Pixels of Prejudice: Decoding Embedded Biases in AI-Generated News Imagery and Their Implications for Visual Journalism—Toward an Algorithmic-Mediated Visual Framing

This exploratory study investigates embedded biases in AI-generated news imagery and their implications for visual journalism. It proposes Algorithmic-Mediated Visual Framing as a conceptual framework to account for the structural and functional transformations brought about by visual generative AI. Utilizing a mixed-method approach of quantitative and qualitative content and thematic analysis, the author analyzed 1200 images generated by DALL-E 3 based on 300 prompts across seven news topics. The findings revealed seven categories of bias, two recurrent visual frames, and three effective prompt refinement strategies to mitigate the biases and shift the visual framing. Theoretical and practical implications of visual framing and visual journalism are further discussed.