The AI glossary.

An AI agent is a system designed to carry out tasks with some level of autonomy. It may plan steps, use tools, search the web, write files, call APIs or complete workflows rather than only answering a single question.

AI governance is the framework a business uses to manage AI safely and effectively. It can include policies, approved tools, data rules, review processes, risk management, staff training and compliance requirements.

An application programming interface is a set of rules that lets one software system connect with another. AI APIs allow websites, apps and workflows to use AI features such as text generation, image analysis or speech transcription.

Artificial intelligence is the broad field of creating computer systems that can perform tasks associated with human intelligence, such as understanding language, recognising patterns, making predictions, solving problems and generating content.

Automation is the use of software or systems to complete tasks with minimal manual effort. In an AI context, automation may involve generating reports, sorting enquiries, drafting content or triggering workflows.

Bias in AI happens when a system reflects unfair patterns from its training data, design or use. It can lead to inaccurate, exclusionary or discriminatory outcomes if not tested and managed carefully.

A chatbot is a conversational tool that can answer questions, provide support or guide users through tasks. Modern AI chatbots often use large language models to create more natural and flexible responses.

Computer vision is a field of AI that enables systems to interpret visual information. It can be used for image recognition, object detection, facial recognition, quality control, medical imaging and visual search.

A context window is the limit on how much text, data or conversation history an AI model can process in a single interaction. A larger context window allows the model to work with longer documents and more background information.

Data privacy is the practice of handling personal, business or sensitive data responsibly. With AI tools, this includes understanding what data is shared, how it is stored, whether it trains models and who can access it.

Deep learning is a branch of machine learning that uses multi-layered neural networks to process complex data. It powers many modern AI systems, including image recognition, speech recognition and large language models.

An embedding converts text, images or other data into numbers that represent meaning or similarity. Embeddings help AI systems search, compare and organise information based on concepts rather than exact wording.

Fine-tuning is the process of taking an existing AI model and training it further on a more specific dataset. This can make the model better at a particular task, tone, domain or type of output.

A foundation model is a large AI model trained on broad data that can be adapted to many different uses. Large language models and image generation models are common examples.

Generative AI refers to AI systems that can create new outputs such as text, images, video, audio, code or designs. Instead of only analysing data, it produces something new based on patterns it has learned.

A graphics processing unit is a type of computer chip originally designed for graphics and video processing. GPUs are widely used in AI because they can handle many calculations in parallel, making them useful for training and running large AI models.

A hallucination happens when an AI system produces an answer that sounds plausible but is wrong, invented or unsupported by the available evidence. This is why important AI outputs should be checked.

Human in the loop means a human remains involved in an AI workflow, usually to review, approve, correct or guide the output. It is important when accuracy, brand voice, ethics or judgement matter.

Inference is the process of using a trained AI model to produce an output. When you ask a chatbot a question and it replies, the model is performing inference.

A large language model is an AI system trained on huge amounts of text so it can understand, summarise, translate, reason and generate language. ChatGPT, Claude and Gemini are examples of products powered by LLMs.

Machine learning is a type of AI where systems improve at a task by learning from data rather than being programmed with every rule manually. It is commonly used for predictions, recommendations, classification and pattern recognition.

A model is the trained system that processes inputs and produces outputs. Different AI models are designed for different jobs, such as writing text, generating images, transcribing audio or predicting behaviour.

Multimodal AI can process and generate different types of information, such as text, images, audio, video and files. For example, a multimodal assistant might read a document, analyse a chart and explain an image in one workflow.

Natural language processing is the area of AI focused on language tasks such as summarisation, translation, sentiment analysis, chatbots, speech-to-text and text generation.

A neural network is a computing model made up of connected layers that process information. It learns patterns by adjusting internal weights, making it useful for recognising images, understanding language and making predictions.

A prompt is the message, instruction, question or context given to an AI tool to guide its response. Better prompts usually include the goal, background, constraints, audience and desired format.

Prompt engineering is the practice of writing, testing and refining prompts so an AI system produces more useful, accurate and relevant responses. It often involves giving context, examples, structure and clear success criteria.

Retrieval augmented generation combines an AI model with a search or knowledge retrieval system. The AI first pulls relevant information from documents, databases or websites, then uses that information to generate a more grounded response.

A token is a chunk of text used by AI models to read and generate language. It may be a word, part of a word or punctuation. Model limits and pricing are often based on token usage.

Training data is the information used to build an AI model. The model learns patterns from this data, which can include text, images, audio, code or structured datasets depending on the type of system.

A vector database stores embeddings so information can be searched by similarity. It is often used in AI systems that need to find relevant documents, FAQs, products or knowledge before generating an answer.