SAISMO: An open-source AI infrastructure for public media
Project
Public media organisations are using AI more and more for tasks such as transcription, translation, text generation and content recommendations. However, the available AI services largely originate from large, non-European technology companies. The SAISMO (Sovereign AI Stack for Media Organisations) project is investigating how public media organisations can gain greater control over their AI infrastructure by developing and testing open-source alternatives that align with public values such as independence, transparency and privacy. The project is funded by SIDN.
Cause
Public media organisations are increasingly using AI for tasks such as transcription, translation, text generation and content recommendations. Many of these technologies are provided by large, non-European companies such as OpenAI, Google and Microsoft. This presents challenges for public broadcasters such as the VPRO. Their journalistic and social mission is based on values such as independence, transparency and privacy, whilst the workings of many commercial AI systems are not fully transparent and data is often processed outside Europe.
In addition, these same technology companies use journalistic content to develop their own AI services. This creates a growing dependency that affects not only the autonomy of media organisations, but also their visibility, reach and role as independent providers of information.
At the same time, an increasing number of European and open-source alternatives are becoming available, such as GPT-NL and Mistral. However, many media organisations still lack the knowledge, infrastructure and practical experience to deploy these solutions effectively. In collaboration with VPRO, this project is therefore investigating how an open and independent AI infrastructure can be developed that aligns with the public values and needs of the media sector.
Goal
The aim of SAISMO is to develop a functional and transferable open-source AI stack for the editorial teams of media organisations. In addition, the project is developing a practical assessment framework that can be used to evaluate AI solutions against public values such as autonomy, transparency, privacy and sustainability.
The project aims to support public media organisations in their transition to open AI solutions that align with their social mission and are less dependent on commercial technology platforms.
Approach
The project brings together researchers, developers and media professionals to collaborate on the development and evaluation of an open-source AI stack for journalistic applications.
First, the project investigates which open-source AI models and services are suitable for common editorial tasks, such as transcription, translation, summarisation, metadata enrichment and text generation. This involves examining performance, technical requirements and costs in comparison with commercial alternatives.
A technical architecture will then be developed to enable these models to be deployed securely, reliably and scalably within media organisations. This will build on existing initiatives relating to digital autonomy and open infrastructures.
In line with this, the existing Public Value Benchmarking Framework will be further developed into a practical tool enabling media organisations to assess AI tools against public values. The framework will be tested and validated in collaboration with the VPRO editorial team. The solutions developed will be put into practice and evaluated within the VPRO editorial environment during the project.
Aimed results
The project will result in a functional open-source AI stack that has been tested in practice within VPRO and will be made available under an open licence. This will enable other media organisations to use, adapt and further develop the infrastructure.
An updated version of the Public Value Benchmarking Framework is also being developed. This offers media organisations a practical tool to assess AI solutions against public values and to balance functionality, autonomy, privacy, transparency and sustainability.
The experiences from the pilot will also be documented in a practical report containing technical lessons, organisational insights and recommendations for scaling up. A supplementary implementation guide offers other public organisations a concrete roadmap for making the transition to open-source AI solutions.
By sharing knowledge through workshops, publications and existing networks, the project contributes to a broader movement towards digital autonomy within the media and cultural sector.
Research partners
- Amsterdam University of Applied Sciences (Centre of Expertise in Applied AI)
- VPRO
- Post-X Society
- Laurens Vreekamp
- Fabienne Meijer
Network partners
- PublicSpaces
- SURF
- NPO Innovation
- Post-Tyranny Tech
- EO
- HUMAN
- EIT Culture & Creativity
Funding
SIDN