Exploring Normativity in Stable Diffusion: Insights for XAI in the Arts

AI systems encode hidden biases that constrain creative expression: The study found that Stable Diffusion exhibited "normative behavior" through invariant patterns and stereotypical outputs, meaning the system repeatedly produced similar results regardless of artist intent, leading participants to feel their creative vision was being compromised rather than enhanced.

Lack of transparency about how AI makes decisions undermines artist control: Participants experienced "unsolicited omission or addition of details" without understanding why the system made these choices. The researchers call for "prompt transparency" so artists can better understand and predict how their instructions will be interpreted by the AI.

Current AI tools can leave creative professionals feeling disempowered: Rather than serving as creative collaborators, the study documents that artists using Stable Diffusion reported feelings of "disempowerment and creative compromise," suggesting that generative AI systems need better design to support rather than constrain human creativity.

Task complexity affects how much AI systems limit creative freedom: The study found that normative behavior impacted practitioners differently depending on task specificity—suggesting that AI systems may be particularly problematic when artists need flexibility and nuance, rather than when following highly specific instructions.

Explainability matters differently in creative contexts than other AI applications: The research highlights that "XAI in the Arts" requires different approaches than traditional AI explainability, focusing on artist empowerment and creative agency rather than just technical transparency.