Akademik Veri Yönetim Sistemi
TURAN-SAM Uluslararası Bilimsel Hakemli Dergisi, cilt.17, sa.68, ss.239-248, 2025 (Hakemli Dergi)
This study examines algorithmic bias in visual outputs generated by contemporary generative artificial intelligence models such as DALL-E 3, Midjourney, and Stable Diffusion. Using a qualitative research design, 450 images were analyzed through content analysis, semiotic interpretation, and critical discourse analysis. Findings reveal that these models systematically reproduce social hierarchies related to gender, race, class, and cultural identity. Leadership and professional roles are predominantly represented by male and Westernlooking figures, while women and non-Western identities are frequently associated with stereotypical or secondary roles. The study also reveals that visual output is shaped by imbalanced training datasets and culturally dominant aesthetic norms, demonstrating that generative AI is not neutral but exists within broader sociotechnical power structures. The research emphasizes the need for more inclusive data practices, transparent model development, and enhanced algorithmic literacy to reduce representational inequalities in AI-generated visuals.