researcher Sahiti Sarva, with partner Leonardo Nicoletti, studied the headlines of major publications in the United States, United Kingdom, India, and South Africa for 10 years to find out how women are represented. In early 2022, he made his clear conclusion in a workshop organized by the London School of Economics (LSE, United Kingdom).

In his opinion, without a timely combination of human talent and artificial intelligence, the results would be useless. “Before starting to implement any algorithm, there was a lot of work to be done,” he recalls. “You have to clean up and remove superfluous words”, he clarifies before adding: “Then you have to find the right code pack for your job”.

He and colleagues developed their own dictionary, capable of calculating what they call “gender bias”. To go beyond the obvious and predictable —“woman”, “girl”, “girl”—, they think that this task should include “combinations of the feminine lexicon, such as ‘wife’, ‘daughter’, ‘actress’, etc. . ., as well as social and behavioral stereotypes, such as ’emotional support’ or ‘care’”.

After this phase, they use a machine learning method called “sentiment analysis”. In this way, they found that, when women do appear in the headlines, “the stories are twice as likely to be violent than empowering.” The assumptions of thousands of journalists and users are verified by data.

Manager from journalismAI, an LSE initiator, Sabrina Argoub, argues that artificial intelligence “can be of great help to journalists.” In the case of women in media, Argoub believes this technology helps increase transparency and accountability, as well as “raise awareness”.

Project AIJO, which automatically compares the rates of men and women cited in journalistic works, being a reference in this field. However, as this professional underlines, artificial intelligence is not the solution to all problems. “The machine provides information, valuable material… That’s the starting point,” he said. The really important decisions are made by the people in the newsroom.