Images on the Internet are even more sexist than texts | Technology

Photos that appear after a Google Images search of professions such as cardiovascular surgeon, mathematician, developer, footballer, commissioner or investment banker are of men. And for a model, hairdresser, nurse, interior designer or art teacher, they are women. A study published this Wednesday by the magazine Nature which analyzed 349,500 photos, 100 for each of the 3,495 most common social categories, observes consistent gender bias on Google and other platforms such as Wikipedia and IMDB. So far, the most relevant studies that have analyzed these biases online They focused primarily on the texts.

Researchers from the University of California at Berkeley say their discovery is important because of the unstoppable increase in visual culture: reading time is reduced and entertainment on “hypervisual platforms focused on the exchange of images » is growing, the article states. Bias in images online is higher than what is observed in surveys and in actual census data in the United States, although, according to Douglas Guilbeault, a professor at Berkeley and one of the authors, “a subtlety is that social groups and generations may differ in the type of visual content they produce and consume, and an important area for future research is to explore how this affects their experience.

Researchers also found that biases in images online It is more frequent than in texts and its effects are more psychologically powerful. At work, they compared the images with text results from Google News. Images demonstrate greater prejudices than texts and their impact is more lasting: people who saw an image with a sexist bias retained its influence for more days than those who read a text with a certain sexism. “What surprises me most about this work is that a simple change, from textual to visual information, can have many implications in the spread of gender bias, threatening decades of progress,” says Guilbeault.

It should be noted that the language used for the research is English, a language in which most professions are not distinguished by gender: there is no surgeon, only surgeon. Photos therefore provide more information than text when we see a footballer or a hairdresser. In Spanish, however, the linguistic distinction would also be evident. But researchers have discovered effects that go beyond the words themselves: “Various psychological research suggests that images may be a particularly powerful way to transmit gender bias. Research on the image superiority effect “They show that images tend to be more indelible and more emotionally impactful than texts,” the article states.

Another advantage would be the use of neutral terminology, which is more complicated in Spanish than in English. Thus, images on the Internet amplify gender bias, both in their statistical figures and in their psychological impact on users.

How do biases appear?

Photos on the Internet reproduce millions of choices of its users, who created or uploaded these photos to their accounts and pages: “Gender bias appears to be driven in part by the content that Internet users choose to post on their blogs. such as audience preferences for what information to consume or images to purchase,” the research explains.

But how have the mechanisms specific to the Internet influenced this bias in online images? “Bias can be intensified because of the network itself,” says Professor Bas Hofstra of Radboud University in the Netherlands, after analyzing the paper before publication. “Maybe the user populations differ, maybe men are the ones who have used or are using the internet the most and as such it’s becoming a more gender sensitive place,” she says.

The article also speculates that the status or bias in hiring at media companies, whose photos are overrepresented on the Internet, may be another cause. “The human preference for a familiar, prototypical representation of social categories is likely to play a role in perpetuating these biases,” the researchers say.

The consequence of this dynamic is to limit women’s access to jobs in which they no longer have a place due to previous social dynamics. But there could be more, according to Hofstra: “The way people talk about certain social categories, for example; we talk more about men in relation to sport, because we see more male athletes onlinewhich could reduce the aspirations of women to take up this sport. »

An obvious fear for the future which is only mentioned in the article according to which the images of the current Internet are one of the main bases of generative artificial intelligence (AI): “Images created with the AI can make the internet a more sexist place if they are based on already generated images that exist online”Hofstra said. “Our work suggests that gender bias in AI may be due in part to the fact that it is trained with public images from platforms such as Google and Wikipedia, which are rife with gender bias,” the study says.

It is difficult to think of concrete solutions to this problem. Google can give biased results because the queries or information already existing on the Internet are biased. “Solutions will have to emerge from technology, academia and civil society,” believes Guilbeault. “Our research aims to start a critical conversation about the implications of visual change for the spread of gender bias. The methods and tools we have created are an important step toward providing transparent information about the changing landscape of bias. onlinewhich is a necessary step to find solutions.

You can follow EL PAÍS Technology In Facebook And X or sign up here to receive our weekly newsletter.

_