presentsUNAIVERSE
The UN AI timeline

    Learn AI /AI Foundations /Artificial intelligence

    Artificial intelligenceAI

    Also called: AI.

    Detail

    In five seconds

    What it is

    An umbrella term for computational systems using symbolic reasoning, search or learned models for tasks such as prediction, classification or content generation.

    In plain words

    AI is a broad name for computer systems that take information in and produce useful outputs in ways that may resemble parts of human work.

    An everyday example

    A small team is translating a short field report from Arabic into English before a meeting. One person uses a phrasebook with fixed entries, another checks past translated reports for similar sentences, and a third drafts a fluent version that the editor then reviews line by line. The team starts with source text, uses different methods in the middle, and produces a translated draft at the end. The result can be useful, but it still needs checking for meaning, tone and missing details.

    Where the story stops working

    A human translation team has judgement, background knowledge and accountability in ways a computer system does not. AI systems process patterns in data or rules; they do not become responsible for the final content, and their errors may be harder to notice.

    Remember thisAI is an umbrella term, not one method. Ask whether a system uses rules, search, optimisation or a model fitted from data, then identify the input, the output and who checks the result.

    This term is contested

    What counts as AI is genuinely disputed. The boundary shifts across time, sectors and communities, so it is better to define the specific technique or system than rely on the label alone.

    Understand it

    Understand it

    Why it matters

    If you hear that a system uses AI, that alone tells you very little about its risks, value or limits. Good decisions require asking what kind of system it is, what data it depends on, what output it produces, how it is tested, and where human oversight remains necessary.

    Different things people call AI

    Help the reader explain that AI is a broad umbrella covering systems built in different ways, with different kinds of outputs.

    1. 01

      Input

      The system receives text, images, numbers, sound or other data.

      Draw one shared box at the top labelled 'Input: text, image, numbers, sound' with arrows splitting into two columns.

    2. 02

      Rules path

      One approach uses instructions written by people, such as if-then rules or expert logic.

      In the left column, draw a box labelled 'Human-written rules' leading to a box labelled 'Output: label, score, action'.

    3. 03

      Learned path

      Another approach uses a model whose settings were adjusted from examples in data.

      In the right column, draw a box labelled 'Model trained on data' leading to a box labelled 'Output: prediction, ranking, generated text'.

    4. 04

      Boundary note

      People disagree about exactly which systems count as AI, so the safer question is how the system works.

      Add a bottom note spanning both columns: 'AI is a broad label; ask what method, what data, and what oversight are used.'

    Two side-by-side columns show common AI approaches. Both begin with input data and end with an output, but the middle stage differs: fixed rules in one column, learned patterns in the other. A caption below says both may be labelled AI, while simple automation may not involve either approach.

    Hovering or tapping each middle box reveals a short note: 'Rules are specified directly by people' or 'Patterns are learned from examples'.

    Describe this without the diagram

    The diagram compares two common approaches that are both often called artificial intelligence. At the top, one input box lists text, images, numbers and sound. This splits into two paths. The left path goes to a box called human-written rules, then to an output box labelled label, score or action. The right path goes to a box called model trained on data, then to an output box labelled prediction, ranking or generated text. A note across the bottom says AI is a broad label and that people should ask what method, what data and what oversight are used.

    With reduced motion: A static two-column diagram shows the same structure with no animation: shared input at the top, rules path on the left, learned-model path on the right, and a note at the bottom about the broad and disputed boundary of AI.

    Artificial intelligence is a broad label, not one single tool. People use it for computer systems that take some input, process it with rules or learned patterns, and return an output such as a score, a label, a suggestion or a piece of text.

    What goes in can be text, images, sound, numbers or other data. What happens in the middle depends on the system. Some AI systems follow hand-written rules. Others use a model, meaning a mathematical system whose internal settings were adjusted using examples from data. What comes out might be a translation, a forecast, a ranked list or a draft paragraph.

