AI Glossary: Plain-English Definitions of Common AI Terms

How to use this glossary#

Use this glossary as a quick reference when you meet an unfamiliar word. The terms below are listed in alphabetical order, each with a short definition and links to related terms. Every term has its own fixed link, so other guides on this site can send you straight to the definition you need.

A few tips make it more useful:

• Search the page. Use your browser's find function to jump to a word, then read the linked terms around it.
• Follow the chain. Many ideas build on each other. A token explains the context window, and an embedding explains semantic search.
• Check the source for rules. Entries on Singapore laws and frameworks summarise the official pages listed in the sources. For a real decision, read the official guidance itself.

Definitions explain concepts, not specific products, so they stay useful as tools change.

Which terms matter for your role?#

Which terms matter depends on what you do with AI. Everyday users mostly need words about prompts, accuracy and data. Builders need the vocabulary of retrieval, tools and testing. Leaders need terms about value, adoption and accountability. The table below suggests a starting set for each role and where to read more.

Most people need a little of every row. A builder still handles personal data, and a manager still needs to know what a hallucination is before approving a tool. Start with your own row, then read across.

Where to go next#

Once a term makes sense, the next step is to see it in use. The guides on this site turn these definitions into practical steps, from writing better prompts to testing an AI app. A structured route through them, ordered from first steps to governance, makes the learning stick faster than reading terms one at a time.

The AI learning pathway orders the guides by stage, so you can work from everyday use to building and governing AI. If you would rather learn by doing, the GenAI Builder Sprint course is a 2-day classroom course where adults plan, build, test and launch a simple AI app. No previous AI or programming experience is needed, although formal entry requirements apply, and funding is subject to eligibility.

Terms A to Z#

AI agent#

An AI system that works towards a goal by planning steps, using tools and acting with some independence, rather than only replying to a single prompt. Agents usually rely on tool calling to take actions.

AI governance#

The policies, roles, processes and controls an organisation uses to make sure AI is used responsibly, lawfully and in line with its goals. An AI use policy and AI risk assessment are common parts of it.

AI model#

The trained program at the core of an AI system that turns inputs into outputs. An app or chatbot is built around one or more models, plus instructions, data and safeguards.

AI risk assessment#

A structured review of what could go wrong with a particular AI use, how likely and how serious each problem would be, and which controls reduce the risk. It is usually repeated when the use or the tool changes.

AI use policy#

An organisation's written rules for how staff may use AI tools: which tools are approved, what data may be entered, which outputs need human review and who to ask for help.

AI Verify#

An AI governance testing framework and software toolkit developed by IMDA. It helps organisations assess their AI systems against 11 internationally recognised AI governance principles through standardised tests, made up of technical tests and process checks, and covers both traditional and generative AI.

API#

Short for application programming interface: a defined way for one piece of software to request data or services from another. Many AI services let other apps send a prompt and receive a reply through an API.

Artificial intelligence (AI)#

Computer systems that perform tasks usually associated with human intelligence, such as recognising patterns, understanding language, making predictions or creating content.

Bias#

Systematic, unfair differences in an AI system's outputs for particular people or groups. It often comes from unbalanced training data or from design choices, and is checked through evaluation.

Change management#

A structured approach to helping people and teams adopt a new way of working, such as a new AI tool, through clear communication, training, support and follow-up.

Chunking#

Splitting long documents into smaller passages before creating embeddings, so a RAG system can retrieve only the part that answers a question.

Context window#

The maximum amount of text, measured in tokens, that a model can take into account at once. It includes your prompt, any documents you add, the conversation so far and the model's reply.

Embedding#

A list of numbers that represents the meaning of a piece of text or an image, so that items with similar meaning have similar numbers. Embeddings power semantic search and RAG.

Evaluation (evals)#

Systematic testing of an AI system's outputs against agreed test cases and criteria, such as accuracy, safety and consistency. Evals are run before release and again after changes.

Explainability#

How well people can understand why an AI system produced a particular output or decision. It matters most when outputs affect people, for example in hiring or lending.

Fine-tuning#

Further training an existing model on a smaller, focused set of examples so it performs better on a particular task, format or style. It changes the model itself, unlike RAG, which adds information at the time of the question.

Generative AI#

AI that creates new content, such as text, images, audio, video or code, in response to an input called a prompt.

Grounding#

Tying a model's answer to specified, trusted sources, such as your own documents, so the answer can be checked against them. RAG is a common way to ground answers and reduce hallucination.

Hallucination#

When an AI model produces information that sounds plausible but is false, invented or not supported by its sources. Checking outputs and grounding answers reduce the risk.

Human in the loop#

A design in which a person reviews, approves or can override an AI system's output before it takes effect. The right level of human involvement depends on how much harm a wrong output could cause.

Inference#

Running a trained model to produce an output from new input, such as generating a reply to a prompt. Training builds the model; inference is using it.

ISO/IEC 42001#

An international standard for an AI management system: the policies, roles and processes an organisation uses to manage the AI it develops or uses. The current edition is ISO/IEC 42001:2023, and the AI Verify Testing Framework is mapped to it.

