Institutional Repository of Vilnius University
Modeling of ai system risk identification using decision graphs /
Abstract
dc:descriptionArtificial Intelligence is a key driver of innovation, yet its probabilistic nature introduces issues that traditional deterministic systems do not face. To mitigate concern over this gap, the European Union has established the AI Act. This regulation categorizes AI systems into four risk tiers and applies obligations based on the classification. Existing compliance tools try to facilitate compliance, however, a significant gap remains for tools that not only encompass all risk groups and identify them, but that are also accessible. This thesis proposes a tool prototype modeled in a multi-agent graph topology that utilizes Retrieval-Augmented Generation (RAG) to fetch specific legal context from the AI Act. It incorporates an Ambiguity Type-Chain of Though (AT-CoT) prompting scheme to initiate clarification when user input is insufficient to form a verdict. The prototype was evaluated quantitatively over 36 individual scenarios and qualitatively through an expert-led analysis. Testing results demonstrated an overall accuracy of 86.1%, while the clarification methodology received a 4 out of 5 utility score from the expert. The thesis concludes that a structured graph-based orchestration works well on legal reasoning tasks. Furthermore, clarification strategies are crucial for bridging the gap between natural language and legal regulatory terms.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Ruginytė, Akvilė,
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- eng
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD308338661&prefLang=en_US