If you keep asking what is artificial intelligence and getting vague answers, you’re not alone. In short, it is a machine-based system that infers from input how to produce predictions, content, recommendations or decisions. Experts still disagree on one exact wording.
| Item | Value |
|---|---|
| OECD countries began work on an AI definition | 2018 |
| OECD revised its AI system definition | 2023 |
| OECD “What is AI?” article published | March 6, 2024 |
| US law cited by Cornell LII for AI | 15 U.S. Code § 9401 |
| Cornell LII entry last reviewed | September 2025 |
These figures come from the OECD.AI article on defining AI and the Cornell Legal Information Institute entry. Definitions and legal references can change, so check the sources before you rely on them.
Why do experts disagree on a definition?
The OECD’s March 2024 article says that many of the world’s smartest people have tried to write a solid definition of AI that everyone accepts, and that they never seem to find consensus. That is a useful warning for a beginner: if one explanation sounds different from another, both can still be reasonable.
The same authors describe a continuum of features rather than a clear red line between AI and non-AI. A system can show a little of those features or a lot, and where you draw the line depends on who is asking and why.
The boundary also moves over time. The OECD article notes that some technologies, such as optical character recognition (OCR), were once widely considered AI but are no longer seen that way by the general public, even though they use textbook AI methods.
Because of this, the OECD chose not to define “AI” itself. Its authors say they decided to define an AI system instead, which they call a more tangible and actionable concept, especially in policy making. OECD countries started that work in 2018 and revised the definition in 2023.
The OECD definition of an AI system, piece by piece
As of the OECD’s March 2024 article, the revised definition reads: an AI system is 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. It adds that different AI systems vary in their levels of autonomy and adaptiveness after deployment.
That sentence is dense, so here it is broken into parts:
- Machine-based: the system runs on machines, such as software on computers.
- Objectives: it works toward goals that can be explicit (written down) or implicit (built into how it behaves).
- Infers from input: it works out what to produce from what it receives. Input can be data, or a user prompt, or a search query.
- Outputs: predictions, content, recommendations or decisions.
- Environments: those outputs can influence physical or virtual settings.
- Autonomy and adaptiveness: these levels vary from system to system after deployment.
The OECD also explains what sits inside such a system. An AI model is described as a core component used to make inferences from inputs to produce outputs, and a system is typically built by combining one or more models before deployment.
Many models come from machine learning. The OECD describes it as a set of techniques that lets machines improve their performance, and usually generate models automatically, through exposure to training data rather than through 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 change a system’s performance and behaviour.
How standards and law describe artificial intelligence
You do not have to rely on a single organisation’s wording. The NIST Computer Security Resource Center glossary lists five numbered definitions of AI, taken from a US federal source, plus a separate machine-based definition from a NIST publication.
The numbered entries each stress a different angle:
- Systems that perform tasks under varying and unpredictable circumstances without significant human oversight, or that learn from experience when exposed to data sets.
- Systems that solve tasks requiring human-like perception, cognition, planning, learning, communication, or physical action.
- Systems designed to think or act like a human, including cognitive architectures and neural networks.
- A set of techniques, including machine learning, designed to approximate a cognitive task.
- Systems designed to act rationally, such as an intelligent software agent or embodied robot that achieves goals.
The NIST glossary’s machine-based definition is close to the OECD wording: a system that, for a given set of human-defined objectives, makes predictions, recommendations, or decisions influencing real or virtual environments. Notice that the OECD, NIST and the law all lean on objectives and on outputs that affect an environment.
The Cornell Legal Information Institute entry says that AI refers to computer-based systems capable of functions that typically require human intelligence, such as recognizing patterns, adapting from experience, or making decisions. As of its September 2025 review, it cites 15 U.S. Code § 9401, which describes a machine-based system that operates toward human-defined objectives.
The same Cornell entry adds that the Fiscal Year 2019 National Defense Authorization Act defines AI more broadly. Those legal definitions exist to guide policy and enforcement, so treat them as reference points rather than a recipe for building a system.
