Appendix & glossary

Your AI dictionary — every key word from the programme, in plain language.

42 terms. Tip: search this page with Ctrl/Cmd + F to find a word fast.

A

Algorithm: A set of step-by-step instructions for solving a problem. The foundation of all software, including AI.
Artificial Intelligence (AI): Computer systems that perform tasks normally requiring human intelligence — recognising patterns, understanding language, making decisions.

Bias (AI): Systematic unfairness in AI outputs, typically caused by skewed training data that over- or under-represents certain groups.
Black Box: An AI system whose internal decision-making process cannot be inspected or explained.

Chatbot: A program that simulates conversation with humans, increasingly powered by large language models.
CNN (Convolutional Neural Network): The standard neural network architecture for processing images — uses layers of filters to detect features at increasing levels of complexity.
Computer Vision: AI that interprets and understands digital images and video.
Context Window: The maximum amount of text an AI language model can "see" at once when generating a response.

Deep Learning: A subset of machine learning using multi-layered neural networks — excels at images, language, and audio.
Deepfake: AI-generated synthetic media that realistically replaces a real person's likeness, voice, or words.
Diffusion Model: AI architecture for generating images — starts with noise and gradually refines it into a coherent image guided by a text prompt.

Epoch: One complete pass through all training data during model training. More epochs = more learning (up to a point).
Explainability (XAI): The ability to interpret and explain how an AI system reached a specific conclusion — critical for high-stakes decisions.

Fairness (AI): The principle that an AI system should treat all groups equitably and not cause disproportionate harm to any group.
Feature: A measurable characteristic that an AI uses to make decisions — e.g. "edge angle" in image recognition.

GAN (Generative Adversarial Network): AI architecture with two competing networks — a generator that creates fake content and a discriminator that tries to detect fakes.
Generative AI: AI that creates new content (text, images, audio, video) rather than just classifying or predicting.
Guardrails: Constraints built into AI systems to prevent harmful, biased, or off-topic outputs.

Hallucination: When an AI language model outputs confident but factually incorrect information — a core limitation of current LLMs.

Label: The correct answer attached to a training example — e.g. "cat" or "not cat" — used in supervised learning.
Large Language Model (LLM): AI trained on vast amounts of text to understand and generate human language — the technology behind ChatGPT, Claude, and Gemini.
Latent Space: The multi-dimensional mathematical space inside an AI model where concepts are encoded as numerical relationships.

Machine Learning (ML): A subset of AI where systems learn from data rather than being explicitly programmed with rules.
Model: The mathematical structure an AI builds from training data — encodes learned patterns and is used to make predictions.

Narrow AI: AI that excels at one specific task but cannot transfer skills to other domains. All current AI is narrow AI.
Natural Language Processing (NLP): AI that understands, interprets, and generates human language in text or speech form.
Neural Network: A computing architecture loosely inspired by the brain — layers of interconnected nodes process information in parallel.

Overfitting: When an AI memorises training data too closely, performing perfectly on training examples but failing on new data.

Parameter: A learnable numerical value inside a neural network that is adjusted during training to minimise errors.
Prompt: The instruction or question you provide to an AI system to guide its output.
Prompt Engineering: The skill of crafting effective, specific instructions to get high-quality outputs from AI systems.

Reinforcement Learning: ML where an AI agent learns by trial and error — rewarded for correct actions, penalised for wrong ones.

Speech Recognition: AI that converts spoken audio into written text.
Supervised Learning: ML where the AI is trained on labelled examples — the most common type of machine learning today.
System Prompt: Hidden instructions given to an AI chatbot that define its persona, rules, and constraints.

Temperature: A setting that controls how creative/random (high temperature) vs predictable/focused (low temperature) an AI's outputs are.
Token: The basic unit an LLM processes — roughly a word fragment (e.g. "unbelievable" = 3 tokens: "un", "believ", "able").
Training Data: The examples used to teach an AI — the quality and diversity of training data largely determines AI quality.
Transfer Learning: Adapting a pre-trained model for a new task rather than training from scratch — dramatically reduces data and compute requirements.

Underfitting: When an AI hasn't learned enough from training data and performs poorly even on familiar examples.
Unsupervised Learning: ML where the AI finds patterns in unlabelled data without being told what to look for.

XAI (Explainable AI): A research field focused on making AI decision-making transparent and interpretable by humans.

Now you know the words — time to see them in action. Ready to start?

Free · No login · Play with friends · Safe for kids