Search the Internet for pictures of CEOs and you’ll quickly notice that almost all of the photos are of men. Of course, it’s not just CEOs. A new study shows that image searches exaggerate gender stereotypes and systematically underrepresent women across dozens of occupations.
The idea for the study, according to lead author Matthew Kay, came from watching research talks and wondering where the images people used in illustrations came from—and what message those images might be sending.
“You can imagine speakers Googling a given topic … and grabbing an image to spice up their slides. That’s one way that representation in image results on Google propagates outward beyond just the person who conducts the search, and was the initial spark that spawned the rest of our work,” Kay, a graduate student in computer science at the University of Washington, writes in an email. For example, if someone searches for images of CEOs, they may be more likely to use an image of a man in their own presentation, advertisement, or website, perpetuating the idea that business executives are usually men.
Search results, the researchers found, exaggerated real gender imbalance in many occupations, typically by about five to 10 percent.
The question remained: What results turn up when searching for images representing a profession? To find out, Kay and fellow researchers Cynthia Matusek and Sean Munson first looked at United States Bureau of Labor Statistics data on the proportion of men and women in each of 45 different occupations. Using the BLS occupation titles as search terms, they downloaded the top 100 search results from Google Images, and gave those to raters on Amazon Mechanical Turk to determine how many men and women appeared in the images.
Search results, the researchers found, exaggerate real gender imbalance in many occupations, typically by about five to 10 percent, though there were some larger discrepancies. For example, while women compose 26 percent of CEOs and 56 percent of authors, they made up just 11 percent and 25 percent, respectively, of image search results for “CEO” and “author.” Then there’s what the team dubbed the “sexy construction worker problem.” For some occupations—construction among them—women who showed up in search results were often depicted in sexualized and unrealistic ways.
Two experiments may have produced the most important findings. First, Amazon users felt search results were of a higher quality when, in fact, the team had manipulated those results to exaggerate gender bias. Second, by manipulating search results’ gender balance, the team found they could change perceptions of the actual gender balance in a field.
The most important takeaway, Kay writes, is that”implicitly or explicitly—the design of search algorithms has some impact on how people perceive the world.” Those algorithms, however, often have more to do with who links to what, who clicks on what, or whether an image can generate ad revenue. Algorithm designers, the researchers argue, should take gender bias into account. The alternative “risks reinforcing or even increasing perceptions of actual gender segregation in careers,” the team writes.
Kay will present the research at CHI 2015 in Seoul later this month.
Quick Studies is an award-winning series that sheds light on new research and discoveries that change the way we look at the world.
