Plain language: An AI system that can take actions on your behalf, such as searching the web, sending emails and booking appointments, with varying degrees of independence.

Technical: An autonomous or semi-autonomous computational system capable of perceiving its environment, making decisions based on defined objectives, and executing multi-step actions. Agents may operate with a human in the loop or with minimal supervision, depending on design.

Misconception: AI agents are independent entities pursuing their own goals and will eventually act against human interests. They operate within parameters set by developers and users and do not have self-determined objectives.

Real-world example: An AI agent that checks your calendar, identifies a free slot, drafts a meeting invitation and sends it, completing a multi-step task without requiring you to do each step manually.

Related terms: AI · Generative AI · Assistive AI

Plain language: A trained system that has learned patterns from large amounts of data and uses those patterns to make predictions or generate outputs.

Technical: A mathematical structure derived from a training process in which parameters are adjusted to minimise error on a dataset. The model encodes learned statistical relationships and maps inputs to outputs at inference time.

Misconception: An AI model 'knows' things the way a human expert does. In practice, a model identifies and replicates statistical patterns in its training data, it does not understand or reason as humans do.

Real-world example: A medical imaging model trained on thousands of scans recognises patterns associated with specific conditions. When shown a new scan, it draws on those learned patterns to flag areas of concern for a clinician to review.

Related terms: AI · Generative AI · Algorithm · Hallucination

Plain language: A set of step-by-step rules or instructions that a computer follows to solve a particular problem or complete a task.

Technical: A finite, well-defined sequence of computational instructions used to process data, make decisions, or generate outputs. Algorithms are building blocks of all software, including AI systems.

Misconception: 'The algorithm' is used as a catch-all for all of AI, as though one all-knowing entity drives every AI outcome. In fact, an algorithm is a specific set of instructions for a specific task.

Real-world example: A social media platform's algorithm follows rules to decide which posts appear first, based on your past engagement. A chess-playing algorithm evaluates the best next move according to defined rules.

Related terms: AI · AI Model · Generative AI · Assistive AI · Machine Learning

Plain language: Technology enabling machines to perform tasks that typically require human thinking, such as recognising images, understanding language, making recommendations, or solving problems.

Technical: An umbrella field of computer science encompassing systems designed to simulate cognitive functions including learning, reasoning, pattern recognition and problem-solving. Covers both rule-based and machine learning approaches.

Misconception: AI is often understood as a single, unified system with its own intentions or desires. In reality, it is a broad category covering many distinct tools and approaches, none of which thinks, decides or wants anything.

Real-world example: When your email filters spam, when Google Maps reroutes you, or when a streaming service recommends a film, AI is at work, and none of these require human intervention in the moment.

Related terms: Algorithm · Generative AI · Assistive AI · AI Model · Intelligence (concept)

Plain language: A generative AI tool developed by OpenAI that generates text responses to user prompts. Currently the most widely recognised example of generative AI among the general public.

Technical: A conversational interface built on OpenAI's large language models (GPT-series), enabling users to interact with a generative AI system via natural language prompts across a wide range of tasks.

Misconception: 'ChatGPT' is frequently used as a synonym for all AI, in the same way 'Hoover' became a synonym for all vacuum cleaners. ChatGPT is one specific commercial tool within the much broader category of generative AI.

Real-world example: Students using ChatGPT to draft an essay outline; a professional using it to summarise a lengthy document; a parent using it to help draft a complaint letter.

Related terms: Generative AI · Prompt · Hallucination

Plain language: Microsoft's AI assistant, integrated into tools like Word, Teams and Outlook. Also functions as a useful conceptual metaphor: AI as a co-pilot that assists rather than a pilot that takes control.

Technical: A suite of generative AI capabilities developed by Microsoft, integrating large language model functionality into productivity software. The copilot metaphor, AI assisting rather than replacing human decision-making, has broader currency in AI communication beyond the branded product.

Misconception: The word 'copilot' can imply the AI is making decisions independently. The metaphor is chosen specifically to convey the opposite: the human remains in control; the AI assists.

Real-world example: Using Copilot in Microsoft Teams to produce a summary of a meeting you missed; using it in Word to suggest edits to a draft document while you retain final editorial control.

Related terms: Generative AI · Assistive AI · Prompt

Plain language: AI built into everyday tools to help with specific tasks, without creating new content from scratch. It works within defined boundaries.

Technical: Non-generative AI systems operating within constrained parameters to perform classification, prediction, recommendation or automation functions. Unlike generative models, these systems produce bounded, task-specific results rather than open-ended novel outputs.

Misconception: Because it doesn't generate text or images, embedded AI is often not recognised as AI at all, or is assumed to be simpler and therefore less significant than generative AI. In fact, it underpins much of our daily digital experience.

Real-world example: Google Maps rerouting a journey in real time; Spotify recommending a playlist; email spam filters; blood sample analysis software that flags patterns for a clinician to review before any human decision is made.

Related terms: AI · Assistive AI · AI Agent · Generative AI · Algorithm

Plain language: When an AI generates information that sounds confident and plausible but is factually wrong or entirely made up.

Technical: A failure mode in generative AI in which the model produces outputs that are statistically coherent, fitting learned patterns of language, but factually inaccurate, unsupported or fabricated. Commonly manifests as invented citations, non-existent sources, or false factual claims presented with apparent confidence.

Misconception: Because the word is borrowed from psychology, people may assume the AI is experiencing something perception-like, or that using AI could affect the user's own mental state. The term refers only to a specific technical error pattern in text generation.

Real-world example: An AI chatbot confidently producing a list of academic references, all with plausible author names, journals and dates, none of which actually exist. Markers encountering student submissions may find such fabricated reference lists.

Related terms: Generative AI · AI Model · Prompt

Plain language: A specialised vocabulary, a curated set of words and definitions for a particular field or subject area, distinct from a general dictionary.

Technical: A domain-specific vocabulary that organises terminology within a particular discipline or community of practice. Unlike a general dictionary, a lexicon is bounded by subject area and typically includes contextual guidance on usage.

Misconception: A lexicon is the same as a dictionary. A dictionary covers general vocabulary of a language; a lexicon is specific to one domain and reflects the conventions of a professional or academic community.

Real-world example: A medical lexicon contains terms like 'tachycardia' and 'myocardial infarction'. This AI lexicon contains terms like 'prompt' and 'hallucination', words with specific meanings in AI contexts that differ from everyday use.

Related terms: AI · Prompt · Algorithm

Plain language: The instruction, question or input you give to an AI system to tell it what you want it to do.

Technical: The input text (or multimodal input) that conditions a generative model's output. The craft of designing effective prompts is known as prompt engineering, and can substantially influence response quality, tone, accuracy and format.

Misconception: A prompt is just a search query. Unlike a keyword-optimised search, a prompt to a generative AI system works more like an instruction to a capable assistant. How it is worded and structured significantly shapes what comes back.

Real-world example: 'Summarise this article in three bullet points for a general audience' is a prompt. 'Write a formal letter declining a job offer, in a warm but professional tone' is also a prompt. Both are instructions that shape the AI's response.

Related terms: Generative AI · Hallucination · AI Model