Profile
Scholarship
Selected recent studies on responsible AI for media and health, from news recommendation and user trust to AI-generated recipes.
Across three studies, the paper examines whether fear–hope reframing can reduce perceived polarization and negative emotional responses while preserving engagement-related outcomes.
A study of how fact-checked LLM support can shape editor trust in personalised news systems.
A cross-country study of how content provenance labels influence trust in digital news platforms.
Examining how people assess the nutritional quality of recipes produced by large language models.
Research into practice
The centre’s outcomes are collective achievements; my contribution combines research and scientific leadership with building the teams and partnerships behind the work.
A three-country study translated content credentials into evidence for newsroom and platform design.
Tools developed for conflict verification were used by fact-checkers and adopted by teams beyond Norway.
Responsible personalisation moved from models and prototypes into partner pipelines and live-platform evaluation.
Research in the public conversation
Recent publications, public writing and events.
As AI takes on a greater role in research, scientists need to be able to inspect, challenge and correct its contributions. An editorial on preserving scrutiny and accountability in AI-assisted science.
Research achievement, leadership competence and institutional power are different things. Universities need open competition and meaningful leadership assessment to give new voices a fair chance.
When AI writes the first draft, newsrooms still need a plan for developing the judgement required to check it. Editors should ask what each automated task used to teach new colleagues—and how they will gain that experience now.
VITAL is a prototype designed to help people assess online health claims, examine supporting evidence and follow transparent sources.
Read the announcementA four-wave study of news recommendation examines how article-selection behaviour relates to people’s stated topic profiles.
ACM DOIPDFDoing better with AI is not the same as retaining the skills needed to check its answers. An argument for keeping the human part of the thinking visible.
Journalists need more than an “AI expert” label. I argue for naming the work behind an expert’s claims and making the limits of that expertise explicit.
Today’s students will teach tomorrow’s professionals. Schools need to assess both what learners produce with AI and what they can understand and do on their own.
An opinion piece on AI-generated health advice and the importance of discussing chatbot use in clinical consultations, drawing on research into the reliability of AI-generated nutrition information.
A changed news feed can change clicks without changing interests. Responsible recommendation should help readers discover more while respecting their decision to say no.
Content credentials document where an image came from. That is not the same as proving a news source deserves our trust. The distinction matters for readers and newsrooms.
An argument for aligning researchers’ risk, control and rewards when universities seek to turn research into new companies.
The Arendalsuka 2026 programme included a MediaFutures session on responsible AI in Norwegian news media, with VITAL, verification and misinformation among the topics.
VITAL is a prototype designed to help people assess online health claims, examine supporting evidence and follow transparent sources.
Promising research can lose momentum between a project, a prototype and everyday use. Shared ownership and sustained collaboration need to survive those transitions.
Labelling synthetic material is only part of the task. Journalism also needs ways to make authentic content traceable when it travels beyond a publisher’s own platform.
AI-assisted work can look excellent without revealing what its author understands. Assessment needs to distinguish polished output from independent capability.
Our Research & Practice Note “Exploring Revealed Behavior and Stated Profiles in Longitudinal News Recommendation,” with Thomas Elmar Kolb and Alain Dominique Starke, reports a four-wave study with 262 participants. Preliminary results show that article-selection behavior shifts with feed composition while stated topic profiles remain comparatively stable—an important caution against treating short-term clicks or likes as stable preference change.
Universities prepare academics to teach. Leadership also needs preparation before people take responsibility for staff, budgets and institutional direction.
Our paper “Reducing Perceived Polarization through Affect-Balanced News Reframing,” with Jia-Hua Jeng, Alain D. Starke, David Elsweiler and Christoph Trattner, has been accepted for the ACM Conference on Recommender Systems 2026.
Students, researchers and lasting institutional knowledge are part of the infrastructure that turns research into value. Innovation policy should not separate them from the research it funds.
At ACM UMAP 2026 in Gothenburg, our team presented three papers spanning trustworthy news personalisation, selective news avoidance and AI-generated recipes. Tobias Wessel presented work on fact-checked LLMs for editor trust, Svenja Lys Forstner presented INRA work on explanations for low-interest news, and Yelyzaveta Lysova presented work on how users overlook nutritional flaws in LLM-generated recipes.
Our SFI MediaFutures paper on why explanations for low-interest news fail to persuade selective news avoiders is now available in the ACM UMAP 2026 workshop proceedings.
Two Research Council of Norway mobility projects connected to SFI MediaFutures and UiB were awarded, strengthening work on AI transparency, media literacy, vaccine beliefs and trustworthy health communication.
A quasi-experimental feasibility study shows how recommender-based support and structured documentation can contribute to operator performance in medical communication centres.
New work examines editor trust in fact-checked LLM-supported news personalisation and how users overlook nutritional flaws in LLM-generated recipes.
NoSoCSS is a new interdisciplinary initiative advancing computational social science research at the intersection of data, AI, platforms and society.