FAIRmedia

Project funding:
Austrian Research Promotion Agency (FFG)
Federal Ministry for Climate Action, Environment, Energy, Mobility, Innovation and Technology

Project website:
FAIRmedia » DIGITAL (joanneum.at)

Project partners
JOANNEUM RESEARCH Forschungsgesellschaft mbH – DIGITAL – Institute for Information and Communication Technologies (Projektleitung)
​​​​​ORF – Multimedia Archive
FH St. Pölten - Institut für Creative\Media/Technologies
Redlink GmbH
University of Vienna

Project details

Most of the data sets used to train AI currently come from organizations outside the EU, particularly from the US and China. This not only leads to dependencies on global technology companies, but also carries the risk that relevant data sets can be withdrawn at any time. Furthermore, there is a risk that AI tools and data sets developed outside Europe do not meet European standards and legal frameworks, such as the GDPR and upcoming AI regulations.
 

The main objectives of FAIRmedia are:

• Creating a framework for the efficient creation of fair and trustworthy data sets: The focus is on current and archived content from Austrian media organizations to ensure a locally produced, trustworthy basis. This also allows regional characteristics (e.g. locally used words or phrases) to be reflected in the created data sets. The creation is supported by tools for analyzing the balance of data sets.
   
• Creation of no-/low-code tools: These tools enable the creation of datasets and the training of AI models without in-depth programming knowledge. The target groups are various actors in the media industry, such as journalists, editorial managers, audience managers and archivists. Through local training, the data remains under the control of the media companies.
   
• Development of post-hoc explanation tools for AI models: These serve to better understand the trained models and provide comprehensible explanations and hints regarding the balance of data sets.
• Demonstration of the tools: using use cases for classification and for content verification using question answering.
   

FAIRmedia has the potential to strengthen data sovereignty in Europe by creating datasets that are fully controlled by Austrian media organizations and developing training methods that can be used by media professionals without in-depth prior knowledge. The project thus opens up new ways for media organizations to effectively use AI technologies and contribute to the European Media Data Space.