Government backs Danish version of ChatGPT

Denmark’s Chamber of Commerce, Dansk Erhverv, is fronting a national consortium of state and private organisations to launch a Danish language version of ChatGPT.

The “responsible artificial intelligence” (AI) model advanced by the newly established Danish Language Model Consortium (DLMC) will be freely available to so-called “principled” organisations and individuals in Denmark.  

The project, which is led by Dansk Erhverv and core partners IBM Denmark and the Alexandria Institute, will initially focus on developing a Danish language model that can be customised to the individual needs of different enterprises and private users.

The Danish language chatbot being developed in parallel to the main project is intended as a problem-solving tool to address the everyday complex problems and tasks encountered by enterprises and private individuals, including the processing and filing of tax returns.

Partners in the DLMC have signed up to an agreement to ensure transparency and copyright protection for Danish data. As a result, the “small language model” developed is protected by a guardrail to enable the technology to be used in a safe and ethical manner, said Thomas Kovsted, the managing director at IBM Denmark.

“There has been exceptionally broad support for the forming of an alliance that can boost the development of a responsible Danish language model. The door is open for more like-minded partners to participate. The initiative will add new momentum to digitisation in Denmark as confidence in the Danish language model we are building grows,” Kovsted said.

The DMLC project is currently working to overcome certain hurdles to certify that the data used and the language model created fully comply with Danish law and pivotal European Union (EU) rules under the General Data Protection Regulation (GDPR) that governs the protection of sensitive personal data.

DMLC expects to develop and present the first of potentially several language models during the fourth quarter of 2024. 

The development journey for a Danish version of ChatGPT is a road laden with problems requiring innovative solutions, said Casper Klynge, the former deputy director of Dansk Erhverv. Klynge left the CoC in August (2024) to join Zscaler, a digital transformation corporation headquartered in San Jose , California .

“We realise that it will be very difficult to compete with the big language models out there. That said, there are certain areas like niche products, or small and tailored language models that solve specific problems, where progress can be made,” Klynges said.

A career diplomat, the ex-IBM executive joined the CoC in 2022. In 2017, Klynges was appointed Denmark ’s first Technology Ambassador to Silicon Valley . In his new role, Klynges will head up Zscaler’s Government Partnerships for Europe, the Middle East  and Africa (EMEA).

The DLMC project gained real traction and momentum in 2023 when the Danish government signalled it would financially support joint initiatives between state and  private sector actors to develop a Danish language chatbox. The CoC, and lead partners IBM Denmark and the Alexandra Institute, immediately began assembling a public-private consortium to finance and run an AI-driven Danish language model project which was launched in May 2024.

Central to the project, the DLMC will provide data for use in the training of Danish common language models. These will be based on the Munin model developed by Danish Foundation Models (DFM), a state-assisted organisation focused on empowering the Danish Language in the digital age.

Munin’s construct is influenced by French company Mistral AI’s open source model of the same name. Established in April 2023 by former employees of Meta Platforms and Google DeepMind, disrupter Mistral produces open source large language models that are freely available to use and customise. The company’s free open source software has positioned it as an alternative to proprietary models.

The DFM’s core mission is focused on pioneering innovation in the field of machine learning and natural language processing (NLP). These initiatives are specifically dedicated to the Danish language and maintaining open access state-of-the-art language models for the language linked to applications within both text and speech.

Common, free and responsible AI-based language models in Danish should be viewed as a national asset in bolstering continued digital development in Denmark, said Jens Kaas Benner, the head of AI Labs at the Alexandra Institute.

“The consortium’s role is to supply Danish data. The development of the basic models is planned to take place under the patronage of Danish Foundation Models. This is a very important project for Denmark ,” Benner said.

The DFM functions as a collaboration between Alexandria Institute and three third-level institutions:  Aarhus University , the University of Copenhagen and the University of Southern Denmark .

Alexandria is one of seven Danish government-approved research and technology Organisations (RTOs) tasked with both developing and delivering expertise to state and private enterprises across seven primary next generation technology area, including AI, computer vision, cyber security, digital sustainability, digital health, internet of things (IoT) and smart products.

The Danish government views the DLMC project as having the potential to strengthen Denmark’s business competitiveness, drive economic growth while supporting open, research and education in the Danish language, said Marie Bjerre, Denmark’s minister for digital government and gender equality.

“This project has brought innovative forces in business, industry and the public sector together to invest in developing a Danish-based language model. The willingness to invest is a critical component of Denmark ’s ambition to reap the benefits of technology and protect our language and culture,” Bjerre said.

On the branding side, the DLMC is exploring possible names for the Danish language model, including MyGPT? and DanGPT.

The roll call of the 20-plus partners in DLMC includes Aarhus Municipality; the Danish pension insurance group ATP; the Bagger Sørensen Group; Cbrain; rescue and emergency group Falck; JN Data; Denmark’s National Association of Municipalities; energy company Norlys; Topdanmark; software developer Visma; and SDC, a supplier of core digital platforms to banks across the Nordic region.

Although OpenSource and generally available to private and public users, the DLMC has implemented measures to prevent piracy and copyright infringement of the specific methods and data configurations used in creating the language model.  

Foremost, the datasets on which the language model is demonstrated and tutored will not be made publicly available. These datasets will remain confidential and protected by restricted access criteria. The DLMC will, however, permit partners to improve on the basic language model made available for training purposes, enabling partners to add further value and customise them to their specific needs.

As a founding partner in DLMC, Topdanmark plans to introduce the Danish language model based chatbots across its entire financial services business. Topdanmark was among the first Nordic finance houses to apply AI technologies when it launched its chatbot Globus in 2019. Globus is currently used to handle over 80% of the group’s inquiries from customers relating to travel insurance.