SDC enters into new collaboration on AI

SDC is part of the new consortium, Dansk Sprogmodel Konsortium, where the goal is to develop a language model that is even better in Danish than the existing international models, and where there is also a focus on transparent use of data.

"Generative AI is a super-exciting area, and we would like to contribute to developing a strong Danish language model that our financial institutions can also benefit from. The opportunities are enormous in that area, and it is a focus area for us to embrace the opportunities that AI provides," says Area Director Marianne Nielsen from Operation & Infrastructure Services.

The Alexandra Institute, IBM Denmark and Dansk Erhverv are leading the new consortium, and a large number of public and private actors have joined the project by opening up to make data available for the new language models. SDC has been involved from the start in this work, which in the long term will be of great benefit to the financial institutions that use SDC's solution.

"Our task is to help feed the new Danish language model with data that it can learn from. Now we have to have a dialogue with the financial institutions about exactly what we can offer. One could, for example be already public texts from websites, letter templates and so on," explains Jesper Nyvold Larsen, area director for the CTO Office, and continues:

"Enormous amounts of data must be used to create a well-functioning model. SDC and our member banks will be able to contribute a small part of this".

Jesper Nyvold Larsen points to opportunities in the future, for example on the possibility that a Danish language model will be able to generate good automatic meeting minutes from customer meetings, so that the advisers' administrative tasks are reduced, just as a Danish language model will be able to be used to build well-functioning Danish chat-bots for e.g. customer support. The possibilities are many if the quality and security behind the language models are in order.

He emphasizes that precisely the delivery of data for the new language model is crucial for the model to be a success and as capable of writing in Danish as possible. But it must be done in a proper and transparent way, so that everyone can be safe in using the model. One of the criticisms of foreign language models is that it is unclear how and what data the models are based on. Therefore, according to IBM Denmark – which is part of the consortium – it has also been essential in the new collaboration to ensure responsible AI with copyright protection and with transparency from data to results.

The next step, according to Jesper Nyvold Larsen, is to meet with representatives from the financial institutions in the SDC community and together take a closer look at what types of data and in what quantities can be contributed to the new model.

You can read more about the new collaboration on Dansk Erhverv's website here.