    AI is often confused with machine learning. Machine learning is one important way to build AI, but not the only one. AI is also often confused with automation. Automation means a system carries out steps automatically; some automation uses no AI at all.

    The boundary of AI is disputed and has changed over time. A spell-checker, a fraud detector and a large language model may all be called AI by different people, while others would reserve the term for only some of them. That is why it helps to ask what the system actually does, what data it uses, and how much human oversight it needs.

    How “A translation desk with different tools” maps onto the real thing

    In the storyIn the AI system
    phrasebook with fixed entriesrule-based system
    past translated reportstraining data or reference data
    drafter producing a fluent versionmodel generating an output
    editor reviewing the drafthuman oversight and quality control

    A worked example

    A programme officer is reviewing de-identified comments from a needs assessment before the coding table goes into a monitoring report. The available categories are access, search, training and other.

    Starting from: Example response: "I can open the portal, but I rarely find the document unless someone sends me the link."

    1. The system receives the response as text input.
    2. A model fitted from previously labelled responses detects language associated with finding material, rather than gaining access to the portal.
    3. The model scores the available categories and proposes search as the highest-scoring label.
    4. The programme officer checks the proposal and reviews any answer for which scores are close or more than one category may apply.

    Result: The answer is coded as search for the draft table. This is a predicted label from one type of AI system, not a definitive interpretation of what the respondent meant.

    In a United Nations workplace

    In a documentation and library service, a programme officer is coding open-ended needs-assessment answers for a monitoring report due to the front office by close of business. An approved AI tool proposes labels such as access, search and training, while uncertain or multi-issue answers remain in the review queue.

    This classifier is one application of AI, not the whole field. It uses a model fitted from labelled examples to map text inputs to category outputs. Other AI systems may instead use symbolic rules, search or optimisation. The programme officer still determines whether the labels and resulting analysis are fit for the report.

    Before you try thisRaw answers containing personal, medical, protection-related or other sensitive information must not be entered into a tool unless it is approved for that type of material. De-identification reduces risk but does not guarantee anonymity. Follow the office's data protection guidance, restrict access and check whether misclassification could distort whose needs appear in the report.

    Where you may hear it

    • At a vendor demo convened by a procurement unit
    • During an ICT governance board review of a proposed tool
    • In the margins of a policy discussion on digital technologies
    • At a programme meeting before an AI-assisted feature is added to the workplan

    Avoid confusion

    Avoid confusion

    Often confused with

    Side-by-side

    Common misconceptions

    • AI always means a system like a chatbot that writes fluent text.

      Chatbots are only one visible type. AI can also classify images, detect patterns in transactions, rank search results or follow fixed expert rules.

    • AI and automation are the same thing.

      Automation means doing tasks automatically. A payroll script can be automated without any AI, while an AI system may support a task without fully automating it.

    • If a system is called AI, it must be intelligent in a human sense.

      The label refers to the kinds of tasks or methods involved, not to human-like awareness or judgement. Many AI systems are narrow tools with limited scope.

    • Anything difficult to explain in software is AI.

      Many complex software systems use ordinary programming, databases or statistics. It is better to ask about the specific method than rely on the label.

    Quick check

    Which statement best describes artificial intelligence?

    Go deeper

    Go deeper

    The simple version

    AI is a broad name for computer systems that take information in and produce useful outputs in ways that may resemble parts of human work.

    The technical version

    AI is an umbrella term for computational systems that map inputs to outputs using techniques such as symbolic rules, search, optimisation and models fitted from data, including machine-learning systems.

    Reviewed external resources

    Where this came from

    Drafted with a language model through an offline pipeline, checked against this site's schema, and every external link above fetched to see whether it answers. Treat it as a starting point and check anything that matters against the primary sources. How this site is built and what that means.

    The one thing to remember

    AI is an umbrella term, not one method. Ask whether a system uses rules, search, optimisation or a model fitted from data, then identify the input, the output and who checks the result.