Generative AI for Automatic Feedback Generation in Serious Games

Abstract

This thesis investigates the use of Generative Artificial Intelligence (AI) for automatic feedback generation in serious games, focusing on identifying the characteristics of effective feedback in serious games, key design considerations, and practical implementation strategies. The narrative-based serious game "Take 5" is used for this study, employing an iterative design science methodology to develop and evaluate various prototype feedback systems. Multiple variants of an automatic feedback generation system designed around generative AI have been developed.

The study’s iterative approach includes qualitative and quantitative evaluations with expert participants, were participants played and discussed output of the implemented systems, leading to insights that refine the feedback systems across multiple iterations.

The research identifies effective feedback as actionable, specific, personalized and motivational which are crucial elements for enhancing content in serious games. Key design considerations for integrating generative AI include leveraging contextual information about the player experience and characterizing goal, employing multiprompt approaches for further consistency and relevance in the feedback provided, enhancing all identified content elements found.

The findings demonstrate that generative AI can improve feedback generation in serious games and that in 84% of the cases this feedback was preferable over the traditional already in-game feedback. This research contributes to the fields of serious games and educational technology by providing practical insights for implementing AI-driven feedback mechanisms in educational contexts.