Computational game content generation and socially responsible design: Current practices, challenges, and open problems

Computational content generation increasingly shapes digital games, visual computing, and interactive environments by automating the production of characters, faces, and environments. Alongside these capabilities, such pipelines raise persistent challenges around fairness, social bias, and representational diversity. This article presents an AI-assisted, PRISMA-informed literature review of 37 recent studies that address these social dimensions across character creation, facial modeling, and procedural environment generation. We synthesize technical approaches – including deep learning, reinforcement learning, generative adversarial networks, and optimization – together with behavioral, participatory, and culturally grounded perspectives. To reduce terminological ambiguity, we provide a consolidated cross-domain conceptual framing of fairness, social bias, and diversity in the reviewed work and map recurring operational proxy families to content types. Our analysis highlights substantial variation in how these concepts are defined and evaluated: only a small subset of studies uses explicit parity- or subgroup-performance formulations, whereas many contributions rely on domain-specific gameplay proxies, distributional/perceptual generative proxies, or qualitative and behavioral evaluation, and are infrequently supported by scalable auditing or benchmarking tools. We further differentiate how fairness concerns manifest across game contexts (e.g., outcome parity in competitively oriented procedural scenarios versus representational inclusion in avatar and face pipelines). Finally, we operationalize these proxy families and recurring limitations in three applied use-cases, illustrating where auditing can attach within end-to-end workflows and where methodological gaps remain when integrating representational and outcome-focused perspectives.