Department of Computer Science and Electrical Engineering

A study by CSEE professor Cynthia Matuszek on gender bias in online images associated with a variety of occupations has recently received a lot of attention. The study, which was carried out with former University of Washington colleagues Matthew Kay and Sean Munson, resulted in a paper to be presented at the ACM CHI Conference on Human Factors in Computing Systems in Seoul, Korea in mid-April.

The paper, Unequal Representation and Gender Stereotypes in Image Search Results for Occupations, was recognized as a best paper of the 2015 ACM CHI Conference. This conference is considered the most prestigious in the field of human–computer interaction, and is one of the top ranked computer science conferences. 

"This project was a joint effort between myself and the two other authors," Dr. Matuszek explains. "The project was originally developed after I saw a presentation in which the images chosen to represent different professions were extremely one-sided and didn't seem representative.  Our primary goal was to learn more about the software design space: does gender in image search results affect whether people think the search results are good?  On the flip side, does showing different genders in different roles (that is, showing a male vs. a female nurse) affect how people think of those occupations?  You need to know the answers to questions like that before you can create, e.g., an image search engine that is really well and thoughtfully designed."

Not surprisingly, Dr. Matuszek's work has been noted by the popular press, with recent stories in the Washington Post Wonkblog, The Atlantic and many online news outlets. The Washington Post article quotes Dr. Matuszek on the genesis of the study.

In addition, many images retrieved by the web’s top search engine happen to be hyper-sexualized caricatures. Some female construction workers in midriff-baring flannel and jean shorts seem better dressed for a Halloween party than, say, a demolition site. (Researchers dubbed this the “sexy construction worker problem.”)

"It’s part of a cycle: How people perceive things affects the search results, which affect how people perceive things," said co-author Cynthia Matuszek, who now teaches computer ethics at the University of Maryland, Baltimore County. Matuszek recalls sitting in a robotics lecture last year at the University of Washington, where she earned her doctoral degree in computer science. A male colleague illustrated researchers in his Powerpoint presentation as “all guys, classic nerds,” she said. But a caretaker was shown in a slide as "a plump woman in her thirties who was wearing a pink suit." The stereotypes irked Matuszek, and she's not the only one wondering about the power of images.

About eight months ago, Matuszek and her colleagues at the University of Washington decided to test the power of popular image. They wanted to know if something as seemingly trivial as search results could sway someone’s perception of how many women work in a certain field, and whether they’re competent. The researchers surveyed 21 people — a pool too small to make any sweeping statement, Matuszek acknowledges, but big enough for a glimpse into our cultural psyche — starting with questions like: What percentage of construction workers are women? Do you believe the person in this photo is good at their job? Two weeks later, they followed up, prompting participants to sift through Google image results before answering the same inquiries. Responses changed after Google images were introduced, according to the study, which was published this week. Search results could determine 7 percent of a participant's subsequent opinion about the number of men and women in a particular field, the authors calculated. And a worker was, on average, deemed more competent if he or she fit into a gender stereotype.

As Adrienne LaFrance notes in a recent Atlantic article about Matuszek's study, "Google image searches don't just reflect the sad state of diversity in corporate leadership; they actually influence the ways in which people think about what it means to be a CEO."

The study concluded that "shifting the representation of gender in image search results can shift people’s perceptions about real-world distributions."[i]