La Nación’s AI Journey to Redefine Media Bias
Florencia Coelho, New Media Research and Training Manager at La Nación, participated in the 2020 JournalismAI Collab Challenges, where she and her Collab team worked to identify and mitigate bias in gender representation in the news. Their collaboration led to using tools and insights to help newsrooms recognise and address their gender bias and work towards more inclusive reporting.
Florencia Coelho, and a group of international journalists from eight news organisations, undertook a journey of innovation and collaboration during the 2020 JournalismAI Collab Challenges. They aimed to use AI to understand, identify, and mitigate bias in news.
The Collab Challenges, a precursor to the JournalismAI Fellowship, was a series of collaborative experiments that brought together media organisations from different countries to explore innovative solutions to improve journalism using AI technologies.
The AIJO Project, born out of the Collab Challenges, conducted two experiments using AI technology to uncover binary gender representations in various news publications. The results underscored the significant imbalance in gender representation in media, highlighting the ongoing need for more inclusive reporting practices.
The AIJO team included: AFP, Reuters, Nice Matin, Nikkei, Schibsted, Reach, Deutsche Welle, and Coelho’s La Nación. Coelho explained AIJO - an abbreviation for AI (Artificial Intelligence) and journalism, reflected their aim to merge these two fields to create more inclusive and insightful news reporting. Coincidentally, Aijō / 愛情 also means love in Japanese.
Image credits: Jamillah Knowles & Reset.Tech Australia / © https://au.reset.tech/ / Better Images of AI / Detail from Connected People / CC-BY 4.0
Exploring Bias with AIJO
Coelho said the idea for the project was a collective decision that emerged naturally. "I don't remember who suggested it, but it was a common-sense decision," she said.
The team’s first experiment focused on exploring gender bias through images. Using facial recognition technology, each news organisation provided a week of photographs to see how many men and women were featured on their publication's homepage.
"I provided 50 portraits of Latin individuals to the database, it was interesting to see the algorithm's results. It often made mistakes, particularly with women of darker complexion or older women, misidentifying them as male instead of female,” said Coelho.
“I tried to make a more inclusive database of photographs for the model,” she said. Explaining that AI relies on statistics, indicating there's always a possibility of mistakes due to how the data is organised.
In the second experiment, the team assessed gender representation using text. With the help of the Gender Gap Tracker, provided by Simon Fraser University, the team measured the proportion of men and women mentioned and quoted in articles. The team found a challenge with using an AI tool that works with multiple languages and decided to focus on one language - English.
“The most difficult part is understanding the nuances of languages, you can't copy-paste one development to other languages,” said Coelho. She explained that AI technology takes time and requires careful consideration of linguistic and cultural nuances.
"It's not just about translating a model to another language. You have to consider regional differences, the difficulty with Spanish-speaking countries is they use different words for the same thing,” she said.
Coelho said gaining internal support from her newsroom was important. “I got support from my newsroom to gather the 50 portraits and monitor our homepage for a week," she said. The support she received included samples for the photographs provided by the visual editor and assistance from the metrics and data team to analyse the ratios. Coelho said the project's success was due to collaboration and communication - their special strength.
“It was like a dream team,” she said. "We had a fantastic designer from France, Aurore Marval, who made inspiring graphics, and team members like Agnes Stenbom and Delfina Arambillet who were great at preparing the website with the team’s findings," said Coelho.
Reflecting on the role of the tool Coelho said, “This is only a tool that helps to measure [gender bias] but what you need is to change culture inside a newsroom”. Explaining that “it will be efficient or impactful if the newsroom decides to use it”. Coelho emphasised that it is only a tool and that real change requires a cultural shift in newsrooms, especially in male-dominated newsrooms.
Adoption at La Nación
Following the Collab Challenges, Coelho and her colleague Delfina Arambillet brought the gender gap tracker project to La Nación’s newsroom. Originally designed to measure the ratio of female-to-male sources quoted in Canadian media, they adapted the tool for Spanish-language media.
“For our newsroom, it's still a challenge as there is some minor progress but it's still a work in progress,” said Coelho. She emphasizes that while their newsroom still struggles with improving gender ratios, winning awards has spurred enthusiasm among editors to continue using these tools for tangible progress.
La Nación’s Spanish version of the Gender Gap Tracker (GENIE) won the WAN-IFRA Digital Latam best use of AI in journalism. GENIE identifies the gender of sources and quotes within its original content, classifying them as female, male, or non-binary.
Coelho acknowledged the challenges of replicating the gender tracker in other newsrooms. “They will have to rebuild it for their own,” she said. Explaining that the tool is not a one-size-fits-all solution and that newsrooms should customise it as each organisation has its unique linguistic and cultural context.
She said even just within their group of newspapers in Latin America, there are differences in language and regional nuances. “Our Spanish is different from other countries and in Brazil they speak Portuguese,” said Coelho.
She said news organisations need substantial resources, including funding, time, and expertise to rebuild the tool for specific newsrooms. She mentions the necessity of developing dictionaries of names and obtaining editorial and financial support to ensure the project's success.
Coelho said collaboration played an important role in AI-based tools. "When we started with our gender gap tracker, we lacked some of the necessary skills internally and partnered with a data scientist,” she said.
She emphasized the importance of a multidisciplinary approach to collaboration. “When we started our gender tracker, we needed a linguistics expert, developers, a project leader, and someone who understands metrics and dashboards,” said Coelho.
Despite the hurdles, the reaction within La Nación’s newsroom to their gender tracker project has been varied, ranging from excitement to frustration. Coelho said collaborative brainstorming sessions with journalists have led to innovative ideas, such as partnering with universities to engage female experts and individuals for news coverage.
Coelho hopes it will become easier to develop solutions with ready-to-use AI tools to lower the barriers for users or project teams.
Industry Impact and Future Directions
La Nación developed another tool, Cómo lo digo (“How do I say it?”) which was created together with the gender tracker. This tool focuses on the correct use of language in news headlines or sentences, particularly for sensitive topics such as disability, suicide, bullying, abuse, and diversity.
Cómo lo digo analyses language to guide journalists in using appropriate and inclusive wording. These tools were developed in response to the insights and challenges encountered during the Collab Challenges.
Reflecting on the Collab Challenges and its impact. “I have the best memories of that team,” said Coelho. "I enjoyed being exposed to perspectives from other countries beyond America. In the first Collab, I was excited to work with people from Japan and Europe, and it was amazing. It was an enriching experience to work with people from other countries,” she said.
“I am a very proud alumnus and thankful for what I learned and some ideas that helped me to develop in our team and my career,” Coelho concluded.