Research Summary: She Works, He Works: A Curious Exploration of Gender Bias in AI-Generated Imagery

A research study titled 'She Works, He Works' integrates feminist visual theory, artistic practice, and AI critique to investigate gender bias in generative text-to-image systems. Evaluating model responses to prompts depicting construction workers, the analysis demonstrates systematic disparities where male figures are associated with authority and professionalism, while female figures are portrayed through lenses of aesthetics and agility, establishing generative AI as a cultural actor that actively perpetuates occupational stereotypes.

Why it matters

As generative visual AI becomes integrated into corporate communications, marketing, and media workflows, unaddressed demographic biases present tangible reputational and brand risks. The evidence of automated stereotyping in basic occupational prompts underlines the necessity of comprehensive bias audits and algorithmic evaluation frameworks. Organizations relying on synthetic media must understand these embedded biases to avoid inadvertently compromising diversity standards and stakeholder trust.

Key insights

• Generative text-to-image models reproduce entrenched societal gender biases within occupational contexts such as construction work.
• Image generation outputs systematically emphasize aesthetics and agility for women, compared to authority and professionalism for men.
• The research frames generative AI systems as active technological and cultural participants capable of reinforcing visual stereotypes.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2407.18524