Women portrayed as younger than men online, and AI amplifies the bias (newsroom.haas.berkeley.edu)

A large-scale study published in Nature analyzed 1.4 million online images and videos, billions of words of web text, and outputs from nine large language models to document a pervasive age–gender distortion: women are systematically portrayed as younger than men across 3,495 occupational and social categories. Using mixed methods—human crowd annotations, objective age cross-references (birthdates vs. timestamps), machine estimates, and text analysis of corpora like Reddit, Google News, Wikipedia and Twitter—the researchers (Berkeley, Stanford, Oxford) found the gap widens for high-status, high‑pay roles. The effect shows up visually and in language, and is robust across platforms (Google, YouTube, Flickr, IMDb, Wikipedia) and models (including GPT-2 and gpt-4o-mini). Crucially, algorithms amplify this cultural distortion in ways that affect decisions: in experiments, ~500 participants exposed to occupation images adopted biased age expectations, and gpt-4o-mini—when generating ~40,000 resumes across 54 occupations—assigned women younger ages (≈1.6 years), more recent graduations, less experience, and later rated older men as more qualified. The study reveals a feedback loop where biased training data and model behavior reshape social beliefs and hiring-relevant inferences. For the AI/ML community this signals a pressing need for dataset auditing, debiasing methods, evaluation protocols that measure age–gender entanglement, and governance to prevent automated systems from cementing discriminatory stereotypes.