AIDA – AI Debriefing on Action
AIDA explores how AI can analyse recorded professional action and guide learners through reflective debriefing in vocational and medical education.
Duration: 1 February 2027 – 29 February 2028
Status: Planned
Education level: Tertiary Level, Upper Secondary Level
Topic: AI – Artificial Intelligence
Keywords: AI-supported debriefing; reflective learning; professional practice; vocational education; medical education
Background
Professional expertise in nursing and emergency medicine develops through repeated cycles of action, feedback and reflection. Debriefing is the key mechanism through which learners turn concrete practice situations into transferable knowledge and action competence. High-quality debriefing, however, is resource-intensive: it requires trained facilitators, structured observation, protected time and, in complex scenarios, several observers. In vocational and clinical education, expert feedback is therefore rarely available at the frequency and depth learners need. Video-assisted debriefing makes professional action observable and discussable. Recent advances in multimodal AI and conversational agents open new possibilities: identifying relevant moments in recordings, structuring observations along predefined quality criteria and supporting individual reflective dialogue. Yet AI-supported debriefing is not automatically educationally valid. Systems may misinterpret situated action, overstate certainty or create data protection risks. What is missing are well-specified, responsibly implemented use cases that combine recorded professional action (including 360-degree video), domain-specific quality criteria, AI-supported analysis and avatar-based reflection, and that are tested with learners and instructors in authentic educational settings.
Objectives
AIDA develops and explores AI-supported debriefing of learners’ own recorded professional action. Objectives: (1) specify two use cases, one in nursing-related vocational education and one in emergency medicine training, with operationalised quality criteria; (2) configure two prototypes combining video analysis and avatar-based reflective dialogue; (3) assess technical accuracy, usability, pedagogical plausibility and acceptability; (4) derive an implementation concept and a roadmap for a follow-up project. Target groups: learners, educators and simulation facilitators.
Method
AIDA follows an exploratory, design-based approach in four work packages. WP1: specify both use cases with practice partners (learning objectives, observable behaviours, quality criteria, dialogue logic and technical requirements) and develop a data-protection-by-design pipeline with de-identification before any AI analysis. WP2: configure two prototypes on afca’s existing XARA platform and test them internally with sample video. WP3: test the prototypes with small groups of learners and instructors, using structured observation, usability questionnaires, interviews or focus groups, and expert review of AI-generated observations against the quality criteria. WP4: synthesise the results into an implementation concept. Analysis is primarily descriptive and qualitative.
Planned Translation
Translation is built in from the start: learners, educators and simulation facilitators co-specify both use cases so the prototypes fit existing training formats. Practice partners are the Department of Emergency Medicine at Inselspital Bern with its simulation training, and [PRACTICE PARTNER] for the nursing-related vocational use case. Reusable outputs include operationalised quality criteria and debriefing structures that other educators can adapt, two demonstrable prototypes, and documented technical, didactic and data protection requirements. Dissemination will include a BeLEARN presentation, prototype demonstrations for educators and simulation facilitators, a conference contribution or short publication, and a roadmap for broader implementation.
AIDA shows which aspects of professional action AI can currently analyse reliably and which require human judgement, giving educators a realistic basis for decisions on AI-supported debriefing. For practice, it offers a scalable complement to facilitator-led debriefing, enabling more frequent, individual reflection on one’s own action. For research, it contributes criterion-based design principles transferable across vocational and clinical education. Impact will be measured through usability and acceptance questionnaires, interviews with learners and instructors on perceived usefulness for reflection, expert ratings of the accuracy and transparency of AI-generated observations against the quality criteria, and documented institutional interest in follow-up implementation.