Age and gender distortion in online media and large language models - Nature

This research article by Guilbeault and colleagues presents a comprehensive investigation into age-related gender bias across digital platforms using multiple methodological approaches.

The study examines how online representations systematically portray women as younger than men across images, videos, and text through observational analyses of major platforms (Google, Wikipedia, YouTube, etc.), controlled experiments with human participants, and algorithmic audits of ChatGPT. Using crowdsourced annotations from over 6,000 coders, machine learning classifications, and word embedding analyses, researchers analyzed more than one million images and extensive text corpora to identify consistent patterns of age-gender bias.

• Reveals pervasive bias where women are consistently depicted as younger across Google Images, Wikipedia, celebrity datasets, and training corpora for facial recognition systems
• Demonstrates through controlled experiments that exposure to biased online images influences human age perceptions and hiring decisions
• Shows that large language models like ChatGPT generate resumes with age disparities that favor younger women and older men across occupations
• Validates findings across multiple data sources, methodologies, and platforms, confirming the systematic nature of age-related gender bias in digital content