This dissertation examines the mechanisms driving the geographical diffusion of Artificial Intelligence through four papers. Despite the significant attention that Artificial Intelligence has received in recent years and its prominence in both public and research discourse, there are still many aspects of Artificial Intelligence where we lack the necessary knowledge. There is a generally established consensus that the diffusion of new technologies, which refers to their geographical spread and adoption among companies and individuals, has a stronger societal impact than the
invention of new technologies. However, researchers and policymakers primarily focus on the latter, leaving a significant knowledge gap regarding the mechanisms underlying the potentially uneven diffusion of Artificial Intelligence. This dissertation aims to address this knowledge gap by examining the following research question: How does the regional context influence the rate and direction of the diffusion of Artificial Intelligence across regions?
The dissertation focuses mainly on traditional/statistical Artificial Intelligence, typically used to create algorithms to make predictions, recommendations, and decisions from outside a given data set. The dissertation approaches how AI technology diffuses spatially from an evolutionary economic theoretical perspective. This entails that how individuals and firms in different regions learn about and ultimately adopt new technology is geographically path dependent, meaning that technology adoption is influenced by several local factors - e.g., regional institutions, resources, capabilities, and the technology already in use - that are self-reinforcing over time. The existing resources in the region, such as worker skills and their experience working with different types of technology, influence the knowledge and learning about new technology that firms can engage with, which in turn affects the demand for diverse resources, experience, and knowledge related to adopting and using the new technology, and thus the mutual relationship continues. The dissertation tests this assumption in the dissertation’s four papers focusing on the Danish case.
Paper A serves as a preliminary study for the thesis as it develops a new regional classification used as the regional scale for the remaining three articles. The main contribution of the article is the use of long time series of different economic and demographic variables related to regional development combined using clustering analysis to develop a more nuanced picture of regional groupings that account for development trends and, thereby, the mechanisms that create different long-term regional development.
Article B is coauthored with Jacob Rubæk Holm and examines regional differences in Artificial Intelligence adoption using data from the TASK survey on AI usage among Danish firm employees. It demonstrates, firstly, that there are regional differences in the degree to which companies adopt Artificial Intelligence and, secondly, that these differences can be explained, among other things, by regional differences in how companies learn and innovate. These characteristics can be described as the firms’ so-called "innovation modes". Specifically, the paper finds that Old industrial regions fall behind the metropolitan regions, likely because they, among other things, are influenced more by the innovation mode based on internal experience generation.
Article C studies whether regional institutions and regional adoption of Artificial Intelligence co-evolve over time. Specifically, the article examines regional informal institutions in the form of regional technological discourse in news media and their mutual evolution with the regional adoption of Artificial Intelligence. The article draws on newly collected data on different regions’ consumption of newspaper articles about Artificial Intelligence. It measures the characteristics of the regional informal institution through the tone and angle in the articles via sentiment analysis. The article demonstrates first that there are regional differences in how Artificial Intelligence is described and, second, that the newspaper article coverage co-evolves with the regional adoption of Artificial Intelligence over a period of almost twenty years.
Article D zooms in on small and medium-sized enterprises within the manufacturing sector outside the metropolitan areas and examines their absorptive capacity concerning Artificial Intelligence. The article is the dissertation’s only qualitative study and uses semi-structured interview data with SMEs in the process of adopting Artificial Intelligence. The article finds that manufacturing SMEs outside the metropolitan regions face particular challenges when they wish to start implementing Artificial Intelligence. Some of the main challenges are a mixture of the lack of relevant skills and difficulties in attracting qualified labor, making AI less abstract and easier to introduce in their business models, overcoming conservatism in the organization, finding inspiration from like-minded companies, and finding information about both new technologies and grant opportunities. However, the firms develop methods to overcome resource scarcity by building on their existing capabilities.
Overall, the dissertation contributes to our understanding of factors that influence an uneven geographical spread of Artificial Intelligence technology. The dissertation demonstrates, by studying the Danish case, that there are barriers to adopting Artificial Intelligence specific to different regions. Different regions have different resources but also cultures, traditions, and institutions of innovation and technology adoption. The dissertation also demonstrates that Artificial Intelligence is not a homogeneous technology, as it has typically been treated in previous studies. This means, as discussed in the Synopsis and illustrated in the various articles, that the observed patterns differ depending on how we treat and define Artificial Intelligence. Still, the conclusion that Artificial Intelligence takes on an unequal geographical spread pattern remains. The dissertation finally argues for greater focus among politicians developing technology policy to take into account that AI
technology usage and the driving mechanisms behind it, are, as shown in this dissertation, regionally specific. The dissertation furthermore argues for continued research efforts among researchers to untangle the mechanisms behind the spatially uneven distribution of AI.