The main techniques and application areas
The OECD says the term “AI” typically covers categories of techniques, such as machine learning and knowledge-based approaches. It also covers application areas, which are the places where those techniques get used:
- Computer vision, which is the topic of what is computer vision.
- Natural language processing, which is covered in what is natural language processing.
- Speech recognition.
- Intelligent decision support systems.
- Intelligent robotic systems.
Most of the beginner questions people ask next are about machine learning. A good starting point is the guide on machine learning vs deep learning, which sorts out two terms that are often used as if they meant the same thing.
If you want to see what happens inside a model, the explainer on how do neural networks work goes through the mechanics step by step. NIST’s glossary mentions neural networks inside one of its definitions, so the idea is worth knowing early.
Learning styles matter too. The article on supervised vs unsupervised learning compares two common ways a machine can learn from data, and the guide to what is reinforcement learning covers a third.
Models can also go wrong in ways that surprise beginners. The page on what is overfitting in machine learning explains one common problem.
Cornell LII also records a distinction drawn by a court between predictive AI, which classifies or forecasts, and generative AI, which creates new content that may not always be reliable. If the second kind interests you, read what is generative ai next.
Everyday life, risks and next steps
The OECD says AI is already a crucial part of most people’s daily routines. That claim comes from its OECD.AI site, which describes itself as a platform dedicated to promoting trustworthy, human-centric AI.
Trust is a real concern, and regulators have said so. As Cornell LII reports, in 2023 the EEOC, DOJ, CFPB and FTC issued a Joint Statement on Enforcement Efforts Against Discrimination and Bias in Automated Systems. It reminded businesses that automated decision-making systems remain subject to anti-discrimination and consumer protection laws.
Cornell LII also notes that in 2023 the White House issued an Executive Order on the safe and trustworthy development and use of AI. At the international level, it says the United Nations’ Global Digital Compact sets a framework for responsible AI governance, multi-stakeholder cooperation, and digital rights.
Reliability is a second concern. According to Cornell LII’s September 2025 review, in a 2025 decision called In re Martin, lawyers were sanctioned for submitting a brief with fictitious citations generated by a large language model. The lesson for everyday users is simple: check what an AI tool gives you before you rely on it.
Unfair outcomes are the third. The collection of ai bias examples shows how automated decisions can treat groups differently, which is the risk the 2023 Joint Statement points to in areas like employment and credit.
When you are ready to study, follow this short path:
- Learn the OECD definition above so you can spot the parts in any product description.
- Read about the main techniques, starting with the machine learning guides linked in the previous section.
- Follow the study plan in how to learn machine learning.
- Widen your view with the hubs on emerging technologies and ai startups.
This site is an independent educational resource. It does not sell services or take enquiries, and nothing here is legal advice.
Frequently asked questions
What is artificial intelligence in simple words?
Show answer
It is a machine-based system that takes input, works out how to respond, and produces predictions, content, recommendations or decisions that can influence the real or virtual world. That follows the OECD’s revised definition, as described in its March 2024 article.
Is there one official definition of AI?
Show answer
No. The OECD says experts never seem to find consensus, and NIST’s glossary lists several definitions. As of the Cornell LII entry reviewed in September 2025, US law also words it in more than one way.
Is optical character recognition still AI?
Show answer
It depends on who you ask. The OECD says OCR was once widely considered AI and no longer is in the eyes of the general public, even though it employs textbook AI methods.
What is the difference between predictive and generative AI?
Show answer
Cornell LII records a court distinction: predictive AI classifies or forecasts, while generative AI creates new content that may not always be reliable. This is as reported in its September 2025 review.
Do anti-discrimination laws apply to automated decisions?
Show answer
According to Cornell LII, a 2023 Joint Statement from the EEOC, DOJ, CFPB and FTC reminded businesses that automated decision-making systems remain subject to anti-discrimination and consumer protection laws. Check current guidance before you act on this.
