OpenAI's Sora exhibits significant bias in video generation

A recent investigation by WIRED has uncovered significant biases in OpenAI's video generator, Sora. The AI system has been found to promote sexist, racist, and ableist stereotypes through the videos it produces. This study involved analyzing hundreds of AI-generated videos and revealed persistent issues with representation and bias. Sora's results often depicted biased portrayals of people in various professions. For instance, all generated pilots were male, while all flight attendants were female. Many videos focused on traditional gender roles, where men filled high-ranking positions like CEOs and political leaders, while women were primarily shown in caregiving roles. The results also indicated a narrow visual representation, with most characters appearing young and conventionally attractive. Research shows that bias in generative AI models often stems from the data used for training. OpenAI acknowledges that bias is a widespread issue in the industry. The company has safety teams dedicated to researching and reducing these biases but has not provided specific details on how it plans to improve Sora's outputs. WIRED's testing also illustrated a pattern of racial representation issues. When prompted for videos depicting various racial identities, Sora struggled to produce diverse results. Additionally, the AI consistently portrayed disabled individuals as wheelchair users, reinforcing ableist stereotypes. The findings highlight how these biases could have real consequences in areas such as advertising and public perception. Experts warn that if left unaddressed, biased AI outputs could perpetuate harmful stereotypes and affect marginalized communities. Addressing the problems with AI bias is complex. Suggestions include enhancing diversity in training data and involving a wider range of experts in the development process. The focus has primarily been on improving AI performance, but researchers emphasize the need for a more inclusive approach that considers the societal implications of these technologies. As Sora becomes more widely accessible, increased pressure may lead to more serious efforts to tackle these biases.