What is artificial intelligence? If the term feels vague, you’re not alone. In plain terms, it is a machine-based system that takes input and infers how to produce outputs such as predictions, content, recommendations or decisions.
This explainer is part of our technology news coverage, and it sticks to how public sources define AI.
| Item | Value |
|---|---|
| OECD first proposed trustworthy-AI principles, with an early AI definition | 2019 |
| Output types named in the OECD definition | 4 (predictions, content, recommendations, decisions) |
| EU Parliament political agreement on the definition, as reported | 8 March 2023 |
| Definition moved from Annex I to its own article in the EU AI Act text, as reported | November 2022 |
These figures come from iubenda and Biometric Update. They can change, so check the sources before you rely on them.
The official definition, in plain words
The most widely quoted wording comes from the Organisation for Economic Co-operation and Development (OECD). As iubenda reports it, the definition reads: “An AI system is a machine-based system that infers, for explicit or implicit objectives, from the input it receives, how to generate outputs like predictions, content, recommendations, or decisions influencing physical or virtual environments.”
That is dense, so here is the everyday version. An AI system is a machine that receives something, works out what to produce, and hands back a result that can affect the world around it.
The OECD proposed a set of principles for trustworthy AI in 2019, and that set included an earlier version of the definition, according to Freevacy’s summary of Euractiv’s reporting. The OECD Council later adopted the new wording quoted above.
What each key word means
Each phrase in the definition does a specific job. Read one at a time, they are easier to follow.
- Machine-based: the system runs on machines rather than being a person doing the thinking.
- Infers: it draws a conclusion from the input it receives.
- Explicit or implicit objectives: the goal may be spelled out, or it may be implied.
- Outputs: predictions, content, recommendations or decisions.
- Physical or virtual environments: the results can affect things in the real world or in software.
The Euractiv summary on Freevacy adds one more line: “Different AI systems vary in their levels of autonomy and adaptiveness after deployment.” So autonomy is a sliding scale, and a system does not have to act entirely on its own to count.
Notice that content is only one of the four outputs. A system that predicts something or recommends something fits the definition just as well as one that writes text or makes video.
Why lawmakers care how AI is defined
A definition decides which products a law covers. Biometric Update put it this way: “As the AI Act is intended to prevent harm from AI, the definition of what AI is proves fundamental.”
iubenda described the EU’s AI Act as “a pioneering legislative proposal aiming to regulate AI based on its potential harm”. It also said the EU Parliament had agreed to align with the OECD’s definition, which keeps the wording consistent with international standards.
The Freevacy summary said the new OECD definition was set to be included in the Act to outline its scope. These are 2023 reports about a text still being negotiated, so they do not tell you what the final law says.
If you want a professional take on the same question, Deloitte Luxembourg has published a perspective titled “Defining AI under the EU AI Act: Clarity for compliance and innovation”. The page is a 10 minute read and covers the European Commission’s guide.
How the EU wording changed along the way
The definition did not appear in one go. The sources describe a few steps, as reported at the time.
- November 2022: the definition was moved from Annex I to its own article, which prevented the European Commission from altering it later.
- March 2023: the European Parliament reached a political agreement on a definition that was effectively the OECD’s.
- Mid-October: the OECD’s Committee on Digital Economy Policy and Working Party on AI Governance discussed the new definition, and the OECD Council then adopted it.
The March 2023 draft text read: “‘Artificial intelligence system’ (AI system) means a machine-based system that is designed to operate with varying levels of autonomy and that can, for explicit or implicit objectives, generate output such as predictions, recommendations, or decisions influencing physical or virtual environments.” Compare it with the OECD wording and you will spot a difference: the draft lists predictions, recommendations and decisions, while the newer OECD text also names content.
Biometric Update added that the definition was expected to change further with the removal of “machine-based”, according to Euractiv sources. It also said the narrower definition suited conservative groupings, while left-of-center politicians had pushed for a broader one.
Where do foundation models fit in?
iubenda noted that the AI Act negotiations introduced obligations for foundation models and General Purpose AI. It said this acknowledged capabilities ranging from generating content like text and videos to evolving after deployment through machine learning techniques.
That connects back to the definition. Generating content is one of the four outputs, and adapting after deployment is part of the autonomy and adaptiveness language.
How to read AI claims in the news
Headlines often use “AI” loosely. A short checklist, built from the definition, helps you test a claim.
- What is the input the system receives?
- What output does it produce: a prediction, content, a recommendation or a decision?
- Who set the objective, and is it stated or implied?
- How much autonomy does the system really have?
- How old is the source, and has the rule or product changed since?
Our guide on how to spot fake tech news shows how to apply that last question to everyday stories. Because an AI system is machine-based, it also helps to understand the hardware side, so read what is a semiconductor chip next.
Frequently asked questions
What is artificial intelligence in simple terms?
Show answer
It is a machine-based system that takes in input and works out outputs such as predictions, content, recommendations or decisions. That is the OECD’s definition, as reported by iubenda.
Does AI have to act fully on its own?
Show answer
No. The Euractiv summary on Freevacy says different AI systems vary in their levels of autonomy and adaptiveness after deployment.
Is AI just chatbots and image generators?
Show answer
No. Content is one of four outputs in the OECD definition, alongside predictions, recommendations and decisions.
Is the OECD definition the same as the EU AI Act definition?
Show answer
As of the March 2023 Biometric Update report, the Parliament’s draft was effectively the OECD’s. As of the 2023 Freevacy summary, the OECD wording was set to be included in the Act. These sources do not show the final text, so check the Act itself.
