The Silent Curriculum: How Does LLM Monoculture Shape Educational Content and Its Accessibility?
Project Overview
The document examines the role of Large Language Models (LLMs) in education, focusing on their potential to influence children's understanding of cultural diversity and occupational stereotypes through what is termed the 'Silent Curriculum.' It warns against an AI monoculture that may arise from similar biases across various models, potentially skewing the information children are exposed to. The study showcases instances where LLMs generate children's stories that inadvertently reinforce implicit biases and stereotypes related to occupations and ethnicities, underscoring the crucial need for incorporating diverse perspectives in the training datasets used for AI. By addressing these issues, the document calls for a more conscientious approach to utilizing generative AI in educational settings to ensure that it fosters a more inclusive and accurate representation of cultural narratives and occupational roles.
Key Applications
Children's story generation using LLMs
Context: Educational context focusing on children's literature and storytelling; target audience includes children and educators.
Implementation: Utilized GPT-3.5 and LLaMA2-70B to generate narratives based on prompts that encourage diversity while examining implicit biases.
Outcomes: Identified cultural representation biases in AI-generated stories; observed a preference for certain ethnicities in occupational roles, which may influence children's perceptions.
Challenges: Potential for reinforcing stereotypes; difficulty in ensuring diverse representation in AI outputs.
Implementation Barriers
Cultural Bias
LLMs may propagate existing stereotypes and biases through their outputs, leading to a narrow understanding of cultural diversity.
Proposed Solutions: Encouraging diverse training datasets and implementing bias detection and correction mechanisms in AI models.
Project Team
Aman Priyanshu
Researcher
Supriti Vijay
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Aman Priyanshu, Supriti Vijay
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