Designing with Generative AI: Ethical and Creative Redesign of Visual Assets in a Statistics Game

This paper describes the redesign of Monster Samples, a team-based, non-digital game that teaches descriptive statistics, particularly mean and median, through reverse reasoning tasks in which learners construct samples satisfying predetermined statistical conditions. The original illustrated monster cards could not be reproduced in academic publications or open dissemination because of licensing restrictions. The authors therefore used text-to-image Generative AI tools to create a new set of visual assets, framing the AI as a tool operating under deliberate pedagogical supervision rather than as an autonomous creator. The work is positioned within discussions of game-based learning in mathematics education, the role of visual representations that encode quantitative information, and responsible AI and intellectual property considerations in educational design.

The redesign was guided by explicit requirements: mathematical fidelity, statistical variability across attributes such as number of eyes, limbs and colours, combinatorial richness, visual clarity, and ethical transparency. Prompts were built incrementally and refined through repeated generate-verify cycles, with each image manually validated before inclusion. A central finding is the gap between visual plausibility and numerical precision. Generating monsters with more than three clearly distinguishable eyes proved especially difficult, and identical prompts could yield inconsistent results, so human validation was indispensable for preserving the game's conceptual clarity.

The authors identify clear practical affordances, including rapid prototyping, flexibility in adjusting attribute distributions, and independence from licensed artwork, which they argue may make non-digital mathematics games more accessible and shareable. These benefits are weighed against limitations in numerical control, prompt sensitivity and stylistic coherence, as well as concerns about opaque training data provenance and authorship. The paper concludes that generative AI can augment, but not replace, pedagogical responsibility, and it calls for future classroom implementation studies and structured frameworks for AI-supported educational game design.