AI-Generated 'Australiana' Images Reveal Racial and Cultural Biases, Study Finds

AI-Generated Images Expose Racial and Cultural Biases

A groundbreaking study published by Oxford University Press has revealed that generative AI tools produce images of Australia and Australians that are riddled with bias, reproducing sexist and racist caricatures more suited to the country's imagined monocultural past 1

1

. The research, conducted in May 2024, challenges the perception of AI as intelligent, creative, and desirable, as promoted by big tech companies.

Research Methodology and Findings

Researchers used 55 different text prompts across five popular image-producing AI tools: Adobe Firefly, Dream Studio, Dall-E 3, Meta AI, and Midjourney 2

2

. The study collected approximately 700 images, which consistently portrayed an idealized version of Australia anchored in a settler-colonial past.

Stereotypical Representations of Australian Families

The AI-generated images of Australian families revealed significant biases:

• "Australian mothers" were typically depicted as white, blonde women in domestic settings 1

  1

  .
• "Australian fathers" were exclusively white, often shown outdoors or engaged in physical activities with children 3

  3

  .
• One peculiar image showed an Australian father holding an iguana, an animal not native to Australia 2

  2

  .

Racial Stereotyping and Indigenous Representation

The study uncovered alarming levels of racial stereotyping, particularly in the representation of Aboriginal Australians:

• Images of "typical Aboriginal Australian families" often depicted regressive visuals of "wild" or "uncivilized" stereotypes 1

  1

  .
• When prompted for an "Aboriginal Australian's house," AI tools consistently generated images of grass-roofed huts in red dirt, contrasting sharply with the suburban brick houses produced for "Australian's house" prompts 3

  3

  .

Persistent Biases in Updated AI Models

To assess whether newer AI models have improved, the researchers tested OpenAI's latest GPT-5 model, released on August 7, 2025. The results showed that biases persist:

• An "Australian's house" prompt generated a photorealistic image of a typical suburban home.
• An "Aboriginal Australian's house" prompt produced a more cartoonish image of a hut in the outback 2

  2

  .

Implications and Concerns

The ubiquity of generative AI tools in various platforms and software makes these findings particularly concerning. The research highlights that:

• AI-generated content can perpetuate inaccurate stereotypes and biases.
• There's a lack of respect for Indigenous Data Sovereignty, where Aboriginal and Torres Strait Islander peoples should have control over their own data 1

  1

  .

This study serves as a crucial reminder of the need for ongoing scrutiny and improvement of AI systems to ensure they do not reinforce harmful stereotypes or misrepresent diverse cultures and communities.