The Essential AI Glossary: Every Key Term You Need to Know
- Stéphane Guy
- Mar 8
- 5 min read
As artificial intelligence becomes increasingly embedded in everyday life, a specialized vocabulary has quietly taken root. Whether you're navigating a product demo, reading a tech brief, or building AI-powered tools yourself, fluency in this language is no longer optional; it's a professional baseline. Here's your definitive guide.
In brief
Artificial intelligence is an umbrella term for techniques that enable software to mimic human reasoning. It hinges on foundational concepts such as artificial neural networks and algorithms.
A prompt is the instruction or query you send to an AI system to generate a response. Its form varies widely, from a casual chat message to a structured set of parameters for generative tools.
Generative AI creates original content (text, images, audio) rather than simply sorting or classifying existing data. Tools like ChatGPT, Gemini, and Suno fall into this category.
Supervised learning requires human oversight to train an AI model; unsupervised learning lets the model discover patterns autonomously. The distinction directly shapes how well a system adapts to new data.
Big Data refers to the massive, continuously accumulating datasets that form the raw material for training and refining AI models at scale.
Artificial Intelligence
Let's start with the foundation. Artificial intelligence, AI, refers to a broad set of techniques designed to make software behave as close to a human as possible: understanding language, recognizing patterns, reasoning through ambiguity. For a deeper look at what AI actually is under the hood, check out our dedicated explainer.
Prompt
A prompt is the input you send to an AI through its interface, a chat window, an API call, a command panel. It can be as simple as "What's the most energy-efficient time to run the dishwasher?" in ChatGPT, or as structured as a multi-field parameter block in a generative music tool like Suno. Prompt quality directly determines output quality, which is why prompt engineering has become a discipline in its own right.
Artificial Neural Network
An artificial neural network is a specific AI architecture built from multiple algorithms arranged in successive "layers" that process information in parallel. The structure deliberately mirrors the human brain: rather than a single computation path, many interconnected units collaborate to tackle complex problems. The result is an AI capable of nuanced pattern recognition, context handling, and inference that linear rule-based systems simply can't match.
Chatbot
A chatbot is a lightweight conversational program designed to interact with users and field questions of varying complexity. In practice, you encounter them as customer service widgets on e-commerce sites, virtual assistants in banking apps, or support agents embedded in SaaS platforms. Modern chatbots powered by large language models (LLMs) have far surpassed their keyword-matching predecessors.
Generative AI
Generative AI designates any AI system capable of producing new content, text, images, video, audio, code. It stands apart from classical AI applications (classification, clustering, anomaly detection) that manipulate existing data without creating anything new. Landmark examples include ChatGPT, Microsoft Copilot, Google Gemini, and Suno.
Language Model
A language model is a program trained to understand and generate human language by analyzing massive text corpora. Early models worked word by word; today's large language models (LLMs) process entire sentences and paragraphs, grasping meaning, tone, and context holistically. This shift, from token prediction to semantic understanding, is what makes modern AI feel conversational rather than mechanical.
Algorithm
An algorithm is the step-by-step procedure that transforms inputs into outputs. Think of it as an assembly instruction manual: the raw materials are your data, the algorithm is the build sequence, and the finished product is your result. Algorithms are the operational core of every AI system, from the ranking logic in your search engine to the recommendation engine on Netflix.
Supervised Learning vs. Unsupervised Learning
Supervised learning is human-guided training. Engineers feed the model labeled examples, spam vs. legitimate email, fraudulent vs. valid transactions, until it learns to classify new cases reliably on its own. Precision is high, but the model's scope is bounded by the data it was shown.
Unsupervised learning removes the human referee. The model ingests raw, unlabeled data and discovers structure independently. This is what powers Spotify's "Discover Weekly" or Netflix's content recommendations, the system detects that you've shifted from rom-coms to Nordic noir and adjusts without anyone telling it to.
The key difference: a supervised model needs manual updates to handle genuinely new categories; an unsupervised model adapts on its own. For a full breakdown of learning paradigms, including reinforcement learning, see our dedicated guide.
Big Data
Big Data describes datasets so vast, fast-moving, or structurally diverse that conventional databases can't handle them. Health records for an entire continent, real-time transaction logs from a global bank, sensor streams from millions of IoT devices, these are Big Data problems.
For AI, Big Data isn't just a storage challenge; it's the training fuel. The more representative and high-quality the data, the more capable the resulting model.
Machine Learning, Continual Learning & Federated Learning
Machine learning is the capacity of an AI system to improve its performance autonomously, using one or more mathematical models as its operational framework. Once the model has internalized its "instruction set," it learns from new data without human reprogramming.
Continual learning (also called lifelong learning) takes this further: the system keeps updating itself in real time, even while actively deployed. No downtime, no retraining cycles, it evolves continuously as it encounters new information.
Federated learning is an architectural approach to privacy-preserving AI training. Multiple organizations, hospitals, banks, research labs, each train the model on their own local data without ever sharing raw records with each other. Only the model updates (gradients) are sent to a central coordinator, who aggregates them into a unified, improved model. The result: collective intelligence without collective data exposure. This framework underpins compliance with regulations like GDPR and the EU AI Act.
FAQ
What is a prompt in AI?
A prompt is the instruction or query you send to an AI system. Its form ranges from a casual chat message ("suggest a dinner recipe") to a structured parameter block in generative tools. Prompt quality directly determines output quality, which is why prompt engineering is now a job title.
What's the difference between supervised and unsupervised learning?
Supervised learning uses human-labeled data to train a model on specific patterns (spam detection, fraud flagging). Unsupervised learning removes the labels, the model discovers structure on its own. Supervised = more precise but rigid. Unsupervised = more adaptive. Your Spotify recommendations run on unsupervised learning.
What is generative AI and how is it different from classical AI?
Generative AI produces new content, text, images, audio, code. Classical AI classifies or sorts existing data without creating anything new. ChatGPT and DALL-E are generative. Your email spam filter is not. That distinction is fundamental to understanding AI's current commercial explosion.
Why does Big Data matter for AI?
AI models learn from data. The more diverse and representative the dataset, the more robust the model. Big Data provides the volume and variety needed to train systems that generalize beyond their training set, separating capable AI from brittle AI.
What is an artificial neural network?
A computational architecture modeled on the human brain. Multiple layers of interconnected processing units work in parallel to tackle complex problems, pattern recognition, language understanding, contextual inference. Neural networks are the structural foundation of every major AI system running today.
What's the difference between machine learning and continual learning?
Machine learning gives an AI a mathematical framework to learn from data autonomously. Continual learning (lifelong learning) keeps that process running in real time, the model updates itself while deployed, without retraining cycles or downtime.