TLDR: New research indicates that generative AI models, such as Sora, are perpetuating colonial stereotypes when creating historical images, particularly when visualizing Aotearoa New Zealand’s past. These AI tools, trained on existing online data, tend to privilege European settler viewpoints, depicting pre-colonial landscapes as empty and indigenous populations as peripheral, thereby naturalizing myths of benevolent colonization.
A recent study has uncovered that artificial intelligence, specifically generative AI models like Midjourney, DALL-E, and Sora, are inadvertently recycling and reinforcing colonial stereotypes in their historical image generation. The research, soon to be published, highlights a significant social risk associated with these increasingly influential tools, which draw on vast amounts of training data scraped from online and other digital sources.
The core issue stems from the AI’s tendency to mirror existing biases present in its training data. This phenomenon has previously been observed in AI models perpetuating sexist and racist stereotypes, such as portraying pilots predominantly as men or criminals as people of color. The new findings extend this concern to historical representations, revealing a distinct colonial bias.
When prompted to visualize Aotearoa New Zealand’s history between the 1700s and 1860s, OpenAI’s Sora model consistently privileged a European settler viewpoint. The AI rendered pre-colonial landscapes as ’empty wilderness,’ depicted Captain Cook as a ‘calm civiliser,’ and cast Māori as ‘timeless, peripheral figures.’ These depictions are not merely aesthetic choices; they carry profound implications.
Experts warn that such AI-generated imagery can naturalize myths of benevolent colonization and undermine critical indigenous claims to political sovereignty, redress through institutions like the Waitangi Tribunal, and efforts for cultural revitalization. By effectively recycling historical sources, these tools reproduce conventions that historically served the project of empire, potentially blunting the perceived urgency of Māori claims.
In response to these challenges, researchers and communities globally are actively working to decolonize AI. This involves developing ethical frameworks that embed Indigenous data sovereignty and collective consent in AI development. While filters could block the most biased outputs, they also risk erasing uncomfortable truths, such as depictions of colonial violence.
Also Read:
- Experienced Professionals Lead AI Adoption, Challenging Generational Assumptions in the Workforce
- ADL Report: Generative AI Video Tools Struggle to Block Antisemitic Content
The most promising path forward, according to the research, lies in fostering AI literacy. A deeper understanding of how these systems ‘think,’ the data they rely on, and how to effectively prompt them is crucial. When approached critically and creatively, AI has the potential to move beyond merely recycling colonial tropes and instead become a medium for re-seeing the past through Indigenous and other diverse perspectives.
