Representation Bias of Adolescents in AI: A Bilingual, Bicultural Study
Project Overview
The document examines the representation bias of adolescents in generative AI, revealing how these models often misrepresent teenagers by linking them to societal issues, influenced by sensationalized media narratives. It highlights the disconnection between AI-generated outputs and the actual experiences of adolescents in the U.S. and Nepal, suggesting that AI frequently perpetuates stereotypes rather than reflecting the true diversity of youth identities. The study advocates for participatory approaches in AI development, emphasizing the need to incorporate adolescents' self-perceptions to achieve fairer representation. Furthermore, it discusses the potential of AI in addressing misconceptions about teenagers by fostering more positive and diverse portrayals, ultimately aiming to shift the narrative around youth in society.
Key Applications
Static Word Embeddings (SWEs) and Generative Language Models (GLMs)
Context: Educational contexts involving adolescents, particularly in understanding their representation in AI and media.
Implementation: Utilized SWEs and GLMs to generate textual outputs based on prompts about teenagers and analyze biases in representation.
Outcomes: Found significant bias in AI outputs associating adolescents with negative traits such as violence and rebellion, while actual adolescent perspectives focused on diversity and everyday experiences.
Challenges: The challenge lies in the sensationalized nature of media training data, which biases AI outputs and misrepresents adolescents' realities.
Implementation Barriers
Societal Bias
AI models absorb biases from sensationalized media portrayals of adolescents, which focus on violence, drug use, and other societal problems. Additionally, training data often comes from media sources that do not accurately represent the diversity and complexity of adolescent experiences.
Proposed Solutions: Encouraging participatory design approaches that involve adolescents in the training and development of AI to better reflect their realities. Developing frameworks for ethical engagement and ensuring data sources include adolescent perspectives directly.
Project Team
Robert Wolfe
Researcher
Aayushi Dangol
Researcher
Bill Howe
Researcher
Alexis Hiniker
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Robert Wolfe, Aayushi Dangol, Bill Howe, Alexis Hiniker
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai