🎓AI Glossary For Beginners

With AI seemingly everywhere, I created this handy glossary to help understand in simple terms what is going on! Last updated December 2025.

Agentic AIAgentic AI refers to how independent a system is - demonstrating autonomy, initiative and problem solving, adapting as it goes without needing input, makes a system Agentic.

Agent WashingWhen companies or products are described as “AI agents” (implying they’re smart, independent systems) but in reality they’re just an automation, simple tools or chatbots. Example: A basic customer support bot being marketed as a full AI agent.

AI (Artificial Intelligence)When computers or machines do tasks that usually need human intelligence, like learning, understanding language, or solving problems. Examples: ChatGPT, Siri, Alexa.

AI AgentAn autonomous system that makes decisions and takes actions on its own to achieve a desired output, rather than needing constant instructions at each stage. Example: a restaurant booking app, that finds the best eatery, knows when you are available, books the table, invites the guests and emails you confirmation.

AI LiteracyThe basic understanding of what AI is, how it works, and what it can and can’t do - you don't need to be a tech expert. Example: Knowing that ChatGPT might “hallucinate” and double-checking facts before using them in a report.

Artificial General Intelligence (AGI)A type of AI with the ability to understand, learn, and apply knowledge across a wide range of tasks at a human level or beyond. Unlike narrow AI, which is designed for specific tasks, AGI can reason, solve problems, and adapt in unfamiliar situations without being retrained. Contrast with Narrow AI below. Examples: None! This is aspirational/hypothetical at the moment.

Artificial Narrow Intelligence (Narrow AI)The type of AI we have today - contrast with AGI above. AI can perform a specific task or limited range of tasks extremely well - but not go beyond that. Examples: Spam filter, chess-playing program. They do what they say on the tin - but no more.

AlgorithmA set of step-by-step instructions a computer follows to solve a problem, like a recipe for a machine.

Algorithmic FairnessMaking sure AI systems make decisions that are fair and don’t discriminate against people.

API (Application Programming Interface)A tool that lets different computer programs talk to each other and share information.

AutomationAs it suggests, this is using computers and technology to perform tasks automatically, with little or no human input. Typically for routine or repetitive tasks. Example: Email sent in response to a form being completed.

Bias in AIWhen an AI system makes unfair or unbalanced decisions, often because of bias or flaws in the data it is trained on. Example: Its hard to generate an image of a watch or clock telling any time other than ten past ten because that is what the majority of images of watches look like.

Big DataVery large collections of information that are too big for normal computers to handle easily.Example: Social media posts, online shopping data.

ChatbotA computer program that can talk to people, usually by text or voice, to answer questions or help with tasks. Examples: ChatGPT, Gemini, customer service bots.

Chain-of-Thought ReasoningUseful for complex tasks, logic or multi-step tasks like strategies and workflows, asking to see the working or "chain-of-thought" in an output lets you check the steps to get to the right answer. Example: "Explain your reasoning before giving the answer".

CopilotA smart virtual assistant that helps users with tasks, decision making and productivity by providing real-time suggestions and automation. Unlike basic chatbots, it understands context. Example: Microsoft Copilot, app-specific copilots for specific software

Computer VisionAI that helps computers “see” and understand images and videos. Examples: Face recognition on phones, self-driving cars.

ConnectorsBuilt-in links that let an AI tool work directly with your apps so it can perform further steps. Example: Instead of just drafting text, it can send the email, update the task, or pull in your calendar.

Context WindowThe maximum amount of text (or data) that an AI model can "hold" at one time when generating responses. Its actually quite a lot, so you can include lengthy documents. Example: ChatGPT's is currently 128,000 tokens (depending on your plan).

Data MiningFinding useful patterns or information in large amounts of data.

Data Facts and information (like numbers, words, or pictures) that computers use to learn, make decisions, or solve problems.

Data LakeA huge storage pool of raw, unfiltered, unstructured data, which is then fished out and used by AI - rather than looking at everything in the entire world, ever. Example: All of a company's data - documents, images, logs, calls, emails - in an unstructured, unindexed format.

DeepfakeAI-generated fake videos or audio that seem real. These can be amusing or satirical, but also extremely concerning use of AI to deceive through fraud, scams and misinformation. Often produced using GANs. Examples: The Pope in a puffer jacket, Keir Starmer "swearing at staff" audio clip

Deep LearningA complex way computers learn from large neural networks with many layers to learn and recognise sophisticated patterns from data. Example: Voice assistants and image recognition.

Edge AI / Edge ComputingRunning AI on local devices (like smartphones or cameras) instead of sending data to the cloud, so things happen faster and more privately.

Ethics Moral rules about what is right or wrong when developing or using AI, such as protecting privacy or avoiding unfairness.

Explainable AIAI that can show or explain why it made a certain decision, instead of just giving a result with no reasoning.

Fine-tuning LLMMaking small adjustments to a large language model (LLM) so it works better for a specific task or company.

Foundation ModelAn extremely powerful general-purpose model, trained on vast amounts of data, not to do one task, but to form the foundation for many tasks across different areas. They can be text only (LLM) or also handle other inputs such as images, videos, audio (multi-modal).

GANs (Generative Adversarial Networks)A type of AI where two computer models compete: one creates fake data, for example like a fake image, and the other tries to spot the fake. The competition between the two creates better and better fakes, until it produces convincing output. Example: Creating realistic photos of people who don’t exist, AI art, Deepfakes.

Generative AI (GenAI)AI that can create new content, such as text, images, music, or videos, which match what humans produce. Examples: ChatGPT (text), Midjourney (images), HeyGen (Video)

GPT (Generative Pre-trained Transformer)A type of large language model (see LLM) that can understand and generate human-like text. Example: ChatGPT is based on GPT.

GuardrailsThe safety rules built into AI systems to stop them from producing harmful, unsafe, or inappropriate results Example: An AI refusing to give instructions for making dangerous substances

Hallucinations When AI gives an answer that sounds correct but is actually made up or wrong. Check out our guide on this here:

Human-in-the-loop (HITL)The principle of keeping a human involved in the process, so the AI component works as an assistant, rather than being autonomous. Think of fraud detection - the AI might flag a suspicious activity, but the human then decides and acts on that information.

Human-on-the-loop (HOTL) The principal of having optional human input - processes see the AI act autonomously, but a human can intervene. Think of a self-driving car - the driver can intervene if things get dangerous.

Human-out-of-the-loop (HOOTL) The principal of having a fully autonomous AI process that runs from beginning to end like an AI Agent. Typically used in low-risk processes, such as approving invoices which are the same each month.

Humanity's Last Exam (HLE)A test designed to see how well an AI model really understands things, not just predicting words. Just like an exam it asks questions on science, maths, law, medicine and more - which a well-educated human might struggle with. Examples: Gemini 3 Pro scored around 37.5% compared to ChatGPT 5.1 at around 26%.

IoT (Internet of Things)Everyday devices (like fridges, watches, or cars) connected to the internet, collecting and sharing data.

Language ModelAn AI program that learns patterns in human language so that it can understand and predict or generate text, but might be trained on limited or very specific data. Examples: Predictive text, autocomplete, early spam filters.

Large Language Model (LLM)A type of language model trained on huge amounts of text data and is able to answer questions, write text, and more, like a human. They are a Foundation Model that only produce text. Examples: GPT-3, Gemma, Claude 1 & 2.

Low-CodeA way to build applications (including AI) using mostly visual tools and minimal coding, making it more accessible. Example: Zapier, Make.com, Relay.app

Machine Learning (ML)A type of AI where computers learn from their own examples and experience, instead of being given exact instructions for every task.

MemoryA small, controlled notebook the AI keeps so it can tailor replies to you. You choose what it remembers, and you can delete anything or update at any time. Example: Ask your chatbot to remember to use non-technical language when memory is activated.

Model Context Protocol (MCP)A method to connect AI tools to your own files, data and systems safely, rather than accessing everything. It asks for only what it needs, when it needs it.

Multi-ModalAn AI system which can work with more than one type of input - not just text, but images, audio, video and code. Examples: GPT-4, Claude 3, Gemini 1.5

Neural NetworkA computer system inspired by the human brain, made up of layers of connected “nodes” that process information and can find patterns. Example: Plant identification app

NLP (Natural Language Processing)AI that helps computers understand and work with human languages, such as reading text, translating or speech recognition.

Open SourceSoftware or AI models that are made freely available for anyone to look at, use, change, and share. The idea is that the “recipe” is open, so a community of people can improve and build on it together, instead of it being locked away by one company. Example: OpenAI’s new open-source releases in 2025 (called gpt-oss-120b & gpt-oss-20b - catchy!), which developers can adapt for their own projects.

PredictionWhen an AI system uses what it has learned to guess what might happen next or to answer a question.

PromptA question, instruction, or statement you give to an AI system to get a specific response.Example: Typing “Write a poem about spring” into ChatGPT.

Prompt EngineeringCrafting effective prompts to get better or more accurate results from AI systems, especially large language models (LLMs).

Prompt InjectionWhen a bad actor hides instructions in a prompt (or in the data the AI reads) to trick the system into doing something that wasn't intended by the user. Example: A website secretly containing text that tells an AI assistant to reveal sensitive company information.

Reasoning ModelAI tools designed to "think out loud" before giving an answer. They slow down, break answers into steps, and usually get more accurate results. Examples: Options to choose a "Reasoning", "Thinking" or "Planning" approach.

Reinforcement LearningA way for AI to learn by trial and error, getting rewards for good choices and penalties for mistakes. Example: Teaching a robot to walk or an AI to play games.

Retrieval Augmented Generation (RAG)A technique where a database or knowledge source is searched first for relevant information - retrieval - which is then used to generate a more accurate or useful response - generation. Example: Summarise our employee benefits - pulls from your HR documents not its general learning.

RoboticsDesigning and building robots, often using AI to help them act intelligently. Examples: Robot vacuum cleaners, industrial robots.

Semantic DatabaseA way of storing data so that computers can understand the meaning and relationships between different pieces of information. Think of it as a smart library that understands concepts and relationships, not just words.

Structured DataInformation organised in a clear, fixed way, like in tables or spreadsheets. Example: Names and addresses in a table.

Supervised LearningA way of training AI where the computer is given examples with the correct answers, so it can learn to make similar decisions on its own.

TemperatureA setting that controls how random or creative the AI responses are. The lower the temperature, the more predictable; higher temperatures give more diverse or unusual responses. Usually on a scale of 0 to 1.

TokenThe smallest chunk of text AI models process - roughly a few letters or part of a word. It's used to help understand how much the model can process in it's context window, or in some AI products to set pricing. Example: Most English words use 1-2 tokens, so 1,000 tokens is about 750 words.

Tool-enabledGiving an AI model access to external "tools" so it can go beyond its own training when responding. Example: connecting to a database, email account or excel.

Turing TestA game-like test proposed by Alan Turing to see if a computer can fool someone into thinking it's human through conversation.

Transfer LearningUsing a model trained for one task to help learn a different, but related, task more quickly.

Unstructured DataInformation that doesn't have a fixed or organised format, making it harder to search or analyse. Example: emails, photos, videos, audio files, social media posts, documents, my notebook

Unsupervised LearningA way of training AI where the computer looks for patterns in data without being told what the correct answers are.

Vibe CodingA playful term for using AI tools to create software without writing having to know or write traditional code. Example: Requesting “make me a website with a contact form and a blue header” and the AI generates the code and builds the website automatically.

Voice AgentAn AI system you can talk to and replies, like having a conversation with a person. Unlike a simple voice assistant (like Siri or Alexa), a voice agent can handle longer, more natural conversations and even take actions for you. Example: Speaking with a voice agent to answer questions about your cv, experience and career as part of an interview process.

About the Author

Connect with me on LinkedIn here

Create your own Scribes here

Check out my Scribe Community here