Colonial Bias In Generative AI Imagery Affects Perceptions Of Māori Culture And History

Generative AI has transformed the way we create and view images. Tools like Midjourney, DALL-E, and Sora can generate anything from realistic photos to paintings with just a short text prompt. These images spread on social media, making it hard to tell they are artificially made. However, this ease of creation and sharing brings significant social risks.

Studies indicate that generative AI models often reflect sexist and racist stereotypes due to their training data sourced from online platforms. For instance, pilots are typically depicted as men, while criminals are shown as people of colour. My upcoming research reveals that generative AI also harbours colonial biases.

Colonial Bias in AI Imagery

When asked to visualise Aotearoa New Zealand's history, Sora tends to favour the European settler perspective. Pre-colonial landscapes appear as untouched wilderness, Captain Cook is portrayed as a peaceful civiliser, and Māori are shown as peripheral figures. Such depictions reinforce myths of benevolent colonisation and undermine Māori claims for sovereignty and cultural revival.

To investigate how AI envisions the past, OpenAI's Sora was tasked with creating scenes from New Zealand's history between the 1700s and 1860s. The prompts were intentionally open-ended to expose the model's inherent visual assumptions rather than dictate specific outcomes.

Recurring Visual Patterns

The results were consistently similar due to generative AI systems predicting probable visual elements based on their training data. For example, in Sora's depiction of "New Zealand in the 1700s," a forested valley is illuminated by golden light with Māori figures as decorative elements. The absence of food plantations or fortifications suggests a land awaiting European discovery.

This aesthetic mirrors the Romantic landscape tradition of 19th-century colonial art, such as John Gully's work, which framed the land as pristine and unclaimed (terra nullius) to justify colonisation. When prompted to portray "a Māori in the 1860s," Sora defaults to a sepia-toned studio portrait resembling late 19th-century cartes de visite photographs.

Recycling Colonial Sources

Visual imagery has historically legitimised colonisation but has faced challenges in recent decades. As part of the Māori rights movement and historical reckoning, statues have been removed, museum exhibits revised, and representations of Māori in media have shifted. Yet old imagery persists in digital archives without critical interpretation.

The exact sources of generative AI training data remain unknown; however, it's likely these archives contribute to what systems like Sora learn from. Generative AI tools recycle these sources, reproducing conventions that once supported empire-building projects.

The Role of AI Literacy

Imagery depicting colonisation as peaceful can diminish the urgency of Māori claims for political sovereignty through institutions like the Waitangi Tribunal and calls for cultural revitalisation. By portraying historical Māori figures as passive and timeless, AI-generated visions obscure ongoing self-determination movements for tino rangatiratanga and mana motuhake.

Globally, researchers are working towards decolonising AI by developing ethical frameworks incorporating Indigenous data sovereignty and collective consent. Visual generative AI presents unique challenges since it shapes perceptions of history and identity through images.

Technical solutions exist but have limitations. Expanding datasets with Māori-curated archives might diversify learning if done under Indigenous principles. Addressing algorithmic bias could balance portrayals when prompted about colonial rule; however, defining "fair" representation remains a political issue.

Filters might block biased outputs but risk erasing uncomfortable truths like colonial violence depictions. The most promising solution lies in enhancing AI literacy—understanding system operations, data sources used by models like Sora—and effectively prompting them.

Critically engaging with these systems allows us to move beyond recycling colonial tropes towards reimagining history through Indigenous perspectives—a practice already embraced by some social media users creatively exploring new narratives using AI technology.