Umeå University's logo

Over the past few years, generative AI has become popular for creating images and text. AI technology has become a staple in modern society, and its tools have grown to be of daily use in both personal and professional settings. However, concerns have been raised about how these AI systems might reinforce societal biases and stereotypes. This study, "Kodade normer: En kvalitativ och kvantiativ studie om bias i generativ AI " investigates whether and how generative AI reproduces biased or stereotypical results using three AI tools: Copilot, Stable Diffusion, and AIEASE. 

Combining semiotic analysis and a quantitative analysis, we examine AI-generated images based on five broad and neutral prompts: Successful person, Happy person, Truck driver, Nurse, and Ambitious person. The broad prompts are intended to compel the AI models to make their own choices, allowing us to analyze their representation decisions.

Our findings reveal clear patterns of bias in generative AI as well as a tendency to assign stereotypical norms concerning success, ambition, gender and ethnicity. The results of this study mirrors the pre-existing awareness of generative AI’s problematic reinforcement of societal norms and that they are far from stereotype-free. Future research should explore a broader range of AI tools, examine changes in AI biases over time, and compare visual and text-based generative models to uncover broader patterns of representation.