A political leader filled with rage, riling up a crowd with a message of hate. A child, tears streaming down her face, grieving the loss of her family in a massacre. Emaciated prisoners, barely hanging onto life, punished simply for their identity. These vivid images, etched into our minds through documentaries, news stories, and history classes, remind us of the world’s harsh realities, painting a picture of past events we never witnessed firsthand.
Images like these hold crucial significance in our understanding of history, shaping our learning and guiding our steps into the future. They capture truths of eras long gone, truths that we might otherwise have missed.
As professionals who sift through archives for documentaries, we — the co-directors of the Archival Producers Alliance — are alarmed by the growing uncertainty surrounding the authenticity of these images. We’re not alone in this concern. Prominent voices in the film industry, including major players like the Motion Picture Academy, are pondering the implications of generative AI and its role in filmmaking, even considering mandating disclosure of AI usage in their works.
While this is relevant for feature films, the stakes are particularly high for documentaries. By the spring of 2023, synthetic visuals and audio started appearing in historical documentaries we were involved with. Without a clear standard for transparency, this blending of real and artificial threatens to undermine the nonfiction genre, which plays a vital role in preserving our collective memory.
In February 2024, OpenAI introduced its text-to-video platform, Sora, featuring a clip titled “Historical footage of California during the Gold Rush.” The video was realistic, showcasing a picturesque scene of a bustling town with hopeful miners. Yet, it was entirely fabricated.
Sora, which was officially launched in December 2024, illustrates how it can create content based on user prompts, simulating reality. However, this clip was far from reality, mixing genuine elements with Hollywood-inspired imaginations and historical biases. Platforms like Sora, Runway, and Luma Dream Machine source their content from the vast digital world, often perpetuating existing biases. Such convincing visuals highlight how easily audiences can be misled by the magic of cinema.
In the film industry, some embrace generative AI enthusiastically, while others share our concerns about its implications. If the trust in visual authenticity crumbles, powerful documentaries might lose their credibility, even those devoid of AI input.
Like the nutritional labels on our groceries, transparency about AI usage might serve as a partial solution. Yet, regulations for AI disclosure seem frustratingly distant.
The promise of generative AI to democratize audio-visual creation is worrisome, especially concerning historical representation. The surge in synthetic images escalates the responsibility of documentarians and researchers to safeguard factual accuracy — a human task irreplaceable by automation. Take, for example, the documentary “Sugarcane,” nominated for an Oscar this year. It demonstrates the importance of meticulous research, authentic archival footage, and compelling narratives to uncover tales, like the tragedies faced by First Nations children in Canadian residential schools.
With AI models continuously advancing and content production quickening, ignoring this technological wave is not an option. While it’s entertaining to experiment with these tools, their outputs lack the essence of authentic human documentation; they’re merely digital mashups.
Our response should include reinforcing AI media literacy both within the industry and among the public. At the Archival Producers Alliance, we’ve developed guidelines for the responsible use of AI in documentary filmmaking, supported by over 50 industry organizations. Additionally, we’re collecting case studies on AI’s application in documentaries to help ensure that the genre remains true to its title, and that its impact on our shared history remains authentic.
In this era, no savior on horseback will protect us from the unchecked rise of generative AI. It’s up to each of us to work collectively to maintain the integrity and diverse perspectives of documented history. True visual records don’t just capture the past; they help us understand and believe in it, providing lessons and insights crucial to our future.
Absent the ability to faithfully witness the events that shaped us, the future we craft together could become nothing more than a distorted remix of the past.
Rachel Antell, Stephanie Jenkins, and Jennifer Petrucelli head the Archival Producers Alliance.
