Artificial intelligence glossary.

Artificial intelligence

A set of techniques that enable a machine to perform tasks that would normally require human intelligence: understanding, reasoning, decision-making, and creation.

Machine learning

A branch of AI in which a model learns rules from data rather than being explicitly programmed.

Deep learning

An approach to machine learning that uses multi-layered neural networks, and lies behind the recent breakthroughs in AI.

Neural network

A model inspired by the brain, made up of layers of 'neurons' that transform input data into predictions.

Algorithm

A sequence of instructions for solving a problem or completing a task; in AI, it governs how the model learns and makes predictions.

Model

The output of training an algorithm on data: it is what produces predictions or generated content.

Training data

The dataset used to teach a model; its quality directly determines the model's performance.

Predictive AI

AI that forecasts an outcome (risk, demand, failure) based on historical data.

Generative AI

A category of AI capable of producing new content — text, images, code, audio — from a given prompt.

LLM (large language model)

Large Language Model: a model trained on vast text corpora, capable of understanding and generating natural language.

ChatGPT

A conversational AI agent from OpenAI built on a large language model, which brought generative AI to mainstream attention in late 2022.

Prompt

An instruction or question given to a generative AI in order to obtain a result.

Prompt engineering

The art of crafting effective prompts to obtain accurate and useful responses from a generative AI.

Hallucination

When an AI produces a confidently worded but factually incorrect response — the key risk to understand and guard against.

Token

The basic unit (a word or word fragment) processed by a language model; also used to measure the length of texts.

AI agent

An AI system capable of chaining actions autonomously to complete a task end to end.

RAG

Retrieval-Augmented Generation: a technique combining a language model with a document knowledge base to produce reliable, source-backed responses.

Fine-tuning

The process of specialising a pre-trained model on data specific to a particular use case or domain.

Embedding

A numerical representation of a piece of text or data that allows an AI to measure its meaning and proximity to other content.

Multimodal

Describes an AI capable of processing multiple types of data (text, images, audio) simultaneously.

AI copilot

An AI assistant integrated into a business tool to support users in their day-to-day tasks.

NLP (natural language processing)

The branch of AI dedicated to understanding and generating human language (also known as NLP).

Computer vision

Technologies that enable a machine to analyse and interpret images and video.

AI speaker

An expert who presents at events to make artificial intelligence accessible and inspire audiences to act.

Keynote

The headline talk at an event — typically the opening or closing session — designed to make an impact and unite the audience.

Plenary

A session that brings together all participants at an event; the natural setting for a keynote.

Masterclass

An in-depth, interactive session led by an expert, focused on building skills and knowledge.

Workshop

A hands-on format in which participants work directly with tools on real-world use cases.

Expert witness

A specialist invited to provide a broader perspective and put topics into context during an event.

Panel discussion

A moderated exchange between several speakers on a given theme, often facilitated by an expert.

Remote session

A presentation delivered online via video conference, as opposed to an in-person event.

AI Act

European Regulation (EU) 2024/1689 on artificial intelligence — the world's first horizontal legal framework for AI, built on a risk-based approach.

GDPR

General Data Protection Regulation: the framework governing the processing of personal data in the European Union.

CNIL

France's national data protection authority (Commission nationale de l'informatique et des libertés), responsible for enforcing data privacy rules.

Algorithmic bias

A distortion in a model that reproduces or amplifies inequalities present in its training data.

Explainability

The ability to understand and justify the decisions made by an AI system — a key requirement in regulated contexts.

Human oversight

The principle that a human retains control and can intervene in the decisions made by an AI system.

Trustworthy AI

AI designed to be robust, transparent, fair, and respectful of fundamental rights.

Responsible AI

An approach that seeks ethical and controlled use of AI, incorporating social and environmental impact alongside regulatory compliance.

Digital sovereignty

The capacity of an organisation or country to maintain control over its data, infrastructure and technology.

Personal data

Any information relating to an identified or identifiable natural person, protected under the GDPR.

Digital transformation

The process of integrating digital technologies into every aspect of an organisation.

Use case

A concrete, identified application of a technology to a specific business need.

POC (proof of concept)

A prototype designed to validate the feasibility and value of a project before significant investment is committed.

AI awareness

An initiative aimed at familiarising all employees with a technology such as AI.

Change management

The actions taken to support teams in adopting a new technology or organisational approach.

Automation

The delegation to a machine of tasks previously carried out by a human.

Augmented productivity

The efficiency gains achieved when AI assists humans in their tasks.

Qualiopi

A French certification attesting to the quality of training programmes, and a prerequisite for public or OPCO funding.

MLOps

A set of practices for deploying, monitoring and maintaining AI models in production reliably.