How do we move AI research from promising methods to trustworthy solutions in practice?

That question was at the center of the NorwAI Forum Spring 2026, held on 15 April in Trondheim. Researchers, PhD candidates, postdocs, and user partners gathered for a full day dedicated to current research across the center.

While the autumn edition of the NorwAI Forum focuses on innovation and partner activities, the spring forum highlights ongoing research: new ideas, emerging PhD projects, recent results, and hands-on exploration of Norwegian large language models.

The program opened with a welcome from Center Director Jon Atle Gulla, who also introduced NorwAI’s new partners HEMIT and Medbric. Short updates from each work package gave participants an overview of recent progress across the center, followed by presentations from researchers and partners.

Trustworthy AI and Responsible Deployment

A strong theme throughout the day was trustworthy AI: Presentations covered explainable anomaly detection through uncertainty decomposition and counterfactual explanations, privacy-preserving AI, fairness in recommender systems, and how trust is shaped in human-AI interaction. The discussions reflected an important shift in AI research: not only asking whether systems perform well, but whether they are understandable, robust, fair and worthy of trust.

AI for Industry and Real-World Data

Several talks focused on industrial AI applications and scalable analytics. Researchers presented work on anomaly detection in mobile networks, streaming graph clustering, contrastive learning for time series, and the use of knowledge graphs and semantics to scale intelligent systems.
Industry perspectives added practical insight into what happens after the prototype stage when AI systems must operate reliably in complex real-world environments.

Generative AI in Organizations

Another key theme was how generative AI is changing organizations and knowledge work: Talks explored how large language models reshape knowledge management, how organizations translate responsible AI principles into practice, and how professionals collaborate creatively with AI through prompting, clarification and iterative dialogue. These contributions highlighted that successful generative AI adoption depends as much on people, workflows and competence as on the models themselves.

AI for Healthcare

The forum also showcased ongoing collaboration on health language models together with healthcare partners, focusing on safe and trustworthy LLMs for clinical settings. The discussions addressed important challenges such as access to sensitive data, evaluation with healthcare professionals, sovereignty, cost, and sustainable innovation models.

Norwegian Language Models in Practice

The day concluded with a hands-on tutorial on NorwAI LLMs, covering the full journey from pretraining to post-training, data quality, fine-tuning and tool use. The session underlined the importance of national language resources, strong datasets, and practical competence in building AI systems for Norwegian contexts.

From Methods to Impact

Across the presentations, one message stood out clearly: NorwAI research is not only about building better models. It is about building AI that works in the real world: AI that is trustworthy, useful, adaptable, and responsible.
 

By Kerstin Bach

Published 2026-04-30