Bias In, Bias Out? Risks and Opportunities of Generative AI in Educational Game Authoring

Generative Artificial Intelligence (AI) is increasingly being integrated into educational game authoring tools, making interactive learning content creation more accessible to educators without programming expertise. However, these AI systems risk perpetuating and amplifying social biases present in their training data, particularly regarding gender, race, and cultural representation. This paper examines the dual nature of AI-powered educational game authoring – as both an opportunity for inclusive storytelling and a potential source of stereotypical content. Through analysis of an AI-based quest generator, we observed systematic biases where male characters were assigned leadership roles while female characters appeared as assistants, scenarios defaulted to Western settings, and diverse backgrounds were underrepresented unless explicitly requested. Despite these risks, we found that large language models can identify and correct bias when appropriately prompted, suggesting untapped potential for supporting inclusive content creation. We propose four design strategies for embedding diversity-aware features into AI authoring workflows: automated stereotype detection, generation of inclusive alternatives, reflective prompts for educator empowerment, and integrated generation-critique processes. Rather than treating bias mitigation as an afterthought, our approach advocates for embedding inclusive design principles directly into the content generation pipeline – from the initial ideation phase through to final output. This work contributes to the development of AI authoring tools that actively support educators in creating narratives that reflect diverse identities and experiences, ultimately empowering all learners to see themselves represented in educational content.