AI Glossary

An indicator of how often a model produces false or fabricated information; used to compare the reliability of models and to manage risks.

Specialized hardware (e.g., graphics or tensor processors) designed to accelerate the computations typical of the training and operation of models.

A technical standard adopted by the European standardization organizations, compliance with which establishes a presumption of conformity of a high-risk AI system with the requirements of the AI Act.

A common framework shared by management system standards (quality, information security, AI), which facilitates their simultaneous implementation within a single organization.

Editorial annotations and summaries of points of law attached to court decisions in the Westlaw database. In the Ross Intelligence case, the court recognized their originality and copyright protection.

A group of class action lawsuits by book authors against Apple, which place it among the technology companies facing copyright disputes over AI training.

An AI system falling into the high-risk category under the AI Act, subject to strict requirements on risk management, data quality, documentation, transparency and human oversight.

A platform and community repository of models and datasets, key to the sharing and deployment of open models.

An arrangement in which the AI proposes a course of action or output, but its execution must be approved by a human before it is carried out; a form of human oversight.

A requirement that fundamental employment decisions not be left solely to AI.

A value whose protection the AI Act emphasizes, particularly in prohibiting manipulative practices and disproportionate social scoring.

Human evaluation of a model's outputs, used to fine-tune its behavior toward usefulness and safety.

A set of measures enabling a human to monitor the operation of an AI system and to intervene where necessary.

An approach to the design and use of AI that places the human being, their rights, interests and dignity at the center.

An arrangement in which a human remains part of the AI decision-making process and can intervene in it, review it or override the system's output.

A search combining classic keyword-based search with vector-based search by meaning, so that the results are more precise and relevant.