Megagames have been used within a serious game context as tools to enhance learning. The megagame “Switching the current” (STC) is one such example, which has been designed by researchers collaborating from the Swedish universities Linköping University, Jönköping University and Högskolan i Skövde. The research team uses it as a platform to educate participants about the Swedish power grid, the involved stakeholders and the challenges involved with implementing a green transition. STC has been a multi-year project which has been iterated upon continuously. One important source for rating how effective a megagame has been and which can be used in guiding future iterations of the game, is the debriefing sessions. Due to the large number of participants in megagames these debriefings are done in smaller discussion groups, and as such it can be hard to make general summaries of the contents of these distributed debriefings. While this has been done, it is a time-consuming process which means that any insights gained will be had several days after a concluded game session. As such it has been impossible to give immediate feedback to participants, which could potentially improve the learning potential of the games. One idea, which this thesis explores, is to speed up the process of creating these summaries by using large language models (LLMs), an AI technology which can process language.

The study interviewed members of the STC research team about what importance the debriefings had for different stakeholders. Then an iterative process developed a program and prompt which could generate summaries which fit the criteria outlined in the interviews. As outlined by best practices in prompt engineering and when working with generative AI, it was important that the facilitators would be able to properly validate the contents of the debriefing summaries. Two summaries were created, one participant-oriented and the other oriented towards the game design team. To test the quality of the summaries a survey was sent out to people who had experience of facilitating STC.

It is important to note the limitations of the study, with a small number of participants and only looking at one specific megagame. It is also noteworthy that the program has not been tested in connection to an actual megagame session. However, the results of the study indicate that the summaries could be a useful addition to the facilitation of megagames and could enhance the learning of participants, as well as being a potent tool for the game design team in iterating upon the game.