This leaves out questions like: Which codes (for labeling the data), themes, and analytical approaches did the AI propose? Which did the researcher accept, revise, or throw out? Where did the analytic argument in the paper actually come from? This gap matters in research, where the credibility of a finding rests on being able to trace it back through the analysis process.
More broadly, addressing this problem will give you the opportunity to engage both practically and theoretically with an issue at the heart of current discussions of human-AI collaborations. As we do more and more work with AI, how can we disclose the use of Gen-AI in a transparent way?
Who is this suitable for
Master students in the DESIGN programme, who are interested in research practices and human-AI collaboration. Having the technical skills to create a working prototype is a prerequisite.
The project
The master project will design and carry out a Design Science Research (DSR) project situated in the Information Systems field. The task is to design, build, and evaluate a working artifact, meaning a system that makes AI-assisted qualitative analysis auditable.
Concretely, this means an artifact that can:
- Capture the analysis process as it happens, like coding decisions, Gen-AI-generated suggestions, researcher acceptances, revisions, and rejections. Optimally, this is done as a structured and tamper-evident trail of evidence.
- Represent that trail in a form that other researchers, supervisors, or reviewers can understand and inspect.
- Integrate into current research practices.
What a DSR approach means:
Following the DSR methodology, the thesis is about building something that operationalises principles for good and ethical use of AI in research, and then evaluating whether it actually works for the people it is meant to serve. Concretely:
- Problem framing and requirements, grounded in the literature on research transparency, provenance, and AI governance in academia.
- Designing and building a working prototype. The scope (e.g., a standalone tool or a logging layer that sits under an existing environment) is open and should be settled with the supervisors early on, matched to your technical interests and the time available.
- Evaluation with real researchers: the artifact should be tried out by a small number of qualitative researchers on, assessing whether it captures the process faithfully, whether the resulting trail is actually legible and useful, and whether it is something researchers would plausibly adopt.
Supervision
Sune Dueholm Müller and Alexander Moltubakk Kempton