Quick answer
In this site's plain words, AI is software that works out an output (a prediction, recommendation, decision or new content) from its input, usually from patterns in data or examples. The US statute, NIST, the OECD and the EU AI Act word it more broadly. This page gives examples, a seller-label decoder, then the five official texts.
Key takeaways
- This site's plain version: AI is software that works out a prediction, recommendation, decision or new content from its input, usually from patterns in data or examples.
- The US statute, NIST's framework, the OECD and the EU AI Act all start from a machine-based system that produces outputs; they differ on content, autonomy and adaptiveness.
- Census Bureau survey, 24 August to 6 September 2026: 24% (95% CI 23% to 24%) of US employer businesses reported using AI in the last two weeks.
What is AI?
This site's 1-sentence plain version: AI is software that works out an output (a prediction, recommendation, decision or new content) from its input, usually from patterns in data or examples, not a rule written for each case. That is a gloss, not a standard. The official definitions below are broader than "usually from patterns in data", and the section on them shows how.
Google Cloud's explainer says you are using AI every day in "navigation apps like Google Maps" and "spam filters in your email", and IBM's lists "AI-powered chatbots and virtual assistants to handle customer inquiries".
Which label is a seller using, and how do you test it?
Hold the seller's label against one sourced sentence and ask the table's question. This is this site's decision aid, built from the texts quoted below; it does not say any product is or is not AI in law.
| The seller says | Label to look for | Sentence to hold it against | Ask the seller | Read on this site |
|---|---|---|---|---|
| "AI-powered" for rules you set (if this, then that) | automation, not necessarily AI | OECD: the system "infers, from the input it receives, how to generate outputs" | "For an input nobody wrote a rule for, what does it do?" | AI automation for small business |
| "Uses machine learning" | machine learning | IBM: "training an algorithm to make predictions or decisions based on data" | "What data trained it, and does it keep changing from mine?" | |
| "Generative AI" | AI that produces content | OECD list names "content"; the statute's list does not | "Who checks what it writes before a customer sees it?" | AI email responder for small business |
| "AI agent" | software acting on a goal | Only one NIST glossary entry names an "intelligent software agent"; the other texts do not use the word | "Which steps can it take without asking me?" | AI agents for small business |
| "AI decides" (hiring, lending, pricing) | a system that makes decisions | IBM: "biased training data used for hiring decisions might reinforce gender or racial stereotypes" | "What does it decide about people or money, and does a person review it before it takes effect?" |
If the answer to the first question is "nothing, it only follows the rule", this site would call the product automation, which may still be useful. This is this site's reading, not a legal conclusion: the statute's and NIST's wording read here does not clearly exclude rule-based software either.
Is AI already common in small businesses?
24% (95% CI 23% to 24%) of all US employer businesses reported using AI in the last two weeks, in the Census Bureau's Business Trends and Outlook Survey for 24 August to 6 September 2026. This is from this site's reviewed study of the Census data: the share answering Yes to whether the business used AI in any of its business functions in the last two weeks. It is self-reported, so it does not show which definition owners applied; the response rate is about 10%, only employer businesses are covered, and it measures use, not results.
How many small businesses use AI gives the breakdown by size and industry.
How do official sources define artificial intelligence?
The US statute, the OECD and the EU AI Act define AI as a machine-based system that produces outputs such as predictions or decisions; NIST's framework says "engineered or machine-based".
- US statute, 15 U.S.C. 9401(3) (read on Cornell LII): "The term "artificial intelligence" means a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments. Artificial intelligence systems use machine and human-based inputs to— (A) perceive real and virtual environments; (B) abstract such perceptions into models through analysis in an automated manner; and (C) use model inference to formulate options for information or action."
- NIST AI Risk Management Framework 1.0 (read on airc.nist.gov): "The AI RMF refers to an AI system as an engineered or machine-based system that can, for a given set of objectives, generate outputs such as predictions, recommendations, or decisions influencing real or virtual environments. AI systems are designed to operate with varying levels of autonomy (Adapted from: OECD Recommendation on AI:2019; ISO/IEC 22989:2022)."
- NIST glossary (read on csrc.nist.gov): the page lists six entries; entry (1), credited to CNSSI 4009-2022 from PL 115-232, Sec. 238, reads "Any artificial system that performs tasks under varying and unpredictable circumstances without significant human oversight, or that can learn from experience and improve performance when exposed to data sets." The SP 800-218A entry repeats the statute's first sentence almost word for word.
- OECD Recommendation on AI (read on legalinstruments.oecd.org; adopted 22/05/2019, amended 03/05/2024): "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. Different AI systems vary in their levels of autonomy and adaptiveness after deployment."
- EU AI Act, Article 3(1): "'AI system' means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and 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" (read on artificialintelligenceact.eu, maintained by the Future of Life Institute; EUR-Lex would not open for us).
