AI Glossary

Plain-language definitions of the AI and technology terms you are most likely to encounter - accurate, neutral and jargon-free.

Foundations

Algorithm: An algorithm is a precise, step-by-step set of instructions for solving a problem or completing a task. In computing, algorithms tell a machine exactly how to transform inputs into outputs. Modern AI systems are built from many algorithms working together, including those used to learn patterns from data.
Artificial Intelligence: Artificial Intelligence (AI) is a broad field of computer science focused on building systems that perform tasks normally associated with human intelligence, such as understanding language, recognising images, or making decisions. AI ranges from simple rule-based programs to systems that learn patterns from large amounts of data. It is a set of techniques and tools, not a single technology.
Classification: Classification is a machine learning task where a system assigns an input to one of several predefined categories. Examples include marking an email as spam or not spam, or sorting a support message by topic. The model learns from labelled examples and then predicts the most likely category for new inputs.
Deep Learning: Deep learning is a branch of machine learning that uses neural networks with many layers to learn complex patterns from large datasets. The “depth” refers to these stacked layers, which allow the system to build up increasingly abstract representations of the data. It underpins many recent advances in language, image, and speech systems.
Generative AI: Generative AI refers to systems that can produce new content such as text, images, audio, or code, rather than only classifying or predicting. These systems learn patterns from large amounts of data and then generate outputs that follow those patterns in response to a prompt. The results are created on the fly and can vary between requests.
Machine Learning: Machine learning is a branch of AI where systems learn patterns from data rather than being explicitly programmed with fixed rules. By processing many examples, a model improves its ability to make predictions or decisions on new data. It is the foundation for most modern AI applications.
Model: In AI, a model is the result of training a system on data so it can make predictions or generate outputs. It captures the patterns learned during training as a set of internal values, sometimes called parameters. The same model can be reused across many tasks once it has been trained.
Neural Network: A neural network is a type of machine learning model loosely inspired by how neurons connect in the brain. It is made up of layers of simple units that pass signals to one another, adjusting their connections during training to recognise patterns. Neural networks are the building blocks behind most modern AI systems.
Supervised Learning: Supervised learning is a machine learning approach where a model learns from examples that are already labelled with the correct answer. By comparing its predictions to the known labels, the model gradually improves. It is commonly used for tasks like classification and prediction where good labelled data is available.

Models

Computer Vision: Computer vision is the area of AI concerned with helping computers interpret images and video. It powers tasks such as detecting objects, reading text from documents, and recognising patterns in visual data. The goal is to turn pixels into useful, structured information that software can act on.
Context Window: The context window is the amount of text a language model can consider at one time, measured in tokens. It includes both the input you provide and the response being generated. If a conversation or document exceeds the context window, earlier content may be left out, which is why managing what goes into the window matters.
Large Language Model (LLM): A large language model (LLM) is an AI system trained on very large amounts of text to predict and generate language. It can answer questions, summarise, translate, and draft content by recognising patterns in how words and ideas fit together. LLMs do not understand the world as people do, so their outputs benefit from review for accuracy.
Natural Language Processing (NLP): Natural Language Processing (NLP) is the field of AI focused on helping computers understand, interpret, and generate human language. It covers tasks such as translation, sentiment analysis, summarisation, and answering questions. Modern NLP relies heavily on models trained on large text collections.
Token: A token is a small unit of text, such as a word or part of a word, that a language model processes. Models read and generate text one token at a time, and limits on input length and pricing are often measured in tokens. Roughly speaking, a token is a few characters of English text.
Transformer: A transformer is a type of neural network architecture that has become the foundation for most modern language models. It uses a mechanism called attention to weigh how different parts of the input relate to one another. This design lets transformers handle long sequences of text effectively and learn rich patterns in language.
Vision Model: A vision model is an AI model trained to work with images, and sometimes video, for tasks such as recognising objects, reading text, or describing a scene. Some vision models are multimodal, meaning they can handle both images and text together. They are widely used in document processing, quality checks, and visual search.

Data

Embedding: An embedding is a way of representing text, images, or other data as a list of numbers that captures its meaning. Items with similar meaning end up with similar number patterns, which lets software measure how related two things are. Embeddings are central to search, recommendations, and retrieval systems.
Training Data: Training data is the collection of examples a model learns from during its training stage. Its quality, coverage, and balance strongly shape how the model behaves and where it may fall short. Because models reflect their training data, choosing and curating that data carefully is an important part of responsible AI.
Vector Database: A vector database is a system designed to store and search embeddings, which represent data as lists of numbers. It can quickly find items whose meaning is similar to a query, rather than matching exact words. Vector databases are a common component in search and retrieval-augmented generation systems.

