Study Finds Gender Bias is More Prevalent in Online Images Than Online Text

A a new study, led by scholars at the  University of California, Berkeley, has discovered online images show more gender bias than online text, as well as elicit more gender bias from viewers.

To compare the prevalence of gender bias in online images versus online text, the research team selected 3,495 social categories, such as various occupations and social roles, from Wordnet, a large database developed by Princeton University. The researchers searched each category in Google Images, and pulled the top 100 images from the search results. They then asked a group of study participants to classify each human face in those images by gender. Next, they used a language-learning model to analyze Google News text and find the frequency of each social category’s occurrence with references to gender. After reviewing results from the Google Images analysis versus the Google News analysis, the researchers found gender associations were more prevalent in images than in text, with images focusing far more on men than women.

“Most of the previous research about bias on the internet has been focused on text, but we now have Google Images, TikTok, YouTube, Instagram — all kinds of content based on modalities besides text,” says Solène Delecourt, professor at the University of California, Berkeley. “Our research suggests that the extent of bias online is much more widespread than previously shown.”

The second part of experiment focused on how online images impact the gender biases of people who view them. The research team asked 450 participants to search via Google for occupations relating to science, technology, and the arts; one group of participants used Google Images and the other used Google News. Next, the participants were asked to rate which gender they associated with each occupation they searched. They were asked to take the same test three days later. The results for both the initial test and the test taken three days after showed the participants who searched for images had stronger gender associations then those who searched for text.

“This isn’t only about the frequency of gender bias online,” says Douglas Guilbeault, professor at the University of California, Berkeley. “Part of the story here is that there’s something very sticky, very potent about images’ representation of people that text just doesn’t have.”