Recent research reveals that generative AI tools like OpenAI's Sora perpetuate colonial biases when visualizing historical scenes. In a study focusing on Aotearoa New Zealand, Sora was prompted to depict scenes from the 1700s to the 1860s. The resulting images consistently portrayed pre-colonial landscapes as empty wilderness and Māori figures as passive or ornamental, reflecting a European settler perspective. These depictions align with colonial-era visual traditions that justified colonization by presenting indigenous lands as uninhabited and Māori people as subjects of European civilization.
The study highlights how AI models, trained on vast datasets scraped from the internet, often mirror societal stereotypes. For instance, AI-generated images might depict pilots as men or criminals as people of color. In the case of Māori history, Sora's outputs defaulted to romanticized wilderness scenes and stylized portraits reminiscent of 19th-century European photography, overlooking the complex socio-political dynamics of the time.
These AI-generated images circulate widely on social media, often without clear indications of their artificial origins. This lack of transparency can reinforce historical inaccuracies and colonial narratives, undermining indigenous perspectives and claims to sovereignty and cultural revitalization. The study calls for a critical examination of AI's role in shaping historical narratives and urges developers to address these biases in training data and algorithms.
As AI becomes increasingly influential in how we communicate and understand history, it's crucial to recognize and mitigate its potential to perpetuate colonial stereotypes. This research serves as a reminder of the importance of inclusive and accurate representations in AI-generated content, ensuring that diverse histories and perspectives are honored and accurately portrayed.
