A groundbreaking study published in Nature on October 8, 2025, exposes a pervasive and troubling bias embedded within online media and large language models: women are consistently portrayed as younger than men, regardless of actual age. This extensive research undertook an unprecedented scale of data analysis, scrutinizing over 1.4 million images and videos alongside nine advanced large language models trained on billions of words. The findings reveal a systematic distortion that skews public perception and reproduces deep-seated stereotypes across digital environments, with profound implications for gender equality in professional spheres.
Researchers from leading institutions—including UC Berkeley’s Haas School of Business, Stanford Graduate School of Business, and Oxford University’s Autonomy Institute—collaborated to dissect the intersection of gender, age, and digital representation. Despite demographic reality: that men and women exhibit no systematic age differences in the workforce and women generally live longer, the digital world constructs a divergent narrative. Online images and videos from platforms such as Google, Wikipedia, IMDb, Flickr, and YouTube repeatedly depict women not just as younger but unambiguously so across nearly 3,500 occupational and social categories.
This phenomenon extends beyond mere visual content. The study’s inclusion of text-based analysis across extensive internet corpora—spanning Reddit, Google News, Wikipedia, and Twitter—demonstrates a linguistic parallel. Language models and textual data associate youth disproportionately with women, while aging remains more attributable to men. This dual bias across modalities confirms the distortion is not an isolated artifact of cosmetics or photographic manipulation but a structural and cultural bias encoded into digital data ecosystems.
Perhaps most strikingly, the bias becomes magnified within high-status, high-paid professions. The study emphasizes that women in prestigious roles such as CEOs and astronauts appear even more disproportionately youthful compared to their male counterparts. This suggests a conflation between gender, age, and power that digitally reflects socio-economic disparities and occupational stratifications, potentially reinforcing systemic inequalities via skewed representation.
Experimental work within the research further illustrates the feedback loop catalyzed by algorithmic amplification. Human participants exposed to occupation-specific images exhibiting gendered age bias subsequently adjusted their perceptions, assuming younger ages for women-dominated roles and older ages for men-dominated ones. These altered perceptions influenced not only age estimates but hiring preferences, demonstrating how online portrayals feed directly into social cognition and decision-making processes about who fits which job and at what career stage.
The study also critically evaluates the behavior of generative AI, particularly ChatGPT’s handling of applicant résumés. When prompted to create and assess nearly 40,000 hypothetical résumés varying only by gendered names, the AI assumed women to be younger by an average of 1.6 years, possessing less work experience, and exhibiting more recent graduation dates than equivalently qualified men. This embedded age-gender bias led ChatGPT to rate older male candidates more favorably, exposing a prejudicial feedback cycle wherein AI systems, trained on biased data, perpetuate discriminatory assessments.
Technical methodologies underpinning these revelations are robust and multifaceted. The researchers employed human raters tasked with classifying gender and estimating age in categorized image samples, cross-referencing metadata with precise birthdates for objective validation. Machine learning models analyzed vast datasets to uncover consistent patterns, confirming statistical significance and revealing entrenched assumptions at scale. These rigorous approaches validate claims beyond anecdotal or isolated findings, situating the distortion as a widespread cultural artifact.
The cognitive implications extend further given society’s growing reliance on digital sources for social knowledge acquisition. As people increasingly learn about social roles and identities via mediated images and AI outputs, the perpetuation of illusory age disparities has the potential to entrench stereotypical gender roles. This dynamic may adversely affect women’s professional recognition, authority, and advancement, skewing public expectations and influencing recruitment and retention in subtle yet consequential ways.
These misrepresentations underscore a broader phenomenon that experts describe as a “culture-wide, statistical distortion of reality.” The distortion infiltrates all layers of digital communication—from imagery and video to spoken and written language—and becomes entrenched through algorithmic processing. As the digital ecosystem evolves, there is urgent need for interventions aimed at recalibrating AI models and content curation mechanisms to reflect empirical truths rather than socially constructed biases.
Researchers caution that without conscious correction, this feedback mechanism will intensify biases, making them more resistant to change. Given the rapid deployment of AI tools in sectors such as hiring and talent screening, unmitigated algorithmic bias could exacerbate existing disparities rather than alleviate them. Awareness and corrective transparency within AI training data and algorithmic decision frameworks must be prioritized to prevent deepening gender and age prejudices.
Solène Delecourt, one of the study’s co-authors and assistant professor at UC Berkeley Haas, emphasizes the societal imperative of recognizing and dismantling these biases. She articulates that the first step towards rectifying pervasive cultural inequalities lies in identifying how stereotypes not only shape individual cognition but are insidiously embedded in collective digital knowledge repositories and the algorithms that process them.
This pioneering research invites an urgent call for both academic and technological communities to scrutinize the ethical design of AI and online platforms. By exposing the entrenched systemic biases, it illuminates the scope of work necessary to ensure equitable representation in digital spaces, an essential prerequisite for fostering fairness in offline realities. Greater transparency, inclusive data sets, and bias-aware AI frameworks are vital for stemming the tide of misrepresentation that currently distorts public understanding.
As online images and language models shape contemporary narratives about gender and age, this study profoundly challenges the veracity of digital depictions. It lays bare how artificial intelligence and online media not only reflect but also reproduce and amplify cultural prejudices, emphasizing the critical intersection of technology, society, and identity formation in the 21st century. This research constitutes a pivotal step in unmasking the digital reflections that reshape and reinforce societal power dynamics.
Subject of Research: People
Article Title: Age and gender distortion in online media and large language models
News Publication Date: October 8, 2025
Web References:
References:
- Guilbeault, D., Delecourt, S., & Desikan, B.S. (2025). Age and gender distortion in online media and large language models. Nature.
Keywords:
- Gender bias
- Computational social science
- Social cognition
- Social psychology
- Mass media
- Communications
