An official website of the United States government
Potential harms from the under-representation of minorities in data, particularly in multi-modal settings, is a well-recognized concern. While there has been extensive effort in detecting such under-representation, resolution has remained a challenge. With recent generative AI advancements, large language and foundation models have emerged as versatile tools across various domains. In this paper, we propose Chameleon, a system that efficiently utilizes these tools to augment a dataset with minimal addition of synthetically generated tuples to enhance the coverage of the under-represented groups. Our system applies quality and outlier-detection tests to ensure the quality and semantic integrity of the generated tuples. In order to minimize the rejection chance of the generated tuples, we propose multiple strategies to provide a guide for the foundation model. Our experiment results, in addition to confirming the efficiency of our proposed algorithms, illustrate our approach's effectiveness, as the model's unfairness in a downstream task significantly dropped after data repair using Chameleon.
more »
« less
This content will become publicly available on March 1, 2026
More of the Same: Persistent Representational Harms Under Increased Representation
To recognize and mitigate the harms of generative AI systems, it is crucial to consider who is represented in the outputs of generative AI systems and how people are represented. A critical gap emerges when naively improving who is represented, as this does not imply bias mitigation efforts have been applied to address how people are represented. We critically examined this by investigating gender representation in occupation across state-of-the-art large language models. We first show evidence suggesting that over time there have been interventions to models altering the resulting gender distribution, and we find that women are more represented than men when models are prompted to generate biographies or personas. We then demonstrate that representational biases persist in how different genders are represented by examining statistically significant word differences across genders. This results in a proliferation of representational harms, stereotypes, and neoliberalism ideals that, despite existing interventions to increase female representation, reinforce existing systems of oppression.
more »
« less
- Award ID(s):
- 2505865
- PAR ID:
- 10631352
- Publisher / Repository:
- https://doi.org/10.48550/arXiv.2503.00333
- Date Published:
- ISSN:
- 2503.00333
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
This study investigates how high school-aged youth engage in algorithm auditing to identify and understand biases in artificial intelligence and machine learning (AI/ML) tools they encounter daily. With AI/ML technologies being increasingly integrated into young people’s lives, there is an urgent need to equip teenagers with AI literacies that build both technical knowledge and awareness of social impacts. Algorithm audits (also called AI audits) have traditionally been employed by experts to assess potential harmful biases, but recent research suggests that non-expert users can also participate productively in auditing. We conducted a two-week participatory design workshop with 14 teenagers (ages 14–15), where they audited the generative AI model behind TikTok’s Effect House, a tool for creating interactive TikTok filters. We present a case study describing how teenagers approached the audit, from deciding what to audit to analyzing data using diverse strategies and communicating their results. Our findings show that participants were engaged and creative throughout the activities, independently raising and exploring new considerations, such as age-related biases, that are uncommon in professional audits. We drew on our expertise in algorithm auditing to triangulate their findings as a way to examine if the workshop supported participants to reach coherent conclusions in their audit. Although the resulting number of changes in race, gender, and age representation uncovered by the teens were slightly different from ours, we reached similar conclusions. This study highlights the potential for auditing to inspire learning activities to foster AI literacies, empower teenagers to critically examine AI systems, and contribute fresh perspectives to the study of algorithmic harms.more » « less
-
This study investigates how high school-aged youth engage in algorithm auditing to identify and understand biases in artificial intelligence and machine learning (AI/ML) tools they encounter daily. With AI/ML technologies being increasingly integrated into young people’s lives, there is an urgent need to equip teenagers with AI literacies that build both technical knowledge and awareness of social impacts. Algorithm audits (also called AI audits) have traditionally been employed by experts to assess potential harmful biases, but recent research suggests that non-expert users can also participate productively in auditing. We conducted a two-week participatory design workshop with 14 teenagers (ages 14–15), where they audited the generative AI model behind TikTok’s Effect House, a tool for creating interactive TikTok filters. We present a case study describing how teenagers approached the audit, from deciding what to audit to analyzing data using diverse strategies and communicating their results. Our findings show that participants were engaged and creative throughout the activities, independently raising and exploring new considerations, such as age-related biases, that are uncommon in professional audits. We drew on our expertise in algorithm auditing to triangulate their findings as a way to examine if the workshop supported participants to reach coherent conclusions in their audit. Although the resulting number of changes in race, gender, and age representation uncovered by the teens were slightly different from ours, we reached similar conclusions. This study highlights the potential for auditing to inspire learning activities to foster AI literacies, empower teenagers to critically examine AI systems, and contribute fresh perspectives to the study of algorithmic harms.more » « less
-
Families are a key institution that reproduce and resist gender inequalities. For instance, families can maintain or challenge cisnormativity—a gender structure that erases, marginalizes, and harms trans people. However, beyond studying highly supportive parents of trans children, scholars lack a full understanding of how family members divest from cisnormativity. Furthermore, overfocusing on parents ignores how children and youth, including siblings, also challenge gender norms within families. Using interviews with 52 trans youth, who are mainly trans youth of color, this article examines how siblings of trans youth divest from cisnormativity and help trans youth achieve gender recognition when parents are unsupportive or ambivalent. We find that siblings recognize and support trans youth’s gender through both passive (such as nonchalantly accepting their trans sibling) and active (such as using correct names and pronouns) gender-supportive practices. We also introduce the concept of counterhegemonic accountability to describe how siblings hold accountable family members who misrecognize trans youth’s gender. Together, siblings and trans youth challenge cisnormativity at home and within the broader society. To understand the complex ways gender norms change in and through families and within society, gender scholars need to study sibling relationships.more » « less
-
Large language models (LLMs) are increasingly integrated into a variety of writing tasks. While these tools can help people by generating ideas or producing higher quality work, like many other AI tools they may risk causing a variety of harms, disproportionately burdening historically marginalized groups. In this work, we introduce and evaluate perceptual harm, a term for the harm caused to users when others perceive or suspect them of using AI. We examined perceptual harms in three online experiments, each of which entailed human participants evaluating the profiles for fictional freelance writers. We asked participants whether they suspected the freelancers of using AI, the quality of their writing, and whether they should be hired. We found some support for perceptual harms against for certain demographic groups, but that perceptions of AI use negatively impacted writing evaluations and hiring outcomes across the board.more » « less
