AI terminology: 50 key AI terms you should know

Core AI terminology: the basics

• Artificial intelligence (AI)Computer systems that carry out tasks that normally need human intelligence, such as understanding language, recognising images or making predictions. The UK Government describes AI products and services as "adaptable" and "autonomous".
• Artificial general intelligence (AGI)A theoretical form of AI that could match or exceed human performance across most cognitive tasks. No system has reached it yet. For now, it remains a long-term goal for some AI developers.
• Frontier AIHighly capable general-purpose AI models that can perform a wide variety of tasks. In practice, the term refers to the most advanced models available at any given time.
• Machine learning (ML)A type of AI that lets a system learn and improve from examples instead of following rules written by a programmer. Our look at how AI and machine learning are reshaping custom software development covers what it means for software projects.
• Neural networkA computer structure inspired by the biological brain, made of layers of connected nodes that adjust as the system learns. In turn, neural networks are the building blocks of deep learning.
• Deep learningA branch of machine learning that uses neural networks with many layers to analyse complex patterns in data. As a result, it underpins most modern image, speech and language AI.
• Cognitive computingAI systems that simulate human thought processes to support decision-making. The term was popularised in the 2010s, notably by IBM for its Watson systems. Today, however, most people simply say AI.

AI models, language and vision

• Foundation modelA machine learning model trained on a vast amount of data so that it can be adapted to many different tasks. For example, large language models are one type.
• Large language models (LLMs)AI models trained on vast amounts of text to understand and generate human-like language. In practice, they are the technology behind chat tools such as ChatGPT, Gemini and Copilot.
• Natural language processing (NLP)AI's ability to understand and generate human language, powering chatbots, voice assistants and translation tools.
• Computer visionA field of AI that enables computers to interpret and process visual data, such as images and videos.

AI terminology for how models learn

• Reinforcement learningA machine learning technique where an AI system learns by trial and error, receiving rewards for actions that move it towards a goal.
• Unsupervised learningA type of machine learning where algorithms identify patterns in data without labelled examples. In contrast, supervised learning trains on data that has been labelled.
• Transfer learningA method where a model trained on one task is adapted for another, saving time and data compared with training from scratch. Fine-tuning is a common form.
• Fine-tuningTraining an existing model further on a specific set of data, such as a company's own documents. The aim is better performance for a particular use.
• Zero-shot learningA model's ability to carry out a task it was not specifically trained for, by drawing on what it learned in general.
• Federated learningA collaborative way of training a model across many devices or organisations without moving the raw data to one place. As a result, it helps preserve privacy.
• AutoML (automated machine learning)Tools that automate steps such as choosing and tuning a model, so non-experts can apply machine learning to real-world problems.
• Synthetic dataData that is generated artificially rather than collected from real-world events. For example, it can replace real data that is scarce or too sensitive to use in training.

Generative AI terminology

• Generative AIAI that creates new text, images, audio, video or other media in response to a prompt. For instance, large language models are one kind, as we explain in generative AI vs LLM.
• Diffusion modelsGenerative models that learn by adding noise to data and then reversing the process. This lets them create new, high-quality images from noise.
• Generative adversarial networks (GANs)AI models made of two networks, a generator and a discriminator. The generator creates synthetic data, such as images, while the discriminator judges whether it looks real.
• InferenceThe stage where a trained model is put to work, producing predictions or outputs from new data. For businesses using AI services, it is usually what the running cost is based on.
• HallucinationWhen an AI model produces output that sounds plausible but is false or nonsensical. Consequently, reducing hallucinations is a priority for improving AI accuracy.

Prompts, tokens and RAG: how AI chat tools work

• Prompt and prompt engineeringA prompt is the instruction or question you give an AI model. Prompt engineering is the practice of writing and structuring prompts to get more accurate and useful results.
• TokensThe small units of text, such as words or parts of words, that language models read and write. AI services usually measure usage, limits and pricing in tokens.
• Context windowThe amount of text, measured in tokens, that a model can take into account at once. It includes your prompt, any documents you supply and the model's own reply.
• Multimodal AIAI that can understand or produce more than one type of content, such as text, images, audio and video.
• Reasoning modelsModels designed to work through a problem step by step before answering, a process often called chain of thought. They tend to do better on maths, logic and coding tasks, but they are slower and cost more per answer.
• Retrieval-augmented generation (RAG)A technique that finds relevant information in a set of documents, such as a company's own files. It then passes that information to the model alongside the prompt. As a result, answers are grounded in that information, which can reduce hallucinations.

