A trained eye : Hybrid intelligence in data-driven decision-making

Sammanfattning

Every day we face situations where we are required to make decisions, and to make a good decision we turn to an emerged field for decision support, artificial intelligence (AI). In other words, large language models (LLM), with the impact they have had, problems have also arisen.To address this, the purpose of this thesis was to explore how large language models can be used to support data-driven decision-making in organizations. To do this a conceptual framework with LLM and RAG was tested on a case regarding fraud in public organizations, since there is need to address the advantages and negatives whit the use of AI in decision-making. The methodology that was used for this thesis is design science research and through this an artifact was produced whit the help of DECAS and hybrid intelligence. The artifact was implemented, demonstrated and evaluated with the help of data and a decision-maker from a municipality. The results of the thesis showed that it is possible to implement the conceptual framework. It showed the importance of preprocessing the data before using it, defining the role of humans and AI in cooperation and investigating the decision-making process where the framework should be implemented. Additionally, identifying the correct data to support the decision-making process in the specific context and designing the prompt.