Engineering

API: An API (Application Programming Interface) is a defined way for two software systems to talk to each other. It lets one program request data or services from another without needing to know how that other system works internally. Many AI capabilities are offered as APIs, so applications can send a request and receive a model’s response.
Fine-tuning: Fine-tuning is the process of taking an existing trained model and training it further on a smaller, focused dataset so it performs better on a specific task or style. It adapts general capabilities to a particular domain without building a model from scratch. Fine-tuning requires good-quality examples and careful evaluation to avoid introducing errors.
Inference: Inference is the stage where a trained model is actually used to produce an output from new input, as opposed to the training stage where it learns. When you send a prompt and receive a response, that is inference at work. Inference speed and cost are practical factors when running AI systems at scale.
Latency: Latency is the delay between making a request and receiving a response. For AI applications, it usually means how long it takes a model to start or finish producing an answer. Lower latency makes systems feel more responsive, which matters for interactive uses such as chat and live assistance.
Prompt: A prompt is the input or instruction you give to a generative AI model to tell it what you want. It can be a question, a command, or a longer set of context and examples. The clarity and detail of a prompt strongly influence the quality and relevance of the response.
Prompt Engineering: Prompt engineering is the practice of designing and refining prompts to get more accurate, useful, or consistent results from a generative AI model. It includes techniques such as giving clear instructions, providing examples, and structuring the request. Good prompt engineering can significantly improve outputs without changing the underlying model.
RAG (Retrieval-Augmented Generation): Retrieval-Augmented Generation (RAG) is a technique that lets a language model use external, up-to-date information when answering. Relevant documents are retrieved from a knowledge source and supplied to the model alongside the question, so its response is grounded in that material. RAG helps reduce hallucinations and keeps answers tied to trusted content.
Structured Output: Structured output is when an AI system returns its response in a defined, machine-readable format such as JSON or a fixed set of fields. This makes the output easier for other software to process reliably, rather than parsing free-form text. It is widely used when AI results feed into automated workflows or databases.

Safety & Oversight

Bias (in AI): Bias in AI refers to systematic and often unfair patterns in a system’s outputs, frequently caused by imbalances or assumptions in the data it learned from. Left unchecked, bias can lead to outcomes that disadvantage certain groups or misrepresent reality. Responsible practice involves testing for bias, documenting limitations, and keeping humans involved in sensitive decisions.
Guardrails: Guardrails are the controls and checks placed around an AI system to keep its behaviour within acceptable, intended limits. They can include filtering unsafe content, restricting what tools a system may use, and validating outputs before they are acted on. Guardrails work best alongside human oversight rather than as a complete substitute for it.
Hallucination: A hallucination is when an AI system produces information that sounds plausible but is incorrect or fabricated. This can happen because generative models predict likely text rather than verifying facts. Because hallucinations can be convincing, important outputs should be checked against reliable sources and reviewed by a person.
Human-in-the-loop: Human-in-the-loop describes workflows where a person reviews, approves, or corrects an AI system’s output before it has an effect. This keeps accountability and judgement with people, especially for decisions that carry risk or affect individuals. It is a core part of responsible AI design, combining the speed of automation with human oversight.

Applications

Agent (AI agent): An AI agent is a software system that can take a goal, break it into steps, and use tools or actions to work towards that goal with limited human input. Agents typically combine a language model for reasoning with access to tools such as search, calculators, or business systems. Because they can act, not just answer, agents are usually deployed with clear boundaries and human oversight.
Chatbot: A chatbot is a software application that holds a conversation with people through text or speech. Earlier chatbots followed fixed scripts and rules, while modern ones often use language models to understand questions and generate more natural replies. They are commonly used for customer support, information lookup, and guiding users through simple tasks.
Workflow Automation: Workflow automation is the use of software to carry out routine, repeatable business steps with little manual effort. When combined with AI, automated workflows can include tasks such as drafting, classifying, or routing information, often with human review at key points. The aim is to save time and reduce errors while keeping people in control of important decisions.

Concepts, connected

A map, not just a list

AI vocabulary is easier to trust when you can see how the pieces relate. We define each term plainly and in context, so the ideas connect rather than pile up.

No hype, no buzzword inflation - just accurate, neutral explanations aimed at business readers who want to make informed decisions.

Plain-language AI

New terms, explained without the hype.

We occasionally publish clear write-ups of the AI concepts that actually matter for business decisions. Subscribe for the next one.

Want this applied to your situation?

Understanding the terms is one thing; choosing the right approach for your goals is another. We are happy to talk it through honestly.

Book a Discovery Call Explore Services