Glossary · Reference

AI Act glossary

Concise, plain-language definitions of the terms used across this site and in the game. Educational simplifications: the regulation's own definitions are in the official text.

See the terms in action Plain-language guide

01The terms

AI Act: Regulation (EU) 2024/1689, the European Union's framework regulation for artificial intelligence, organised around a risk-based approach.
AI system: A machine-based system that, with varying autonomy, infers from inputs how to generate outputs — predictions, content, recommendations or decisions — that can influence environments.
Artificial intelligence: The broad field of techniques that let machines perform tasks associated with human intelligence, such as pattern recognition, prediction and language.
Risk-based approach: The AI Act's organising logic: the stricter the potential impact of a use on safety and rights, the stricter the rules — from outright bans to no specific obligations.
Prohibited AI practices: The small set of uses banned outright — such as generalized social scoring — because they are judged incompatible with fundamental rights regardless of safeguards.
High-risk AI system: A system whose use can significantly affect people's safety, opportunities or rights (education, employment, essential services, health…), allowed only with strict obligations.
Transparency obligation: The duty, in defined situations, to tell people they are interacting with AI or looking at AI-generated content.
General-purpose AI (GPAI): A model trained for broad capability and usable across many downstream tasks and systems, rather than for one purpose.
Generative AI: AI that produces content — text, images, audio, video; the most familiar form of general-purpose models.
AI literacy: The competences needed to understand, use and critically evaluate AI systems — including a basic grasp of the rules that govern them.
Serious game: A game designed primarily for learning or training rather than pure entertainment.
Privacy by design: Building privacy protections into a system's architecture from the start — collecting nothing by default — rather than adding them afterwards.
Human oversight: Supervision of an AI system by a person with the competence, information and authority to genuinely understand, question and override it.
Data governance: The practices ensuring the data used to build and run a system are relevant, representative and properly managed.
Conformity assessment: The process of verifying, before deployment, that a high-risk system meets the requirements that apply to it.
Fundamental rights: The rights protected in the EU legal order — dignity, non-discrimination, privacy, effective remedy — that serve as the AI Act's yardstick.
Fundamental rights impact: The assessment of how a system's deployment affects people's fundamental rights, especially for high-risk uses.
Accountability: Being answerable for a system's effects: identifiable, responsible, and exposed to consequences and remedies.
Documentation: The written record — purpose, data, design choices, validation — that makes a system explainable to authorities, buyers and affected people.
Risk management: The continuous process of identifying, evaluating and mitigating what can go wrong with a system, before and during use.
Provider: The actor that develops an AI system or model and places it on the market under its own name.
Deployer: The actor that uses an AI system under its own authority — the school, office or company that deploys it on real people.
User: In everyday language, the person interacting with or affected by a system; note that the regulation's defined roles are provider and deployer.
Biometric identification: Recognising a person from physical or behavioural traits — face, voice, gait; heavily restricted in public spaces.
Emotion recognition: Inferring a person's emotional state from their signals; prohibited in workplaces and schools, with narrow exceptions.
Recommender system: A system that ranks or suggests content and options; a daily-life example of AI shaping choices without deciding formally.
Automated decision-making: Decisions taken wholly or partly by systems without meaningful human involvement — the situation most of the AI Act's safeguards exist for.

02How to use this glossary

In class, hand out five terms per group before playing and have each group flag when "their" terms appear in a case. For the legal definitions, always defer to the official text on EUR-Lex. Context for the concepts: the guide and risk categories.