Artificiell intelligens, ett paradigmskifte för arkiven? : En diskursanalys av hur Riksarkivet skriver om AI på sin webbplats

Sammanfattning

On the 4th of July 2024 the Swedish government decided to assign the Agency for Digital Government and the Swedish Authority for Privacy Protection to give guidelines for increasing use of generative artificial intelligence in the public sector. One year later, on the 10th of September 2025, the government released their budget proposition for the year 2026. Here the National Archives of Sweden (NAS) was assigned 20 million Swedish crowns for the year to digitize archival materials and make possible the training of AI-models on Swedish data. This will influence how the archive describes this technology and since the NAS have an expertise on information management, it is relevant to understand in what ways they describe AI on their website. This thesis therefore uses a discourse analysis to study how posts on the website of the NAS describe AI. This is coupled with a literary review regarding AI and archives, where the resulting different subjects and subject groups are created to present different descriptions of AI than those displayed in the NAS’ posts. The conclusion of this thesis is that the NAS’ posts display two discourses: an archival official AI-discourse and an archival official data governance discourse. The first one strengthens the NAS’ role as an authority where AI is a positive force, improving the accessibility for the general public and researchers which leads to innovation and effectivity. The second one presents how AI’s improving role benefits other authorities, agencies, institutions and users who can have use of the data and AI-models published. In the first discourse, the NAS presents itself as a part of the larger society, where democratization is mentioned and where the archives are mentioned as shared, which I interpret as an attempt for NAS to legitimize its use of AI. The literary review resulted in four categories: 1) improvements as a result of AI, 2) risks and setbacks as a result of AI, 3) needs as a result of AI, and 4) potentials as a result of AI. The comparison between the discourse analysis and the literary review showed that the NAS, while mentioning improvements and potentials as a result of AI, was mostly silent on needs and risks and setbacks.