UW Study Reveals AI Models' Negative Portrayal of Teenagers, Highlighting Cultural Biases

AI Models Show Bias in Portraying Teenagers

A groundbreaking study by the University of Washington has uncovered significant biases in how artificial intelligence (AI) systems portray teenagers. The research, led by Information School doctoral student Robert Wolfe and a UW team, examined two common open-source AI systems trained in English and one in Nepali, revealing stark differences in their representations of teens 1

1

Negative Associations in AI Responses

The study found that in English-language systems, approximately 30% of AI responses about teens referenced societal problems such as violence, drug use, and mental illness. In contrast, the Nepali system showed fewer negative associations, with only about 10% of responses containing such references 2

2

Disconnect Between AI Portrayals and Teen Self-Perception

Researchers conducted workshops with teens from the U.S. and Nepal, discovering a significant disconnect between how AI systems portrayed teenagers and how teens viewed themselves. Wolfe stated, "The way teens continued the prompts we gave AI models were incredibly mundane. They talked about video games and being with their friends, whereas the models brought up things like committing crimes and bullying" 1

1

Media Influence on AI Training Data

The study suggests that the skewed portrayal of teenagers in AI systems stems partly from the abundance of negative media coverage about teens. News stories, often considered "high-quality" training data due to their factual nature, frequently focus on negative stories rather than the everyday experiences of most teens 2

2

Cultural Differences in AI Representation

The research also highlighted cultural differences in how teens want to be represented in AI systems. U.S. teens emphasized the importance of diversity, while Nepalese teens suggested that AI should present them more positively 2

2

Related Stories

64% of Teens Use AI Chatbots as Schools Grapple With Cheating and Learning Impact

The AI Generation Gap: Teens Embrace AI Tools While Parents Remain in the Dark

72% of U.S. Teens Use AI Companions: Study Reveals Widespread Adoption and Potential Risks

Teens Increasingly Turn to AI Companions for Advice and Friendship, Raising Concerns

Call for Changes in AI Training

Senior author Alexis Hiniker, a UW associate professor in the Information School, emphasized the need for significant changes in AI model training. She advocated for a community-driven approach that incorporates teens' perspectives and everyday experiences as the primary source for training these systems 1

1

Implications for Future AI Development

While the study focused on open-source systems that may not represent the most current versions, the researchers warn that upstream biases can persist implicitly in more advanced models. This persistence could affect outputs as AI systems become more integrated into people's lives, influencing decisions in schools or personal recommendations 2

2