The AI jargon buster.

Artificial Intelligence (AI)

Software that performs tasks normally requiring human intelligence — understanding language, spotting patterns or making decisions.

Machine Learning (ML)

A branch of AI where systems learn patterns from data and improve with experience, rather than being explicitly programmed with rules.

Deep Learning

Machine learning that uses many-layered neural networks to learn complex patterns — the engine behind most modern image and language AI.

Neural Network

A model loosely inspired by the brain, made of interconnected "nodes" that process data in layers to recognise patterns.

Large Language Model (LLM)

An AI model trained on vast amounts of text to understand and generate human-like language — the technology behind tools like ChatGPT.

Generative AI

AI that creates new content — text, images, code, audio or video — rather than only analysing or classifying existing data.

Foundation Model

A large, general-purpose model trained on broad data that can be adapted to many tasks — the "foundation" for specialised applications.

Frontier Model

The most advanced, largest-scale models at the cutting edge of capability — typically from leading labs and under the greatest safety scrutiny.

Transformer

The neural-network architecture behind most modern language models, designed to weigh the importance of different words in context.

Parameters

The internal values a model learns during training. More parameters generally means more capacity to capture complex patterns — often billions of them.

AGI (Artificial General Intelligence)

A hypothetical future AI that could perform any intellectual task a human can. Today's AI remains "narrow" — excellent at specific tasks only.

Prompt

The instruction or question you give an AI model to get a response.

Prompt Engineering

The practice of crafting and refining prompts to get more accurate, reliable or useful results from an AI model.

Token

The small chunk of text (roughly a word or part of a word) that models read and generate. Usage and limits are usually measured in tokens.

Context Window

How much text a model can consider at once — its short-term "memory" for a given conversation or document.

Temperature

A setting that controls how creative or predictable output is. Lower is more focused and consistent; higher is more varied.

Inference

The act of running a trained model to produce an answer — the "using" phase, as opposed to training.

Fine-tuning

Further training a general model on your own examples so it performs better on a specific task, tone or domain.

RAG (Retrieval-Augmented Generation)

A technique that lets a model pull in relevant information from your own documents before answering — improving accuracy and reducing guesswork.

Embedding

A numerical representation of text or images that captures meaning, letting software compare and search by similarity rather than exact words.

Vector Database

A store designed to hold embeddings and find the most similar items quickly — the backbone of semantic search and RAG systems.

Multimodal AI

Models that work across more than one type of input or output — for example text, images, audio and video together.

Chain-of-Thought

Prompting or model behaviour that works through a problem step by step, often improving reasoning on complex tasks.

AI Agent

An AI system that takes actions to achieve a goal — using tools, calling systems and making decisions — rather than just answering one question.

Agentic AI

AI designed to operate with a degree of autonomy: planning, deciding and acting across multiple steps with limited human intervention.

Agentic Orchestration

Coordinating multiple AI agents and tools into reliable workflows, with checkpoints and controls so the whole system stays safe and accountable.

Model Context Protocol (MCP)

An open standard for connecting AI models to external tools and data sources in a consistent, governable way.

Human-in-the-Loop (HITL)

Designing AI processes so a person reviews, approves or can override key decisions — keeping accountability with humans.

Hallucination

When a model produces confident but incorrect or fabricated information — a key risk to manage in any deployment.

Prompt Injection

A security attack where hidden or malicious instructions trick an AI system into behaving in unintended ways.

Shadow AI

Unsanctioned use of AI tools by staff outside official policy, creating hidden security, compliance and data risks.

Model Drift

The gradual decline in a model's accuracy over time as the real world changes and moves away from its training data.

Bias

Systematic unfairness in AI outputs, often reflecting imbalances in the data the model learned from.

AI Governance

The policies, controls and accountability structures that ensure AI is used safely, legally, ethically and in line with business goals.

Governance as Code

Turning AI governance policies into automated, enforceable technical controls embedded directly in delivery pipelines.

Responsible AI

An approach to designing and using AI that prioritises fairness, transparency, accountability, privacy and safety.

Guardrails

Technical and policy controls that constrain what an AI system can do or say — keeping outputs safe, on-brand and compliant.

AI Assurance

Independent checks, evidence and testing that give leaders confidence an AI system does what it should — and nothing it shouldn't.

Red Teaming

Deliberately stress-testing an AI system by trying to make it fail or misbehave, to find and fix weaknesses before attackers do.

Explainability (XAI)

The ability to understand and communicate why an AI system produced a particular output — essential for trust and regulation.

Alignment

The challenge of ensuring an AI system's goals and behaviour match human intentions and values.

Sovereign AI

An organisation's or nation's ability to build and run AI using its own infrastructure, data and controls, reducing dependence on external providers.