AI trained to detect gender stereotypes in films

Scientists have developed a machine-learning model to automatically recognise the actions of different characters in films. as reported in a study from the University of Southern California.

To determine the characters’ actions, the model analyses film scripts. To this end, the researchers assembled a dataset comprising 1.2 million scene descriptions from 912 films released between 1909 and 2013.

In total, the algorithm identified 50,000 actions performed by 20,000 characters.

They then conducted statistical analyses to identify differences in the types of actions performed by characters of different genders. In doing so, they identified well-established gender stereotypes.

“Having collected 1.2 million scene descriptions from 912 scripts, we were able to study systematic gender differences in the portrayal of characters in films at scale,” the study says.

According to the study, female characters more often display affection and have less freedom of movement than male characters. They are also more likely to be under observation by others in the scene, underscoring an emphasis on their appearance.

Male characters, in turn, are less likely to cry.

According to the researchers, the model’s ability to capture context and nuances within the overall narrative is limited. However, their results are in line with previous studies on gender stereotypes in popular media.

They believe their research may help raise awareness of how mass media perpetuate harmful stereotypes and thereby influence people’s beliefs and actions in real life.

They also note that the model identifies bias only in dialogues.

“These methods do not account for the harmful stereotypes conveyed through the actions of characters,” they wrote.

In the future, the researchers plan to improve the model and incorporate notions of intersectionality, such as gender, age, and race, for a deeper understanding of the issue.

In August 2021, the Twitter image-cropping algorithm was accused of bias against slim women .

In October 2022, researchers found that Facebook’s advertising tools use recognition of race, gender and age to deliver targeted sponsored messages.

Subscribe to ForkLog’s Telegram news: ForkLog AI — all the news from the world of AI!