Which text was written by an AI? Journalism researcher Antonia Eichenauer appeared as a guest on radio3 (rbb), where she spoke with host Katja Weber about an experiment conducted by the media outlet Correctiv. In the “Spotlight” newsletter, readers were given one week to guess which text was written by an AI and which was written by editor-in-chief Annette Dowideit. Readers were also asked to rate their preferred version. The verdict? Readers were not always able to reliably distinguish between the texts. In some cases, even colleagues from the editorial team were incorrect in their assessments.
There was no clear answer when it came to whether AI-generated texts are generally perceived as better or worse. This did not surprise Antonia Eichenauer. As part of the project “Journalism: The Automation of News and Journalistic Autonomy,” she studies AI in journalism and is well aware of the persuasive power of AI-generated texts.
So, what does this mean for journalism? Her assessment: If the use of AI isn’t transparent, it can damage the credibility of journalists and media brands. If the audience gets the impression that something is being “slipped in” on them, trust suffers. This is precisely why, according to Antonia Eichenauer, it will be even more important for journalists in the future to demonstrate that they have conducted their own research and actually observed or experienced events firsthand, thus taking on the role of witnesses.
For the experiment, the artificial intelligence “Lumo,” developed by Proton, generated the texts using only Dowideit’s previous writings.
Photo: Antonia Eichenauer