Back

Documentary 'Ghost in the Machine' Examines Racial Roots of Generative AI

Background and Motivation

Valerie Veatch, a filmmaker, first noticed OpenAI’s Sora text‑to‑video generative AI model when it was released in 2024. While she was curious about the creative possibilities, she quickly observed that the model frequently produced images that were steeped in racism and sexism. The lack of concern from fellow AI‑enthusiast peers deepened her unease and prompted her to investigate the underlying causes of these harmful outputs.

Historical Roots of Generative AI

Veatch’s documentary, Ghost in the Machine, argues that the problems seen in modern generative AI are not new. It traces a lineage that begins in Victorian‑era England with the development of eugenics by Francis Galton, a cousin of Charles Darwin. Galton’s belief in improving humanity through the elimination of “inferior” races informed the early social sciences. His work on multidimensional modeling influenced Karl Pearson, who later created statistical tools such as logistic regression—techniques that now form a core part of machine‑learning algorithms.

The film also highlights that the term "artificial intelligence" was coined in 1956 by computer scientist John McCarthy as a marketing phrase to attract funding, a label that has persisted despite its vague meaning.

Veatch’s Personal Encounters with Bias

While experimenting with an early version of Sora in an artists’ Slack channel, Veatch witnessed a woman of color’s attempts to generate images of herself being white‑washed. The model retained her braids and fashion but placed her in a “white space” gallery, prompting Veatch to react with disbelief. She contacted OpenAI to report the racist, sexist, and misogynistic outputs—such as images that added extra breasts or forced a twerking pose—but received a dismissive response that the issue could not be changed.

Documentary Content and Perspective

Ghost in the Machine brings together AI researchers, historians, and critical theorists who argue that the bias in generative AI is rooted in its historical connections to scientific fields built on discriminatory worldviews. Veatch refuses to present a flattering portrait of industry leaders, stating she would not hug Sam Altman on camera because that would become propaganda.

Release Information

The documentary will be available for streaming on Kinema from March 26th to March 28th, after which it is slated to air on PBS in the fall. A correction note in the original article clarifies that the proper term is "logistic regression," not "logical regression."

Used: News Factory APP - news discovery and automation - ChatGPT for Business

Source: The Verge

Also available in: