2027: Agentic Computational Science

Ski track through snow-covered trees

The 27th Geilo Winter School, funded by DNV and SINTEF Digital, will take place Sunday January 17 to Friday January 22, 2027.

Agentic artificial intelligence is driving a profound transformation in science. By enabling systems that can reason, plan, and execute complex research tasks with increasing autonomy, agentic AI is redefining the scientific process itself. Applied mathematics has proven particularly well suited to this emerging paradigm, where structured reasoning and programming converge.

The 2027 Geilo Winter School explores the future of agentic computational science. Through hands-on sessions, case studies, and discussions of emerging trends, we will examine how AI agents are changing the way we work. The school will provide both practical tools and a strategic perspective on the opportunities ahead, in a field that evolves at unprecedented speed.

Cost and accommodation

Thanks to support from DNV and SINTEF, there is no fee to attend the school. Attendees cover their own travel and accommodation. The school takes place at Dr. Holms Hotel, which offers the following rates (in NOK):

PackageDouble roomSingle room
Course package, Sunday–Friday (per person per night, incl. full board)1,8452,300
Extra night, Friday–Sunday (per night, incl. breakfast)1,9901,790

Accompanying persons pay NOK 1,430 per person per night, including full board.

Register

More lecturers will be announced in the coming weeks.

Program

Please note that the program, including train departure times, is tentative and will be finalized in the coming days. Participants are responsible for arranging their own travel.

Topics

Trustworthy research with agentic AI (Geir Kjetil Sandve)

Lecture 1: A second brain for your research, built with agentic AI

Much of what makes a researcher effective never ends up in a paper: the articles you have read and what you thought of them, your own notes and drafts, the topics you know well enough to build on, your opinions about where your field is heading, and figures, explanations and plans. Agentic AI makes it practical to collect all of this into a personal knowledge base that grows over time. It also lets the AI draw on that knowledge base when it helps you write, plan or analyse. A short introduction (20–30 min) shows some ideas for how such a “second brain” can be organised and maintained, and where the AI should and should not take the lead. Participants then set up a first version for their own research and discuss design choices with each other and in plenary exchanges.

Lecture 2: Outer harnesses for transparent and reproducible research with agentic AI

An AI agent will happily run thirty analyses and report the best one. The question is how to make it do research you can actually trust. This lecture is about the “outer harness”: the skills, instruction documents, conventions and checks that can help steer an agent towards analyses that are transparent, reproducible and stable to reasonable alternative choices. A short introduction (20–30 min) is followed by participants building a small harness for an analysis of their own, then discussion in plenary.

Lecture 3: Setting the direction for machine-learning method development with agentic AI

Agentic AI can generate and test machine-learning methods far faster than a person can, which makes it easy to end up with many variants and no clear idea of which results to believe. This lecture is about setting the direction before the agent starts. Datasets, evaluation set-ups and ways of working are established first, so that you end up with a method whose results you trust and a clear overview of the methods developed, how they performed and how the development went. A short introduction (20–30 min) draws on concrete examples from immunology and time-series forecasting. Participants then set up such a framework for a method-development problem of their own and discuss it with each other.

A practical introduction to agentic AI for scientific computing (Jakob Torben)

TBA

Lecturers

Geir Kjetil Sandve

Geir Kjetil Sandve is Professor of Machine Learning at the Department of Informatics, University of Oslo, and co-director of TRUST, the Norwegian Centre for Trustworthy AI. His research spans methodology development and applied work, with a particular focus on deep learning, spatio-temporal modeling, simulation, benchmarking and reproducibility, with a growing focus on veridical data science and on how agentic AI changes the practice of research. On the application side he works on immune-receptor repertoire analysis, through the open-source immuneML platform, and on climate-sensitive disease forecasting in low- and middle-income countries, through the Chap modelling platform.

Geir Kjetil Sandve

Jakob Torben

Jakob Torben is a PhD fellow in the Applied Computational Science group at SINTEF Digital, Department of Mathematics and Cybernetics. He works on simulation and modelling of physical systems using numerical and data-driven methods. His current focus is agentic AI for scientific computing: AI systems grounded in physics-based simulators that make simulation-driven modelling faster and more accessible. This builds on his expertise in hybrid methods combining machine learning with differentiable simulators, and in high-performance computing, especially GPU computing. He holds an MSc in Applied Computational Science and Engineering (2021) from the Department of Earth Science and Engineering at Imperial College London.

Jakob Torben

Funding

The 2027 Geilo Winter School is funded by DNV and SINTEF Digital. The funding from DNV is provided by DNV Group Research and Development.