These days, we’re bombarded with images on picture-packed news sites and social media platforms. And much of that visual content, according to new research, is reinforcing powerful gender stereotypes.
Through a series of experiments and with the help of large-language models, Assistant Professors Douglas Guilbeault and Solène Delecourt found that Google Images exhibit significantly stronger gender bias for both female- and male-typed categories than text from Google News. What’s more, while the text is slightly more focused on men than women, this bias is over four times stronger in images.
Delecourt says that most of the previous research about bias online has focused on text. “But we now have Google Images, TikTok, YouTube, Instagram—all kinds of content based on modalities besides text,” she says. “Our research suggests that the extent of bias online is much more widespread than previously shown.”
To zero in on gender bias in online images, Guilbeault, Delecourt, and colleagues designed a novel series of techniques to compare bias in images versus text and to investigate its psychological impact in both mediums.
First, the researchers pulled 3,495 social categories—which included occupations like “doctor” and “carpenter” as well as social roles like “friend” and “neighbor”—from Wordnet, a large database of related words and concepts. To calculate the gender balance within each category of images, the researchers retrieved the top hundred Google images corresponding to each category and recruited people to classify each human face by perceived gender.
Large-language models measured gender bias in online text by noting the frequency of each social category’s occurrence alongside references to gender in Google News text. The researchers’ analysis revealed that gender associations were more extreme among the images than within text. There were also far more images focused on men than women.
In another study, 450 participants searched Google for apt descriptions of occupations relating to science, technology, and the arts. One group used Google News to upload textual descriptions; another group used Google Images to upload pictures of occupations. Compared to those in the text and control conditions, the participants who worked with the images displayed much stronger gender bias associating women with arts and men with science (a bias linked to systemic inequalities in academia and industry)—even three days later.
“This isn’t only about the frequency of gender bias online,” says Guilbeault, the paper’s lead author. “There’s something very sticky, very potent about images’ representation of people that text just doesn’t have.”
Catalyzing American entrepreneurship to look like America
Even though professional faculty member Maura O’Neill, BCEMBA 04, started her first company 40 years ago, the hurdles she faced as a female entrepreneur are largely the same ones currently faced by her students who are women and/or students of color. “These Berkeley students are super smart, have great ideas with grit oozing out of their DNA,” says O’Neill. What are we missing as a community and as a nation, she wondered, by leaving this talent on the sidelines, particularly when both the problems and the opportunities are huge?
O’Neill analyzed big data sets and found that if American business ownership reflected the race, ethnicity, and gender of the nation’s population, the country could have 2.4 million more businesses, 20 million more jobs, and a $4 trillion increase in annual GDP. At the state level, those whose business ownership more closely reflects their population have 2.5 times higher GDP than states lagging further behind. “Everyone wins when this happens,” O’Neill says.
At current rates, it will take over 700 years for American entrepreneurship to look like America. But O’Neill launched The Decade Project to accelerate the path to parity in 10 years. “If we can put a man on the moon in a decade and bring him back safely without first knowing the technology,” she says, “we can do this.”
TDP’s research shows that four areas are key to success: role models, financial capital, knowledge, and connections. In addition to mobilizing a national ecosystem of partners in each of the areas, TDP is identifying ways to leverage the flood of federal and private investment dollars (estimated to be $8 trillion) associated with the transition to a sustainable economy to seize this opportunity and catalyze local leaders across the nation.
What’s in it for Your State?
The Decade Project data indicates that if American business ownership reflected the local population, the country could have over 2 million more employer firms. How many would your state gain?
How Long to Reach Equality?
At current rates, if nothing changes, it will take over 700 years for American business ownership to look like America.*
*Trends based on 2019 Census Data.
**As a whole, Asian Americans are at parity, but it is not true for every Asian subgroup.
Assistant Professor Kiera Hudson studies schadenfreude and the psychological and biological roots of power hierarchies.
Assistant Professor Sa-kiera “Kiera” Hudson has received a 2024 National Science Foundation CAREER award, the NSF’s most prestigious awards program in support of early-career faculty who have the potential to serve as academic role models in research and education.
Hudson said she is thrilled to receive the award and will use the $850,000 grant to fund new research on schadenfreude, which is pleasure derived from another person’s misfortune.
Empathy is often hailed as the emotion to target in intergroup conflicts, as it predicts consequential behaviors that can help reduce inequality, said Hudson, who earned a PhD in the (social) psychology department at Harvard University in 2020. “In many social conflicts, people struggle to feel empathy for those not part of their social groups,” she said. But in the study of empathy, behavioral scientists have perhaps overlooked schadenfreude’s relevance to conflict among groups of people, which is why it’s crucial to learn more, she said.
“If we better understand what drives intergroup schadenfreude—and the consequences—we can better understand how to design interventions to decrease the harm it causes, particularly to marginalized groups,” she said.
How schadenfreude harms
In her new research project, Hudson, a member of the Management of Organizations Group (MORS) at Haas, will investigate how schadenfreude contributes to harm, attempting to understand the cognitive mechanisms that allow it to flourish. The project will put a strong emphasis on research and education, including training minoritized scientists, collaborating with organizations focused on equity and social justice, and disseminating research to interdisciplinary communities.
Hudson said her goal is to bring a broader understanding of people’s more “nasty, harmful behaviors,” at a particular time in history.
