Building trustworthy AI: public and private sectors join forces
Existing public- and private data creates AI and data protection
The desired consortium output is to develop a language model that over time can solve concrete and specific tasks for a companies or public organisations. To get there, it requires datasets, commitment and trust.
The Danish language model will be trained using data from consortium members, which is part of their commitment to the initiative (see commitments below). Denmark’s advanced network, cloud, and data infrastructure provide extensive datasets from both public and private sectors, ready to be utilised for AI training. This robust data foundation is key to develop effective AI solutions.
However, handling large amounts of data from the private and public sectors naturally brings up concerns about data protection. To address these issues, the Danish language model will be developed as a smaller model, thereby enhancing data protection through five key principles (detailed below). The Danish Chamber of Commerce believes this approach will encourage more companies and public institutions to adopt AI, promoting trust and safety in AI technology use.
All companies have the opportunity to become a member of the consortium, but only if they sign the principals and commitments below.
Five principals for members
To secure data protection all members of the Danish Language Model Consortium must sign five principals:
- The developed Danish language models must at all times be freely accessible to everyone.
- The utilised data and developed language models must at all times comply with Danish law and EU legislation, including GDPR in terms of the protection of sensitive personal information.
- The models must be trained on datasets approved for the purpose by copyright holders. Data must not leave the EU.
- There must be transparency regarding data sources and model training.
- Robust security protocols will be implemented to protect data and the models from potential vulnerabilities.
Member commitment
All Consortium members shall commit to the following:
- Publicity: To promote the initiative and agree to publicise their participation.
- Use cases: To share their relevant use cases within the Consortium for further development of the model. In addition, members are encouraged to share use cases with the public to inspire widespread adoption of the models.
- Data access: To share datasets on an ongoing basis with Danish Foundation Models (DFM), which can be used to train Danish models.
- Adherence to principles: To agree that the principles for the development of the Danish language models are also the foundation for a responsible implementation and use of artificial intelligence.