Last week, Svenska Dagbladet depublished an essay by sociology professor Emma Engdahl after the newspaper's editors made the assessment that the text "may have been written to a significant extent with the help of AI".
The announcement came after the site Kvartal also analyzed the essay and confirmed the use of AI.
Emma Engdahl herself has denied the accusations, but SvD's head of culture Ida Ölmedal is confident with the editors' assessment.
"At SvD, we care that it is people who produce the journalism," she wrote in an email to DN.
But how do you reason on other newspapers? Where, and how, is the AI boundary drawn?
At Göteborgs-Posten, the starting point is that AI can be used to improve texts, but that the newspaper's texts must be written by humans.
But when it comes to texts by external writers, it is not always easy to determine where the line should be, says editor-in-chief Christofer Ahlqvist. The assessment is made on a case-by-case basis.
- Using AI to improve a sentence here and there does not have to disqualify a text from being published with us, continues Christofer Ahlqvist.
- You can compare it to debate articles by politicians. There, it is not always the politician himself who wrote, but often several ghostwriters. What's the difference between that and taking help from AI to put words in the right order, but keep your own opinions?
He believes that the higher the height of a text, the less reasonable it is to use AI.
- So it was probably reasonable for SvD to unpublish an essay. But it might not have been reasonable if it had been a different type of text.
Do you ever turn down texts due to suspicious AI use?
- Yes, but it is not about a huge number of cases. It has also happened that we liked the topic and asked the writer to rewrite with less AI, says Christofer Ahlqvist.
Even at Sydsvenskan, they have turned down texts because of AI. The newspaper's principle is that the texts should be written by people, says editor-in-chief Lena Philipson.
But like Christofer Ahlqvist, she believes that different types of texts should be assessed differently.
- We are very likely to take in submissions and debate articles that are written with the help of AI. And I'm not against it. There is a democratization aspect there - AI means that more people can write in an understandable way, says Lena Philipson.
At Dagens Nyheter, AI use is permitted in parts of the work process, but the newspaper does not publish AI-generated articles.
- In the communication to external writers it is stated that all formulations and conclusions must be the writer's own, says editor-in-chief Anna Åberg.
At Expressen, there is a quality policy that many freelancers also take advantage of. It states that AI can be used for, for example, language review and research, but that the writer must be responsible for the text, fact-checking and publication in its finished form.
- We can certainly work more with our communication around that, but our attitude is clear, says editor-in-chief Klas Granström.
The newspaper stops recurring texts where AI is considered to have done too much of the writing job.
- It is something that we need to become good at keeping track of.
In the review of Emma Engdahl's essay, Kvartal used the analysis tool Pangram to assess the presence of AI. SvD also used, according to a column, "checks with tools for recognizing AI-produced text" - alongside "an own editorial assessment".
Discussions about which methods and programs are best for the purpose are currently heated in the Swedish newsrooms.
At Expressen, analysis tools are used.
- But they must function as a complement and second opinion. We are not in a situation where we let the technology alone make systematic assessments, says Klas Granström and believes that the editorial assessment carries the most weight.
The same applies to, for example, Sydsvenskan and DN, where suitable programs and methods are discussed but currently no specific tool is used. The editors are allowed to use their human judgement.
Christofer Ahlqvist at GP sees a risk with the analysis tools:
- They can label an entire text as AI-generated even though the writer may have only used it to improve certain wording. But we are continuing to look at different methods, he says.
At the same time, he emphasizes that those of the newspaper's editorial offices that receive a lot of text from the outside, such as the culture editorial office, have a developed relationship with their writers and can thus conduct a dialogue with them about the limits of AI use.
- It is difficult to say that the way we work today will always apply. I think it will change a lot going forward. Basically it will be about the general perception and what our readers think is reasonable and not.
What do readers think about AI?
- When we tested AI, the reactions were few. But it probably looks different today than it did just six months ago. Lately, the risks of AI have come up on the agenda in a completely different way.
