AI Glossary
D
OpenAI's text-to-image generation model, one of the pioneering tools of generative visual AI.
AI-supported user interface design techniques that deceptively push users toward decisions against their own interests.
An EU regulation governing access to and sharing of data generated by connected devices, which also affects the availability of data for training and operating AI.
We summarize what you should not overlook in our article Data Act – what you should not overlook.
The voluntary sharing of data by individuals or organizations in the public interest, for example for research, as regulated by the Data Governance Act.
The infrastructure essential for training and running models; its energy demands and location are becoming the subject of political and regulatory debate.
An entity that determines the purposes and means of the processing of personal data.
A change in the characteristics of input data over time that can reduce the reliability of a model without the model itself changing.
A set of rules and processes for managing data within an organization, ensuring its quality, availability, security and legal compliance throughout its lifecycle.
An EU regulation promoting trustworthy data sharing, laying down rules for data intermediaries and the re-use of public sector data.
An entity that controls data generated by a connected product or service and decides on making it available under the rules of the Data Act.
An entity that intermediates the sharing of data between those who provide it and those who use it, under the conditions laid down by EU law.
A requirement to process only the data necessary for a given purpose, which conflicts with the tendency of models to benefit from as large a volume of data as possible.
A targeted attack in which malicious or misleading samples are deliberately inserted into training data in order to disrupt the behavior of the resulting model.
The ability to move data between different service providers.
An agreement under data protection rules governing the processing of data by an AI service provider on behalf of the controller.
A requirement that data used for the development and operation of AI be relevant, representative, sufficiently accurate and, as far as possible, free of systematic bias.
The provision of data to another entity under conditions laid down by contract or by law.
A natural person whose personal data is processed.
A freely given, specific, informed, and unambiguous expression of will regarding the processing of personal data.
An entity that, by law or contract, gains access to data and uses it under conditions laid down by the Data Act.
The sui generis right of a database maker to protect content in which they have made a substantial investment; relevant when large datasets are used for training.
A structured collection of data used for training; the conditions of its use are set by its license, the breach of which may give rise to liability.
The question of whether, and to what extent, a dataset used for training is protected by copyright or by the sui generis database right, and on what conditions it may be used.
License terms determining how a dataset may be used, shared and further distributed; breaching them may give rise to liability.
The dataset equivalent of a model card; it documents the origin, composition, collection method and limitations of the data so that its suitability can be assessed.
Fixed deadlines that the Czech draft sets for the processes of authorizing the operation and testing of high-risk AI systems in real-world conditions.
The part of a model that generates output, such as text, from an internal representation; typical of models geared towards content creation.
A subfield of machine learning that uses neural networks with many layers, capable of recognizing complex patterns in large datasets.
Targeted tailoring of content to an individual based on extensive data about their behavior, which can increase the effectiveness of manipulative practices.
Realistic-looking but artificially created or altered image, audio or video content depicting a person saying or doing something that never happened.