Research Project > DATAPUBLICS - Transforming Journalism and Audiences in the age of datafication > News

AI in newsroom – experiences from across the world at the AI-Journalism festival

At the Journalism AI festival, Agnes Stenbom from Shipsted Media Stockholm and Issei Mori from Nikkei in Japan explained how they have collaborated with other newsrooms to use AI to combat biases in Journalism. Datapublics’ Jannie Møller Hartley reports from the conference.
Published by:
  • DATAPUBLICS - Transforming Journalism and Audiences in the age of datafication (Research Project)
    • Screen shot from digital of conference of speakers Agnes Stenbom from Shipsted Media Stockholm, Issei Mori from Nikkei in Japan and Mattia Peretti who was Managing the JournalismAI conference


    Bias in AI is a fact and well documented, but what can be done about it though AI? This was explored in the AIJO-collab project, presented at the POLIS conference on AI in Journalism in December 2020.

    “There’s a significant risk of bias when implementing AI in the newsroom. When coming together, we wanted to explore what AI could do for us, not to us,” says Agnes Stenbom, who is responsible Data & AI Specialist and Industry PhD Candidate, Schibsted.

    Gender bias a common concern

    One of the projects explored in the AI-Journalism Collab was how to better serve diverse audiences, which became the focal point of this talk at the JournalismAI Festival.

    “Bias in Japan, might not be the same in Sweden, but gender is an overall concern for many newsrooms across the world,” Agnes explained and said this led them to wanting to uncover gender representations in images used for news stories.

    Though face detection, filtering they showed that 77 percent of the images were men. This first step was more difficult than expected, especially as many people in the images are wearing facemasks. By adjusting the model with a human review, they showed that 72 percent of people in the images were male.

    The model is free and openly available from the AIJO-git hub, but Issei Mori stressed that the model should be reviewed by humans, as the model does sometimes make mistakes.

    They also examined if men and women were given the same length of quotes, and showed that men were given significantly more space than women.

    Future collaborative projects remain essential

    “On bringing this back to the newsroom, datasets of diversity are needed, systems that AI can learn from,” Agnes Stenbom said, remaining hopeful for the future, and more collaborative projects:

    “Machines in the industry will only be as good as the people in it,” Stenbom said.

    She argued that there is still a lot of work ahead, but that this project has been a good start:

    “AI is statistics on steroids, but the kind of AI that we have been using for this project, was a good starting point, and from there move on to more complicated technologies.”

    JournalismAI is a project run by POLIS – the journalism think-tank at the London School of Economics and Political Science – in collaboration with the Google News Initiative.