Not legal advice. These are quotations, not an opinion on whether a law covers a product. The EU Article opens "For the purposes of this Regulation, the following definitions apply:". The US definition is one of the terms "In this chapter", and the division that enacted that chapter may be cited as the National Artificial Intelligence Initiative Act of 2020; it is not a general definition for all US law. NIST says its framework "is intended for voluntary use" and "The AI RMF 1.0 is being updated. A revised version is in progress." Ask a lawyer how any applies to you.
What do the definitions include, and where do they differ?
Of the five texts quoted, the four other than the NIST glossary agree on a system whose outputs influence an environment; they differ on content, autonomy and whether learning is named.
| Text | Outputs it lists | Autonomy mentioned | The word "learn" in the quoted sentence |
|---|---|---|---|
| US statute, 15 U.S.C. 9401(3) | predictions, recommendations or decisions | not stated on the pages we read | not stated; says "model inference" |
| NIST AI RMF 1.0 | predictions, recommendations, or decisions | "varying levels of autonomy" | not stated |
| NIST glossary | varies by entry; SP 800-218A lists predictions, recommendations, or decisions | the word is not used; entry (1) says "without significant human oversight" | entry (1): "can learn from experience"; entries (2) and (5) also name learning |
| OECD Recommendation | predictions, content, recommendations, or decisions | "vary in their levels of autonomy and adaptiveness after deployment" | not stated; says "infers" |
| EU AI Act, Art. 3(1) | predictions, content, recommendations, or decisions | "varying levels of autonomy" and "adaptiveness after deployment" | not stated; says "infers" |
"Content" appears in the OECD and EU lists but not in the statute's or the NIST framework's. An OECD.AI explainer by Russell, Perset and Grobelnik (29 November 2023; its disclaimer says the opinions are the authors', not necessarily the OECD's) says "content" was added to make clear the Recommendation covers generative AI, and calls it "technically, a sub-set of" predictions, recommendations or decisions, so the gap may be wording. None of those four quoted sentences (statute, NIST framework, OECD, EU) contains the word "learn". This is this site's reading, not a legal conclusion.
What do two vendor explainers add?
They add everyday examples and a stated limit, but they are vendor explainers, not legal definitions. IBM's page shows "Updated 02 October 2026"; Google Cloud's date is not stated on the pages we read.
| Page | Its definition sentence | A limit it states |
|---|---|---|
| IBM | "Artificial intelligence (AI) is technology that enables computers and machines to simulate human learning, comprehension, problem solving, decision making, creativity and autonomy." | in its ethics and legal risks section: "biased training data used for hiring decisions might reinforce gender or racial stereotypes" |
| Google Cloud | "Artificial intelligence (AI) is a set of technologies that empowers computers to learn, reason, and perform a variety of advanced tasks in ways that used to require human intelligence" | "poor training data can lead to biased or inaccurate outputs"; the page says this of what it calls Artificial Narrow Intelligence, which it says is "the only form of AI that currently exists." |
What should you do before you buy something called AI?
Get the answer to the decoder's question in writing before you pay.
- Ask what goes in, what comes out, and which decoder row the label matches.
- If the product decides things about people or money, ask whether a person reviews each decision before it takes effect.
- Check its output on a task whose answer you know.
- Start with one narrow task from the small business automation guide.
How was this page made?
This page was drafted with Claude, an AI assistant made by Anthropic, and every quotation was checked against the page it came from on 7 October 2026. Anthropic is not one of the organizations covered. Pages were read from outside the US. Gaps in the tables read "not stated on the pages we read".
Our data
24% (95% CI 23% to 24%) of all US employer businesses reported using AI in the last two weeks, Census Bureau BTOS period 202619 (reference period 24 August to 6 September 2026).
Method: Census-published national AI-use share for all US employer businesses in the Business Trends and Outlook Survey, as registered in this site's statistically reviewed study (key national_US). The survey question asks whether the business used AI in any of its business functions in the last two weeks; the denominator includes No and Do not know answers. One reference period, employer businesses only, self-reported, current question wording only, response rate about 10%; non-response bias cannot be assessed from the public files.
FAQ
Is there a legal definition of AI in the US?
Yes, in at least one place. 15 U.S.C. 9401(3) defines the term for its chapter; the division that enacted the chapter may be cited as the National Artificial Intelligence Initiative Act of 2020. This page quotes it and does not apply it to any business. Not legal advice.
What is an example of AI?
Google Cloud lists navigation apps, shopping recommendations, spam filters in email and virtual assistants. IBM lists voice assistant apps and chatbots that handle customer inquiries.
Is machine learning the same as AI?
No. IBM says "Directly underneath AI, we have machine learning", and Google Cloud says "Machine Learning (ML): This is a type of AI".
Do these definitions decide whether my tool is AI?
Only inside the text that uses them: the statute for its chapter, the EU AI Act for that Regulation, the NIST framework for those who choose to use it. A seller's brochure is bound by none.