Over 50 clear and up-to-date definitions to understand Artificial Intelligence, the European AI Act regulation, GDPR and the key concepts of digital ethics.
From the AI Act to explainability, from algorithmic bias to human oversight: this glossary brings together the essential vocabulary of Data Science, Machine Learning and the ethical governance of AI systems. A reference designed for decision-makers, legal professionals, data scientists and compliance officers.
The first European regulation on artificial intelligence (EU Regulation 2024/1689). It governs the placing on the market and the use of AI systems through a risk-based approach (unacceptable, high, limited, minimal) and imposes obligations of transparency, governance and human oversight.
A finite, ordered set of instructions designed to solve a problem or accomplish a task. In AI, the algorithm learns rules from data rather than receiving them explicitly.
An independent assessment of an AI system aimed at verifying its compliance, fairness and robustness. This is at the heart of the ADEL label approach and the DEO Platform.
A process of sensitising teams and building their skills around the challenges of Data Science and AI ethics — a prerequisite for responsible adoption.
A Machine Learning method in which the model learns from labelled data (inputs paired with known outputs) to predict new cases.
A method in which the model independently detects structures, clusters or anomalies in unlabelled data (clustering, dimensionality reduction).
A systematic distortion of an AI system's outputs, reproducing or amplifying discriminations present in training data or the model's design. Detecting it is a central challenge of responsible AI.
Datasets of such volume, velocity and variety that they exceed the capacity of traditional processing tools. The raw material for many AI models.
The dataset used to train a model. Its quality, representativeness and governance directly determine the reliability and fairness of the system.
Commission Nationale de l'Informatique et des Libertés: the French authority responsible for ensuring the protection of personal data and the enforcement of GDPR.
The principle of integrating regulatory requirements (GDPR, AI Act) from the design stage of a system, rather than retrospectively.
The full set of obligations applicable to AI systems according to their risk level: technical documentation, risk management, transparency, human oversight, logging.
The step of cleansing and normalising data (correcting errors, handling missing values, deduplication) prior to any reliable modelling.
A discipline combining statistics, computer science and domain expertise to extract value from data, from collection through to predictive modelling.
A sub-field of Machine Learning based on deep neural networks with multiple layers, responsible for recent advances in vision, language and generative AI.
Under the AI Act, an organisation that uses an AI system under its own authority. It bears obligations of human oversight and incident monitoring.
A platform for self-assessment and auditing the ethical maturity of AI systems, underpinning GoodAlgo's ADEL labels.
Data Protection Officer: the designated person responsible for steering GDPR compliance within an organisation.
Reflection on the values, duties and responsibilities that should govern the design and use of digital technologies in the service of people.
An approach developed by Jérôme Béranger: evolving ethics continuously alongside technology, rather than freezing it in static rules.
The integration of ethical principles from the design stage of a digital system: transparency, fairness, respect for fundamental rights.
The ability to explain in an understandable way the decisions made by an AI system. A key requirement for trust, audit and AI Act compliance.
The property of a model that treats individuals and groups without unjustified discrimination, measured through dedicated statistical tests.
A very large model (foundation model) pre-trained on vast corpora, adaptable to multiple tasks — the backbone of modern generative AI.
Targeted retraining of a pre-trained model on specific data to adapt it to a business use case.
The full set of policies, roles and processes ensuring the quality, security and compliance of data throughout its lifecycle.
A family of AI systems capable of producing original content (text, image, audio, code) from instructions. Covered by the ADEL-AI Use label.
General-Purpose AI: AI models intended for general use, subject to specific transparency and documentation obligations under the AI Act.
An AI Act category covering AI systems that may affect health, safety or fundamental rights (recruitment, credit, healthcare, justice…). Subject to the strictest obligations.
The production by a generative AI of information that is plausible but false. A major risk to be managed through human oversight.
A set of techniques enabling machines to perform tasks associated with human intelligence: perception, reasoning, learning, decision-making.
The design and use of AI that respects ethics, fundamental rights and regulation, with transparency and accountability.
AI that is lawful, ethical and robust, according to the European Commission's framework: a system worthy of trust throughout its entire lifecycle.
The operational phase of a trained model: it produces predictions or content on new real-world data.
GoodAlgo certification attesting to the ethical commitment of an actor or AI system, available as ADEL-Project, ADEL-AI Act and ADEL-AI Use (Bronze, Silver, Gold levels).
Large Language Model: a Deep Learning model trained on vast text corpora, capable of understanding and generating natural language.
Automated learning: a branch of AI in which a model learns rules from data rather than explicit instructions.
The construction of a mathematical model representing a phenomenon from data, in order to predict or explain it.
The level of advancement of an organisation in its responsible AI approach, measured by the DEO Platform and materialised through ADEL labels.
A defect in a model that is too closely fitted to its training data, causing it to generalise poorly on new data.
Any information relating to an identified or identifiable natural person, protected under GDPR.
The use of statistical models and AI to anticipate future events or behaviours based on historical data.
Under the AI Act, the entity that develops or places on the market an AI system. It bears the majority of compliance obligations.
GoodAlgo's proprietary tool for self-assessment and monitoring of the digital ethical maturity of AI projects and systems.
General Data Protection Regulation (EU 2016/679): the European framework protecting citizens' personal data and governing its processing.
The ability of an AI system to remain reliable and safe in the face of errors, unusual data or adversarial attacks. A requirement of the AI Act.
A model inspired by the brain, made up of layers of artificial neurons, at the heart of Deep Learning.
The application of Corporate Social Responsibility to digital challenges: algorithmic ethics, digital sobriety, inclusion.
A principle ensuring that a human retains control over an AI system: the ability to intervene, correct or stop a decision. Mandatory for high-risk AI systems.
Making use of small volumes of quality data, often more relevant and resource-efficient than Big Data for many industrial use cases.
Under the AI Act, an automated system that, for given objectives, generates outputs (predictions, recommendations, decisions) that influence its environment.
Technical and organisational measures protecting data against loss, alteration or unauthorised access.
The ability to make the functioning and purposes of an AI system intelligible to its users and to authorities.
Any operation performed on personal data (collection, storage, analysis, sharing), governed by GDPR.
The logging of events in an AI system enabling its decisions to be reconstructed and audited after the fact.
The transformation of raw data into business value: insights, automation, decision support.
The graphical representation of data and AI results to make them readable and actionable (e.g. GoodAlgo's VIZIAA tool).
eXplainable AI: a set of methods making AI model decisions interpretable by humans, essential for trust and compliance.
The ability of a model to handle a task without prior specific examples, drawing on its generalised knowledge.
Our experts guide you from ethical audit to ADEL certification. Assess the maturity of your AI systems for free.