Age and gender distortion in online media and large language models

★

☆

SHARE

Overview

Paper Summary

Conflicts of Interest

Identified Weaknesses

Rating Explanation

Good to know

Topic Hierarchy

File Information

Paper Summary

Paperzilla title

Online Worlds and AI are Ageist & Sexist: Women are always 'Younger' and Men are 'Older' in Pixels and Prompts!

This study reveals widespread age- and gender-related distortion across online media (images, videos, text) and large language models (LLMs). It found that women are consistently depicted as younger than men, especially in higher-status roles, and demonstrated how mainstream algorithms amplify these biases, influencing people's beliefs and even AI-driven hiring recommendations. The research combined large-scale data analysis with human experiments and LLM audits to show these biases are pervasive and disconnect from real-world demographics.

Possible Conflicts of Interest

None identified. Authors are affiliated with academic institutions (Stanford University, University of California Berkeley, University of Oxford, Autonomy Institute) and disclose academic funding sources. The study audits products from Google and OpenAI, but authors do not appear to be employees of these companies.

Identified Weaknesses

Reliance on Crowdsourced Data for Image Annotation

While the authors mention controls (three annotators, attention checks, validation tasks), human coders on platforms like Mechanical Turk can introduce their own biases or misinterpretations, even with careful design. The subjectivity of age and gender classification, even when averaged, might not perfectly reflect objective truth.

Generalizability of 'Nationally Representative US Sample' from Prolific

Although Prolific aims for national representativeness, a sample of n=459 for an experiment making claims about 'people's beliefs' might still have limitations in fully capturing the diversity of the US population, especially given the self-selected nature of survey participants.

ChatGPT Resume Audit is Synthetic

The study audits ChatGPT's generation and evaluation of synthetic resumes, not real-world applications. While valuable for isolating AI bias, it doesn't directly measure the impact on actual hiring processes or reflect the full complexity of human hiring decisions.

Observational Nature of Initial Findings

Many initial findings about age and gender distortion in online media (images, videos, text) are observational correlations, not causal. While the experimental component addresses amplification, the underlying 'distortion' is described statistically.

Focus on Binary Gender

The study primarily uses a binary classification of gender (male/female) in its analyses, acknowledging that gender is non-binary but justifying the binary approach based on previous literature and methodological consistency. This simplification might not fully capture the nuances of gender representation online.

Rating Explanation

This paper presents strong, multi-modal evidence of age and gender bias across a vast array of online data and algorithms, including LLMs and search engines. The methodology is robust, combining large-scale observational data, a human participant experiment, and an AI audit. The findings are highly significant for understanding and addressing algorithmic bias and societal inequality, despite some limitations inherent in crowdsourced data and synthetic AI experiments.

Good to know

This is our free standard analysis. Paperzilla Pro fact-checks every citation, researches author backgrounds and funding sources, and uses advanced AI reasoning for more thorough insights.

Explore Pro →