In collaboration with Aalto University, we explored how artificial intelligence can help knowledge brokers facilitate Science Sparring discussions. The Finnish Academy of Science and Letters has spent several years testing the potential of AI across different knowledge brokering methods.
AI’s potential is frequently explored, in practice and in the research literature, as a way of synthesising evidence: How can policymakers get better knowledge syntheses faster? The potential of AI in interactions between researchers and policymakers, however, has received less attention.
The three cornerstones of science advice are scientific credibility, political legitimacy, and practical usefulness for policymakers – requirements that do not disappear when AI is used as a supporting tool (Tyler et al. 2023). This was the starting point for the pilot: AI was introduced into the established Science Sparring model which emphasises trust between participants and relevance to an ongoing policymaking process. The project team included Professor Antti Oulasvirta, who leads Aalto University’s Computational Behavior Lab, and Master’s student Nina Aro, who carried out the work as her thesis.
“Public discussion often overlooks that AI use has to be designed to fit its specific context. Developing human-centred AI methods that suit different situations is far from easy. In this project we studied participatory design methods for shaping AI services. This was particularly interesting from a research perspective because Science Sparring is demanding in terms of content but also fast-paced,” says Professor Antti Oulasvirta.
The new tool: features and capabilities
The tool can be used in real Science Sparring workshops starting in autumn 2026. The development work was guided by a human-in-the-loop approach in which the knowledge broker decides which AI-generated information is used the discussion.
The prototype runs in the background throughout the Science Sparring session. It automatically analyses the discussions and offers knowledge brokers the following insights:
- Knowledge gaps: What are the open questions in this policy drafting process?
- Assumptions: What is being taken for granted and left unquestioned?
- Synthesis: Which themes stand out?
- Disagreements: What are participants disagreeing about?
AI brings additional analytical capacity to the three-hour intensive workshop. Drawing on the tool’s preliminary analysis the knowledge broker offers observations that challenge participants to consider the topic from new viewpoints.
“I was surprised by how much work it takes to understand users’ needs. We had to gain a thorough understanding of knowledge brokering before we could even begin to sketch out technical solutions. The development has required continuous experimentation, fine-tuning and close collaboration. The most exciting part was that AI was never before applied in this kind of context to support science-policy dialogue. That was liberating for our thinking.” says Nina Aro who was responsible for the prototype’s technical development and testing.
AI will not replace the knowledge broker
In the further development the goal is not a tool in which a human is one component (human-in-the-loop) but rather a system in which AI plays a role alongside many other components (AI-in-the-mix). AI is brought in only where it supports methodological development and the method is not shaped around AI’s requirements. Instead, AI is given the right place within the process so that trust and legitimacy are not compromised. As development continues, we are looking at how AI could challenge workshop participants’ thinking even more effectively to add value to its existing role of identifying knowledge gaps and assumptions for the knowledge broker.
Furthermore, we believe that in the future AI could move from simply analysing the discussion to systematically challenging the reasoning behind it. This could mean, for example, testing the theories of change underpinning a policy proposal: Which causal links are uncertain? Where might the implementation fail? AI could also look for shared assumptions that remain invisible precisely because everyone in the room takes them for granted. It could introduce counterarguments, alternative explanations, overlooked perspectives, or plausible future scenarios. These are the possibilities we want to explore next.
Text: Laura Väliniemi
Aro Nina (2026) Co-Designing AI Support for Science-for-Policy Facilitators. Aalto University.
Tyler et. al. (2023) AI tools as policy advicers? The potential and the pitfalls. Nature.
Read more about our Science and Policy work