatology model does not understand chest X-rays.
If they want to use it for a new task, it must be retrained or further trained, or even rebuilt, because it does not transfer its knowledge to other areas on its own.

An everyday example

When your phone unlocks the screen using facial recognition, you are using narrow artificial intelligence. This program identifies your face with extreme precision, but if you asked it what the weather will be tomorrow, it would not know what to do with the question.

Who is affected

everyoneThrough search engines, recommendation systems, maps and spam filters, you use them dozens of times daily, mostly unnoticed.

employeeNarrow AI typically takes over individual subtasks from you, such as data entry or pre-screening, rarely the entire job. → Work & HR

patient, doctorDiagnostic-findings analysis programs are typically built for a single, narrow examination task, so the doctor still puts together the full picture. → Health

entrepreneur, executiveWhen procuring, you must know exactly what task the system was built for, as you cannot expect reliable results outside that scope.

A common misconception

MythNarrow artificial intelligence is weak, stupid, or outdated technology.

In reality'Narrow' refers to the scope, not the performance. It can outperform the best humans in its specific task, but it cannot be used for anything else.

Limits and risks

Unreliable outside its training set: it can confidently err in unusual situations that differ from its examples.
It does not understand in the human sense what it does: it recognizes patterns, lacking common sense and broader background knowledge.
It inherits the biases of its training data, so a system that learned from a poorly compiled database may make unfair decisions against certain groups.
The boundary between narrow and general AI is not sharp, so labels appearing in news should be treated with caution.

Similar, but not the same

general AI – AGI would solve any intellectual task at a human level, while narrow AI only solves the one task it was built for.
general-purpose AI model – This term, also used in the EU AI Act, refers to a model capable of various tasks, such as a large language model. It is broader than narrow AI, but not human-level general intelligence.
superintelligence – Superintelligence would surpass humans in every field, while narrow AI covers at most a single field.

Related concepts

Broader

artificial intelligence

Narrower

expert system, recommendation system, image recognition, speech recognition

On the edge

general AI, superintelligence, machine learning, general-purpose AI model, AI-agent

A brief history

The terms 'weak' and 'strong' AI were introduced by philosopher John Searle in 1980 in his paper on the Chinese room thought experiment. For him, it was not about the scope of tasks: 'weak AI' meant that the machine only imitates thinking, while 'strong AI' meant that it truly thinks and understands. The term 'narrow AI' spread later, following debates on artificial general intelligence, and today common language often uses it as a synonym for 'weak AI'.

Where things stand

At the end of 2026, most AI systems working in practice are narrow AI, present everywhere from phones to factories. However, because of large language models and agents, the boundary between narrow and general AI is increasingly disputed.

Status: 2026 November

In the news

Appears in 8 Agent Age stories; the curve shows weekly occurrence in the news magazine.

Analyses examine AI training and copyright in India amid ANI v OpenAI disputeOct 20, 2026 · Jog
ANI v. OpenAI: Indian analysis of AI training and fair dealingOct 6, 2026 · Jog
Delhi High Court refuses immediate stay in ANI v OpenAI copyright caseSep 15, 2026 · Jog
Booking.com, Windscribe and Shopee sellers draw criticism over AI-generated advertisingAug 24, 2026 · Marketing
AI and copyright: Thomson Reuters win, ANI v OpenAI, Apple shareholders, BörsenvereinAug 17, 2026 · Jog
Indian banks move from AI experimentation to deployment at scaleAug 11, 2026 · Money
Delhi High Court rules in India's first AI copyright case, ANI v OpenAIAug 11, 2026 · Jog
Delhi High Court refuses interim injunction in ANI v OpenAI copyright caseAug 3, 2026 · Jog

Also known as

narrow AI · narrow artificial intelligence · gyenge MI · narrow AI · weak AI · artificial narrow intelligence · ANI

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.