AI terminology for business

• AI agents (agentic AI)AI systems that plan and carry out multi-step tasks towards a goal with limited human input. They use tools such as search, email or business software. Our guide to AI agents for business explains how they work, and multi-agent solutions cover agents working together.
• AI copilotAn AI assistant built into software to help with tasks such as drafting, summarising or coding. The person using it stays in control. Microsoft uses Copilot as the brand name for its assistant.
• Model Context Protocol (MCP)An open standard, introduced by Anthropic, for connecting AI assistants to a business's existing tools and data. Similarly, Google's WebMCP brings the idea to websites, as we explain in our look at Google WebMCP.
• Virtual agentsAI-powered systems that hold conversations with people, commonly used in customer support and sales.
• Vibe codingBuilding software by describing what you want to an AI tool and accepting the code it writes. Often, nobody reviews that code line by line. Our piece on agentic engineering and vibe coding compares it with a more controlled approach.
• Workflow automationSoftware that carries out a sequence of business steps automatically. Most of it follows fixed rules rather than AI. That often makes it the cheaper option, as we explain in AI vs automation.
• Low-code and no-code AIPlatforms that let people build AI-driven applications without extensive programming knowledge.

Data and infrastructure: AI terminology

• Predictive analyticsUsing historical data with statistical and machine learning models to forecast future outcomes.
• Big data analyticsExamining very large or complex datasets to uncover patterns and trends. However, AI is only one way to do it, alongside traditional statistics and database tools.
• MLOps (machine learning operations)A set of practices for deploying, monitoring and maintaining machine learning models once they are in use.
• Edge AIAI processing that happens directly on devices, such as phones or sensors, rather than in the cloud. As a result, responses are faster and data stays local.
• Sovereign AIAI that a country or organisation controls itself, with its data, models and infrastructure kept within its own jurisdiction. For a business, it often means running a private model. We cover this in when a private LLM beats an API.
• Quantum AIThe combination of quantum computing and AI to tackle problems too complex for classical computers. For now, though, it is still largely at the research stage.

AI terminology for trust and risk

• Bias in AISystematic errors that make an AI system treat some people or groups unfairly. It can come from unrepresentative training data or from choices made in designing the system. Our piece on the ethics of AI covers how to address it.

• Explainable AI (XAI)AI systems designed to give clear, understandable explanations for their decisions, which supports trust and accountability. It overlaps with what is sometimes called interpretability or transparency. We look at why it matters in AI transparency is no longer optional.
• DeepfakeAI-generated or AI-altered images, video or audio that realistically show someone saying or doing something they did not. They are also one reason for the push to label AI-generated content. We cover that in our piece on AI content labelling.
• AI washingOverstating or misrepresenting how much a product or service uses AI. For example, in March 2024 the US Securities and Exchange Commission fined two investment advisers a combined $400,000. Both had made misleading claims about their use of AI.

Governance and responsible AI use

• Human in the loopAn approach where people review, correct or approve an AI system's outputs at key points. In other words, the AI does not act alone. It is a common safeguard where AI decisions affect customers.
• AI governanceThe policies, roles and controls an organisation uses to make sure its AI is used safely, legally and as intended. For AI agents in particular, see our guide to AI agent governance.
• AI literacyAn understanding of what AI can and cannot do, and how to use it responsibly. The EU AI Act requires AI providers and deployers to take steps to build AI literacy among their staff. This can also apply to UK firms whose AI is used in the EU.

Why AI terminology matters for your business

These 50 AI definitions cover the opportunities and challenges that AI brings to businesses and individuals alike. Knowing them will not make anyone an AI engineer. However, it helps you separate what a tool actually does from how it is marketed.

Building your AI vocabulary is less about being tech-savvy and more about preparing your organisation for what comes next. Whether you are looking at workflow automation, virtual agents or predictive analytics, start with the problem you want to solve. Our look at the AI adoption challenges UK business leaders raised sets out what most often holds businesses back.

Frequently asked questions

Sources

• Artificial intelligence (AI) glossary, Parliamentary Office of Science and Technology
• Data science and AI glossary, The Alan Turing Institute
• Artificial intelligence in UK businesses: 2023 to 2026, Office for National Statistics, 20 July 2026
• Introducing the Model Context Protocol, Anthropic
• Regulation (EU) 2024/1689 (the EU AI Act), EUR-Lex
• SEC charges two investment advisers with making false and misleading statements about their use of artificial intelligence, US Securities and Exchange Commission, 18 March 2024