“Across the world, there has been an increase in rigid beliefs of who belongs to ‘us’ versus ‘them’ fueled by perceived threat and competition, leading to intensified intergroup animosity,” she said. “These are the exact conditions under which schadenfreude thrives, suggesting that we are not only in an empathy deficit as a nation, as proposed by Obama in 2006, but perhaps also in a schadenfreude surplus.”
More broadly, Hudson’s research at Haas is focused on two main areas: the psychological and biological roots of power hierarchies, and how these hierarchies intersect to influence experiences and perceptions.
A paper published today in the journal Nature finds that online images show stronger gender biases than online texts. Researchers also found that bias is more psychologically potent in visual form than in writing.
Image copyright Solène Delecourt
A picture is worth a thousand words, as the saying goes, and research has shown that the human brain does indeed better retain information from images than from text.
These days, we are taking in more visual content than ever as we peruse picture-packed news sites and social media platforms. And much of that visual content, according to new Berkeley Haas research, is reinforcing powerful gender stereotypes.
Through a series of experiments, observations, and the help of large language models, professors Douglas Guilbeault and Solène Delecourt found that female and male gender associations are more extreme among images retrieved on Google than within text from Google News. What’s more, while the text is slightly more focused on men than women, this bias is over four times stronger in images.
“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 Delecourt. “Our research suggests that the extent of bias online is much more widespread than previously shown.”
Not only is online gender bias more prevalent in images than in text, the study revealed, but such bias is more psychologically potent in visual form. Strikingly, in one experiment, study participants who looked at gender-biased images—as opposed to those reading gender-biased text—demonstrated significantly stronger biases even three days later.
As online worlds grow more and more visual, it’s important to understand the outsized potency of images, says Guilbeault, the lead author on the paper.
“We realized that this has implications for stereotypes—and no one had demonstrated that connection before,” Guilbeault says. “Images are a particularly sticky way for stereotypes to be communicated.”
The extent of bias–and its effects
To zero in on gender bias in online images, Guilbeault and Delecourt teamed up with co-authors Tasker Hull from Psiphon, Inc., a software company that develops censorship-navigation tools; doctoral researcher Bhargav Srinivasa Desikan of Switzerland’s École Polytechnique Fédérale de Lausanne (now at IPPR in London); Mark Chu from Columbia University; and Ethan Nadler from the University of Southern California. They designed a novel series of techniques to compare bias in images versus text, and to investigate its psychological impact in both mediums.
First, the researchers pulled 3,495 social categories—which included occupations like “doctor” and “carpenter” as well as social roles like “friend” and “neighbor”—from Wordnet, a large database of related words and concepts.
To calculate the gender balance within each category of images, the researchers retrieved the top hundred Google images corresponding to each category and recruited people to classify each human face by gender.
Measuring gender bias in online texts was a trickier proposition—though one perfectly suited for fast-evolving large-language models, which noted the frequency of each social category’s occurrence alongside references to gender in Google News text.
The researchers’ analysis revealed that gender associations were more extreme among the images than within the text. There were also far more images focused on men than women.
Sticky images
The experimental phase of the study sought to illuminate the impacts that biases in online images have on internet users. The researchers asked 450 participants to use Google to search for apt descriptions of occupations relating to science, technology, and the arts. One group used Google News to find and upload textual descriptions; another group used Google Images to find and upload pictures of occupations. (A control group was assigned the same task with neutral categories like “apple” and “guitar.”)
After selecting their text- or image-based descriptions, the participants rated which gender they most associated with each occupation. Then they completed a test that asked them to quickly sort various words into gender categories. The test was administered again after three days.
The participants who worked with the images displayed much stronger gender associations compared to those in the text and control conditions—even three days later.
“This isn’t only about the frequency of gender bias online,” says Guilbeault. “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.”
Interestingly, when the researchers conducted their own online survey of public opinion—and when they looked at data on occupational gender distributions reported by the U.S. Bureau of Labor Statistics—they found that gender disparities were much less pronounced than in those reflected in Google images.
Opening doors to new research
Delecourt and Guilbeault say they hope their findings lead to a more serious grappling with the challenges posed by embedded bias in online images. After all, it’s relatively easy to tweak text to be as neutral as possible, whereas images of people inherently convey racial, gender, and other demographic information.
Guilbeault notes that other research has shown that gender biases in online text have decreased, but those findings may not reveal the whole story. “In images we actually still see very prevalent widespread gender bias,” he says. “That may be because we haven’t really focused on images in terms of this movement towards gender equality. But it could also be because it’s just harder to do that in images.”
Guilbeault and Delecourt are already at work on another project in this vein to examine gender-age bias online using many of the same techniques. “Part of the reason this paper is so exciting is that it opens the door to many, many other types of research—into age or race, or into other modalities, like video,” Delecourt says.
These days, we’re bombarded with images on picture-packed news sites and social media platforms. And much of that visual content, according to new research, is reinforcing powerful gender stereotypes.
Even though professional faculty member Maura O’Neill, BCEMBA 04, started her first company 40 years ago, the hurdles she faced as a female entrepreneur are largely the same ones currently faced by her students who are women and/or students of color. “These Berkeley students are super smart, have great ideas with grit oozing out of their DNA,” says O’Neill. What are we missing as a community and as a nation, she wondered, by leaving this talent on the sidelines, particularly when both the problems and the opportunities are huge? 
