<b>Glitched Realities</b>: Biases in AI image generation

The problem of biases in AI image generation roots from wide dependency on AI tools such as DALL-E, midjouney, Adobe Firefly and Stable diffusion to generate image based content. These tools use generative artificial intelligence to create images based on text prompts. The image generators are trained on massive, often uncurated datasets picked from the internet reflecting real world biases. The data in reference reflects online and real-world media content, making the AI models often absorb and replicate cultural, racial and gender biases present in that material (Skilton and Cardinal, 2024). The inner working of these models are largely opaque to regular users, making it difficult to predict or manage their biased outputs.<br>