Conversations about artificial intelligence move quickly between terms that sound similar but mean different things. This glossary covers 20 of the most common, in the order a newcomer is likely to meet them, with definitions short enough to remember and precise enough to be useful.
The basics
Artificial intelligence (AI). The broad field of building computer systems that perform tasks normally associated with human intelligence, such as recognizing speech, interpreting images, making decisions and generating language.
Machine learning (ML). A branch of AI in which a system learns patterns from data rather than following rules written by hand. Most modern AI, including every product described below, is built on machine learning.
Neural network. A computing structure loosely inspired by the brain, made of layers of connected units (“neurons”) that each apply a simple mathematical operation. Training adjusts the strength of the connections so the network maps inputs to useful outputs.
Deep learning. Machine learning that uses neural networks with many layers. The depth lets the network learn increasingly abstract features, which is why deep learning drove the advances in image recognition, speech and language of the past decade.
Generative models
Generative AI. AI that produces new content, such as text, images, audio, video or code, by learning the patterns in existing material. ChatGPT, Gemini, Claude and image generators such as Midjourney are generative systems.
Large language model (LLM). A neural network trained on very large amounts of text to predict and generate language. LLMs underpin most chatbots and writing assistants.
GPT (Generative Pre-trained Transformer). OpenAI’s family of large language models, and by extension a widely used shorthand for LLMs built on the transformer architecture. “Pre-trained” refers to the initial training on general text before any task-specific tuning.
Multimodal AI. A model that can take in and reason over more than one kind of input, such as text, images and audio together, and often produce more than one kind of output.
How models learn and run
Training data. The collection of examples a model learns from. Its size, quality and composition largely determine what the model can do and what mistakes it will make.
Token. The unit of text a language model actually processes: usually a word, part of a word or a punctuation mark. Model limits and API pricing are typically expressed in tokens.
Inference. The stage at which a trained model is used to produce an output for a new input. Training happens once (or occasionally); inference happens every time someone sends a prompt.
Fine-tuning. Further training of an existing model on a smaller, specialized dataset so it performs better on a particular task or in a particular style.
Prompt. The instruction or input given to a model. Writing prompts that reliably produce the intended result is often called prompt engineering.
Building applications
Retrieval-augmented generation (RAG). A technique in which the system first retrieves relevant documents from an external source, then passes them to the model along with the question, so the answer is grounded in that material rather than only in what the model memorized during training.
AI agent. A system that uses a model to plan and carry out a sequence of actions toward a goal, such as browsing websites, running code or calling other software, with limited human intervention between steps.
AI automation. The use of AI to perform repetitive digital tasks, such as sorting messages, filling forms or summarizing reports, that previously required a person.
Model Context Protocol (MCP). An open standard, introduced by Anthropic in November 2024, that defines how AI applications connect to external tools, data sources and services. The protocol’s documentation describes it as a common interface so that any compliant model or app can use any compliant tool, and it has since been adopted by other major AI developers.
Trust and perception
Bias in AI. Systematic unfairness in a model’s outputs, usually inherited from imbalances or prejudices in its training data, which can lead to worse results for particular groups of people.
Explainable AI (XAI). Methods for making a model’s decisions understandable to people, so that users, regulators and developers can see why a system produced a given result.
Computer vision. The area of AI concerned with enabling machines to interpret images and video, from recognizing faces and reading text to guiding self-driving cars.
These definitions are deliberately general. In practice, terms such as “agent” and “reasoning” are used loosely in marketing, so it is worth asking what a product actually does when it claims one of these labels.
