If you keep asking what is artificial intelligence, you’re in good company, because even experts disagree. In short, it’s a machine-based system that infers outputs, such as predictions, content or decisions, from the input it receives.
AI is one of the many types of technology that shape daily life, and this page sticks to the basics.
| Item | Value |
|---|---|
| OECD countries began work on a shared definition | 2018 |
| OECD revised its definition of an AI system | 2023 |
| OECD post explaining the definition | March 6, 2024 |
| What the OECD chose to define | An “AI system”, not “AI” itself |
| Glossary used for common AI terms below | NIST glossary, March 2023 |
These figures come from the OECD.AI explainer and the North Carolina Department of Information Technology AI glossary. Definitions and dates can change, so check the sources before you rely on them.
What artificial intelligence means in plain terms
Official bodies tend to define an AI system rather than AI as a vague whole. In its March 2024 post, the OECD gives the revised definition: a machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments.
That sentence packs in a lot, so it helps to take it apart:
- Machine-based: it is a system built from machines, not a person.
- Objectives: it works toward goals that can be stated openly or built in implicitly.
- Infers from input: it works out how to produce an output from what it receives, such as data, a user prompt or a search query.
- Outputs: predictions, content, recommendations or decisions.
- Influence: those outputs can affect physical or virtual environments.
The same source adds that AI systems vary in their levels of autonomy and adaptiveness after deployment. Some do a narrow job under close human control, while others act with more independence.
Notice what the definition leaves out. It describes what a system does with input and output, and it does not require the system to think the way a person does.
Why is AI so hard to define?
The OECD authors describe the answer as elusive, even though the question is simple. Many of the world’s smartest people have tried to find a definition everyone agrees on, and they have not reached consensus. The post says there is no clear red line but a continuum of features that characterise what we think of as artificial intelligence.
The boundary also moves over time. Optical character recognition (OCR) was once widely considered AI, yet the general public no longer sees it that way, even though it employs textbook AI methods.
OECD countries started working on a shared definition in 2018. They revised it in 2023, and the OECD calls the resulting consensus a first, though not the “holy grail” of definitions.
The practical lesson is simple: when someone says a product “uses AI”, ask what it takes in, what it produces and how it was built. The label alone tells you very little.
How AI systems are built and used
Every AI system starts with input. During development, people supply knowledge, rules, code or data, and machine learning can use training data to produce a model. After deployment, the system uses new input, for example a user prompt or a search query, to infer its output.
The OECD describes an AI model as a core component of an AI system, used to make inferences from inputs to produce outputs. An AI system is typically built by combining one or more models, developed manually or automatically.
Machine learning is one way of getting there. It is a set of techniques that lets machines improve their performance, and usually generate models automatically, through exposure to training data rather than explicit instructions from a human.
Building does not always stop at launch. The OECD notes that deployers may fine-tune or continuously train models during operation, which can significantly affect a system’s performance and behaviour.
Common kinds of AI and where you meet them
The OECD says the term “AI” typically covers technique categories such as machine learning and knowledge-based approaches. It also covers application areas, which are the easiest way to spot AI in everyday tools:
- Computer vision
- Natural language processing
- Speech recognition
- Intelligent decision support systems
- Intelligent robotic systems
The OECD.AI site goes as far as saying AI is already a crucial part of most people’s daily routines. Whether or not you notice it, it often sits behind search, recommendations and voice tools.
Generative AI is the group that has drawn the most attention. The North Carolina glossary, drawing on NIST’s March 2023 glossary, defines it as the class of AI models that emulate the structure and characteristics of input data in order to generate derived synthetic content, such as images, videos, audio and text.
You steer such a tool with a prompt, which the glossary defines as an instruction or command given to an AI tool to carry out a task or function.
What AI is not, and how to use it sensibly
AI is not a single product or a single technique. It is an umbrella for many methods and uses, and its edges shift as the public’s view of what counts as “intelligent” changes.
It is also not a plain algorithm. The glossary describes an algorithm as a set of computational rules followed to solve a mathematical problem, whereas an AI system infers how to generate outputs from the input it receives.
And it is not automatically correct. The glossary defines a hallucination as generated content that is nonsensical or unfaithful to the provided source content, so check anything that matters.
The same glossary lists what trustworthy AI systems should aim for:
- Valid and reliable
- Safe, secure and resilient
- Accountable and transparent
- Explainable and interpretable
- Privacy-enhanced
- Fair, with harmful bias managed
Many AI tools run on remote servers, so it helps to understand what is cloud computing and where your data goes. Likewise, when you weigh up privacy, it is useful to ask how does encryption work and what it does and does not protect.
Frequently asked questions
Is AI the same as machine learning?
Show answer
No. Machine learning is one set of techniques within the wider field. The OECD lists machine learning and knowledge-based approaches as categories of techniques that the term AI typically covers.
Are all AI systems fully autonomous?
Show answer
No. According to the OECD, different AI systems vary in their levels of autonomy and adaptiveness after deployment.
What is a prompt?
Show answer
A prompt is an instruction or command given to an AI tool to carry out a task or function, according to the North Carolina glossary. The OECD also notes that input can take the form of a user prompt or a search query.
Why does AI sometimes produce wrong answers?
Show answer
Generated content can be what the glossary calls a hallucination: nonsensical or unfaithful to the provided source content. That is why it pays to verify important output yourself.