Large language model (LLM)#

A language model with a very large number of parameters, trained on large amounts of text, that can understand and generate human-like text. Chat assistants are built on LLMs.

Machine learning#

A branch of AI in which systems learn patterns from data instead of following only rules written by people. Most modern AI, including generative AI, is built this way.

Model AI Governance Framework#

Singapore's guidance for private sector organisations on key ethical and governance issues when deploying AI, first released by the PDPC on 23 January 2019, with a second edition on 21 January 2020. Its guiding principles are that decisions made by AI should be explainable, transparent and fair, and that AI systems should be human-centric.

Model Context Protocol (MCP)#

An open-source standard for connecting AI applications to external systems, such as data sources (for example local files and databases), tools and workflows. Its documentation compares it to a USB-C port for AI applications.

Parameters#

The internal numerical values a model learns during training, which together decide how it turns inputs into outputs. Model size is often described by its number of parameters.

Personal data#

Under Singapore's PDPA, data about an individual who can be identified from that data, or from that data and other information to which the organisation has or is likely to have access.

Personal Data Protection Act (PDPA)#

Singapore's law that provides a baseline standard of protection for personal data. It governs the collection, use, disclosure and care of personal data, and complements sector-specific laws such as the Banking Act and Insurance Act.

Prompt#

The instruction, question or content you give an AI model to get a response. Clear prompts state the task, the context and the format you want.

Prompt engineering#

The practice of designing and testing prompts so an AI model gives useful, reliable results, for example by adding context, examples, constraints and a required output format.

Retrieval-augmented generation (RAG)#

A technique in which a system first retrieves relevant passages from a knowledge source, such as your documents, and then has a language model generate an answer using them. It was described in a 2020 research paper, which found RAG models produced more specific, diverse and factual language than a comparable model relying only on its parameters.

Return on investment (ROI)#

A measure of the benefit gained from an investment compared with what it cost. For AI, benefits might include time saved or fewer errors, set against tool, training and staff costs.

Semantic search#

Search that matches by meaning rather than exact keywords, so a question about annual leave can find a passage about vacation days. It usually works by comparing embeddings.

Small language model (SLM)#

A language model with far fewer parameters than an LLM. It needs less computing power, can run on smaller hardware and is often suited to narrower tasks.

System prompt#

Standing instructions set by the builder of an AI app that shape every response, such as its role, tone, rules and limits. Users usually do not see it.

Temperature#

A setting that controls how much randomness a model uses when choosing its next token. Lower values give more predictable output; higher values give more varied output.

Token#

A small unit of text, such as a word or part of a word, that a language model reads and writes. Context limits and usage are often measured in tokens.

Tool calling#

A model's ability to ask for an external function or service, such as a search, calculator or database lookup, to be run and to use the result in its answer. The application runs the tool, not the model itself.

Training data#

The examples a model learns from during training, such as text, images or records. Its quality, coverage and balance shape what the model can do and where it may show bias.

Use case#

A specific task or problem where AI is applied, described by who uses it, what it does and how success will be measured.

Vector database#

A database designed to store embeddings and quickly find the ones closest in meaning to a query. It is a common building block of RAG systems.

Vibe coding#

Building or changing software by describing what you want in plain language and letting an AI-assisted tool write the code. You still test the result and stay accountable for it.

Workflow automation#

Using software to run a sequence of tasks automatically when something happens, such as saving form entries to a spreadsheet and sending a notification. AI steps, such as summarising or classifying text, can be added to these workflows.

Sources#

1. What is the Model Context Protocol (MCP)?, Model Context Protocol, accessed 8 November 2026
2. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks, arXiv, accessed 8 November 2026
3. PDPA Overview, Personal Data Protection Commission (PDPC), accessed 8 November 2026
4. Singapore’s Approach to AI Governance, Personal Data Protection Commission (PDPC), accessed 8 November 2026
5. What is AI Verify Testing Framework, AI Verify Foundation, accessed 8 November 2026

Frequently asked questions#

What is a large language model (LLM)?+

A large language model is a language model with a very large number of parameters, trained on large amounts of text, that can understand and generate human-like text. Chat assistants are built on LLMs.

What is retrieval-augmented generation (RAG)?+

RAG is a technique in which a system first retrieves relevant passages from a knowledge source, such as your documents, and then has a language model generate an answer using them. It is a common way to ground answers and reduce hallucination.

What is the Model Context Protocol (MCP)?+

MCP is an open-source standard for connecting AI applications to external systems, such as data sources, tools and workflows. Its documentation compares it to a USB-C port for AI applications.

What counts as personal data under the PDPA?+

Under Singapore's PDPA, personal data is data about an individual who can be identified from that data, or from that data and other information to which the organisation has or is likely to have access.

What is AI Verify?+

AI Verify is an AI governance testing framework and software toolkit developed by IMDA. It helps organisations assess their AI systems against 11 internationally recognised AI governance principles through technical tests and process checks.

Which AI terms should I learn first?+

It depends on your role. Everyday users should start with prompt, hallucination, personal data and AI use policy. Builders need RAG, embedding, tool calling, MCP and evaluation. Leaders need use case, ROI, change management and AI governance.