AI Video Tool Sora Sparks Controversy Over Biased and Stereotypical Outputs
Staff Editor OpenAI’s Sora Faces Criticism for Bias in Video Generation
OpenAI’s recently unveiled video generation tool, Sora, has come under scrutiny for exhibiting racial, sexist, and ableist biases in its outputs. This development adds to the growing concern over generative AI models perpetuating harmful stereotypes, particularly in visual media.
A recent investigation by Wired magazine has shed light on the extent of these biases. Using 25 carefully crafted prompts, researchers tested Sora’s ability to generate diverse and inclusive content across various scenarios, job roles, and identity aspects.
The investigation revealed significant gender and occupational biases. Sora consistently failed to depict women in traditionally male-dominated professions such as pilots and CEOs, instead defaulting to stereotypical roles like flight attendants and nurses. Additionally, representations of gay couples predominantly featured conventionally attractive white men.
Racial biases were also evident in Sora’s outputs. The tool struggled to accurately represent racial diversity, often limiting its depictions to either Black or white individuals. Prompts for interracial couples yielded inconsistent and often stereotypical results, with a tendency to default to Black couples.
Sora’s handling of disability and body type representation raised further concerns. The tool’s portrayal of disabled individuals was largely limited to stereotypical wheelchair depictions, failing to capture the broader spectrum of disabilities. Similarly, prompts for “a fat person running” produced inaccurate and potentially offensive results.
Experts in the field have warned of the real-world implications of such biased AI-generated content. There are growing calls for improved safety and ethical measures in AI development to address these issues.
In response to these findings, OpenAI has acknowledged the industry-wide challenge of bias in AI models. The company stated it is actively researching ways to adjust training data and prompts to reduce bias, emphasizing its commitment to safety and ethical AI development.
As AI-generated visual content becomes increasingly prevalent, the tech industry faces mounting pressure to address these biases and ensure fair and accurate representation across all demographics.
