Data-driven journalism for the future
Data-driven journalism is here to stay – and AI tools are only getting better and better. Måns Magnusson, researcher in machine learning and statistics, monitors developments in this area with interest.
Magnusson is a researcher at the Department of Statistics at Uppsala University. In 2016, he co-founded the data-driven news agency Newsworthy. Its business idea is to process and analyse large amounts of data and break it down municipality by municipality or county by county. News items and ideas are then sold as a service to local newspapers.
“Last year, Newsworthy broke down the climate budget, showing how much carbon dioxide was released at the municipal level and what that meant for different areas of Sweden. It had quite an impact in the local newspapers.”
AI image tools
and statistics. Photo: Tobias Sterner
Recently, AI tools for producing text and images have attracted great interest and provoked much debate. In a collaboration between Newsworthy and the Department of Statistics, two students have written a Bachelor’s dissertation on the possibility of using AI-generated images to illustrate news articles.
“The students concluded that some images were good enough, but they were in the minority. Although impressive images, they were still not good enough for journalistic use. For example, people in the pictures sometimes only had four fingers,” Magnusson says with a laugh.
Can make silly mistakes
In the future, he hopes to investigate further whether auto-generated text can be used in journalism, but guesses there is still some way to go before such AI tools become truly useful.
“The generative models can make rather silly mistakes today. If you want to pursue automated journalism, you must have quality guarantees. It doesn't work when one out of a hundred articles is completely wrong. That aspect is extremely vital within journalism.”
Do you think journalism will become increasingly automated in the long run?
“Yes, but the idea of merely generating text is less intriguing. Journalism has a different role: to pinpoint what is newsworthy and interesting, to offer in-depth analysis and assessments. In that regard, I think the idea of automatically generating articles is still a long way off.”
Two different research tracks
At the Newsworthy agency, the use of AI tools is closer at hand. However, he would not say that they use AI today; instead he talks about data-driven journalism, which he believes has a real future.
“Each individual local newspaper’s editorial office cannot afford to do an in-depth statistical analysis of local data, but if the cost is shared among many local newspapers, that can be achieved.”
Magnusson is pursuing two different tracks in his research. One is examining statistical inference or probability, where probability models of reality are produced. The second is machine learning for textual data, whereby large amounts of text are statistically analysed.
Both of these fields of research feed into Newsworthy, even though the company was founded separately from the research. And developments are moving rapidly, he notes.
“This is a very fun area to research, as you can see how it affects society on a broad scale.”
Annica Hulth