XAI-CRED: Explainable Artificial Intelligence and Credibility in Asylum Decision-making
XAI-CRED pioneers a new type of explainable artificial intelligence focused on credibility assessment to improve transparency, fairness, and decision-making in asylum procedures while pushing the limits of what we know about human decision-making processes. The project addresses both the technical 'black-box' problem in AI systems and the human 'black-box' of intersubjective judgment and bias in legal credibility assessments.
XAI-CRED is a five-year project (2025-2029) led by Prof. Thomas B. Moeslund from Aalborg University and Prof. Thomas Gammeltoft-Hansen from the University of Copenhagen, with co-PI Prof. Henrik Palmer Olsen. The project leverages recent methodological advances from Explainable AI (XAI) to develop algorithms that can provide meaningful explanations of AI decision processes. The aim is two-fold: 1) to unpack both the human and technical black boxes of decision-making in asylum law, and 2) to pioneer the development of models generalizable beyond the legal domain where intersubjective credibility assessment plays an important role.
Drawing on an established interdisciplinary collaboration and unique access to a large dataset of Danish asylum decisions, XAI-CRED will advance insights for XAI and legal understandings of asylum law. The project will analyze approximately 16,000 full-text Danish asylum case files (1995-2020) through five progressive research stages: C1-DEFINE, C2-COMPUTE, C3-UNDERSTAND, C4-EXPLAIN, and C5-FEEDBACK. This structured approach will employ natural language processing, network analysis, eye-tracking, and other advanced techniques to develop a comprehensive XAI model for credibility assessment in asylum decision-making.
The project is organized around five interrelated challenges:
C1 – DEFINE (Credibility in asylum decision-making): focuses on obtaining a deeper understanding of credibility and developing an AI model to classify this.
C2 – COMPUTE (Computing credibility information): focuses on how to compute information regarding credibility decision-making from a technical point of view.
C3 – UNDERSTAND (Understanding credibility concepts): tackles the challenge of defining concepts from a legal perspective.
C4 – EXPLAIN (Explaining credibility): combines previous learnings to develop an XAI algorithm that makes predictive AI meaningful to domain experts.
C5 – FEEDBACK (Enriched decision-making through XAI) will develop feedback tools for domain experts and embed research in education across both disciplines.
The project will utilize a unique dataset of approximately 16,000 full-text Danish asylum case files covering the period 1995-2020. Using advanced network analysis, eye-tracking experiments, and amnesic probing, the researchers will identify patterns in credibility assessments and develop tools to improve decision-making.
The project is led by Prof. Thomas B. Moeslund (Aalborg University, Faculty of IT and Design) and Prof. Thomas Gammeltoft-Hansen (University of Copenhagen, Faculty of Law), with Co-PI Prof. Henrik Palmer Olsen (University of Copenhagen, Faculty of Law).
XAI-CRED includes international collaboration with Søren Jørgensen (Center for Human Rights and International Justice, Stanford University). As former Danish Consul General and head of Innovation Center Denmark in Silicon Valley, Jørgensen contributes expertise in ethics, law, and technology. He will host a two-week retreat for the team at Stanford and provide continuous input on real-life challenges associated with XAI and law.
The project brings together the AI for the People and REPAI Centers at Aalborg University with the MOBILE Center of Excellence and Nordic Asylum Law & Data Lab at the University of Copenhagen, building on strong complementary competencies in data science and asylum law through interdisciplinary collaboration.
Researchers
University of Copenhagen
| Name | Title | |
|---|---|---|
| Gammeltoft-Hansen, Thomas | Professor with special responsibilities (PI1, XAI-CRED; PI, NordASIL; PI, DATA4ALL; PI, AFAR; PI, XAIfair) |
|
| Olsen, Henrik Palmer | Professor in Jurisprudence (PI2, XAI-CRED; co-PI, NordASIL) |
Aalborg University
| Name | Title | |
|---|---|---|
| Moeslund, Thomas B. | Professor (CO-PI, XAI-CRED; XAIfair) |
 |
| Name | Title | Phone |
|---|
Funded by:
Project: XAI-CRED: Explainable Artificial Intelligence and Credibility in Asylum Decision-making
Period: 2025-2029
Contact
PI Professor
Thomas Gammeltoft-Hansen
South Campus,
Building: 6B.4.40
DK-2300 Copenhagen S
Phone: +45 50 20 34 00
E-mail: tgh@jur.ku.dk