Creating an xAI Framework for Diverse Game Characters

The rapid advancement of Generative AI (GenAI) introduces challenges in representation and diversity in games. Game companies increasingly use GenAI for brainstorming and collaboration, but AI systems trained on internet data often reflect societal biases, leading to the reproduction of stereotypes. This issue is compounded by game designers' limited awareness of misrepresentation and the lack of diversity in development teams. As the gaming audience becomes more diverse, there is a pressing need for inclusive representation. Explainable AI (xAI) offers a solution by allowing designers to understand how AI models generate outputs. xAI can help identify misrepresentations and stereotypes in outputs, particularly in character design. By revealing how AI interprets inputs, xAI can increase designers’ awareness and critical reflection, helping them create more diverse and inclusive characters. This paper proposes a framework with five objective categories for explaining game characters generated by AI: Age, Face, Body, Hair, and Style. These categories are developed through a deductive approach and pretested with seven game characters generated using DALL-E. Pretesting ensures the categories are as objective and unambiguous as possible, so even designers with limited knowledge of diversity can use them to analyze AI-generated characters. This framework is a first step towards training game artists to use GenAI effectively to create diverse and representative games. Future iterations will expand and test the categories. By integrating xAI, we will also explain to designers how to make necessary corrections in their prompts based on their own analysis, supporting the co-creation of inclusive game characters with AI.