# What is Narrow AI?

Also called weak AI.

Narrow AI, also called weak AI, refers to systems built and trained to perform a specific task or a bounded set of tasks, such as recognizing faces, ranking search results, or transcribing speech. Performance can be very high inside that scope and typically degrades sharply outside it. Every artificial intelligence system in production use today is narrow in this sense.

The term functions as a contrast with the hypothetical notion of general intelligence. It says nothing about how capable a system is, only about the breadth of the domain over which its competence holds. A narrow system may reach very high accuracy on its specific task while being unable to answer a simple question one step outside the distribution it was trained on.

Boundaries have become harder to draw. A large language model handles translation, classification, summarization, and code generation without task specific training, which is broader than a classical narrow system, yet it remains bounded by the distribution of its training data and by its interfaces. Some authors describe such models as general purpose without being generally intelligent.

The distinction matters for deployment. Because a narrow system's competence is scoped, its evaluation, monitoring, and guardrails can be scoped as well. Documenting the intended operating domain, testing for inputs that fall outside it, and defining behavior when the system is uncertain are standard practices, and model cards exist to record exactly these boundaries.

## Key points

- Systems competent within one bounded task domain.
- Also called weak AI; contrasts with hypothetical general AI.
- High skill inside scope, brittle immediately outside it.
- All deployed AI systems today are narrow.
- Scoping enables focused evaluation and guardrails.

## In practice

A model trained to detect defects on one production line reaches very high accuracy on the parts and lighting it was trained with. Move the same camera to a different line with different components, and accuracy collapses, because nothing in the training data described those parts. The system is expert within a narrow domain and uninformed just outside it.

## Related terms

- [Artificial General Intelligence](/en/glossary/artificial-general-intelligence)
- [Machine Learning](/en/glossary/machine-learning)
- [Model Card](/en/glossary/model-card)
- [Large Language Model](/en/glossary/large-language-model)
- [Computer Vision](/en/glossary/computer-vision)

[Back to the AI Glossary](/en/glossary)