Designing an AI-Enhanced Interface for Cognitive Support in High-Stakes Interrogations

Sammanfattning

Interrogation leaders in high-stakes investigative settings must convert long interview sessions into accurate written protocols under time pressure, stress, and cognitive bias. While AI has the potential to streamline this work, generic large-model assistants risk replacing one form of cognitive effort with another by inviting passive acceptance of generated output. To address this, we de- signed, developed, and evaluated FENRIR, a locally hosted AI product that supports post-interview documentation. We then assessed how well FENRIR fits the workflow of interrogation leaders at an authority that conducts such investigations. We followed a user-centered design process and developed FENRIR in cooperation with a domain expert and interrogation leaders from the studied organization. Participants agreed that transcript traceability of AI- based analysis was essential, and reported that AI-assisted analysis shifted their effort from generation to verification rather than removing effort alto- gether. Building on this, we argue for further work on so-called frictional design, interface patterns that make verification a required step of the work- flow rather than one the user can skip, so that the human remains the final decision-maker, particularly in high-stakes investigative settings.