Using AI to automatically track bias and stereotypes in journalistic content
AFP has been looking at this issue for six months, in partnership with seven other media groups (Reuters, Schibsted, La Nacion, Reach PLC, Deutsche Welle, Nikkei, Nice Matin). The working group, brought together by Polis -- a media think tank at the London School of Economics -- presented its findings on December 8 at the JournalismAI Festival.
Two pilot projects were carried out over a week in November 2020 to test tools that detect gender bias in text content (in English) and photos collected by the participating groups.
The results show that there is still work to be done: the ratio of women appearing in the photos was 27.2 percent. We analysed 1,430 articles from 16-20 November showing that women made up 21 percent of people cited and 22 percent of sources quoted. The algorithm also enabled us to calculate the length of quotes, which were slightly shorter for women. The tool, which can run automatically, has other potential applications: for example to ensure balanced coverage of candidates in an election campaign.
To find out more, please visit the group's website at: https://www.aijoproject.com/
The project was presented on December 8 a 1600 GMT here: https://www.journalismaifestival.com/