The Male Abstraction Bias in Internet Culture and Large Language Models

摘要变化

TwoConceptual foundationalabstraction and gender are two equally central aspects of the human cognitionmind, abstractnessyet andthey gender,have aretraditionally typicallybeen examinedstudied in isolation from one another, as separablethough they were fundamentally separate. Here, we find theygender and abstraction are systematically linked throughout the entire space of concepts spanning single (N = 40,000) and multi-word expressions (N = 67,000): across billions of words and millions of images on the internet --- and across numerous measurement techniques, including human and AI classifications --- more abstract concepts are more associated with men, ana effectpattern whichwe show replicates in languages around the world. This patternmale abstraction bias not only holds across conceptualsemantic domains, but it also maps onto grammatical structure: verbs and adverbs are more abstract and more male-coded, revealing a system-wide trace of stereotypes accentuating agentic qualities in men. Furthermore, we show this bias readily emerges in humanthe free associations humans and AIAIs responsesproduce in freecontrolled associationonline tasks, indicating its psychological internalization and its algorithmic reproduction. TheThese malefindings abstractionidentify biasa maylatent underliestructure andin unifyconceptual space that unifies a range of stereotypes: for example, evidencingframings of women as more embodied, emotional, and warm, versus men as more cognitive, rational, and competent; as well as erroneous views of men as better at mathematics, science, and engineering. More broadly, by analyzing an unprecedentedly large, multimodal sample of online data, we provide culture-wide evidence of gender as a core organizing dimensionprinciple inshaping conceptual space far beyond the characterization of people and social roles.