arXiv subjects

Frederike Zufall

Publications and source records attributed to Frederike Zufall.

4 recordsLinked to original sources

arXiv2026

Qualifying and Quantifying Risk Under the EU AI Act

The EU AI Act uses a risk-based approach to regulate AI systems, calibrating the intensity of regulation according to the risks they pose. While the term 'risk' implies quantification, resulting from the combination of the probability and severity of harm, the AI Act refers to risks to fundamental rights, thereby engaging a qualitative perspective. In this piece, we address this puzzle using a two-step framework under which the EU AI Act balances risks with the protection of fundamental rights, the legitimate purposes of providers and deployers, and the impacts of regulatory measures on providers, deployers, and regulators. We discuss this framework against the backdrop of potential approaches to quantifying risks, with a specific focus on defining and measuring the main components of the concept of risk: probability, severity, and their combination. We suggest that the protection of fundamental rights and risk quantification can be aligned by incorporating quantification methodologies into the proposed framework. In particular, the AI Act implies a balancing analysis that uses a severity-first approach to classify and quantify the risks posed by AI systems. The integrated framework not only helps to clarify the AI Act's risk-based approach, but can also inform technical and implementation choices. Finally, we conclude that if the risk quantification methodology or its application to the protection of fundamental rights is left to providers and deployers, there is potential for 'risk hacking', which could lead to the underclassification of AI systems and subsequent regulatory shortcuts.

cs.CY↗

arXiv2025

Conditioning Large Language Models on Legal Systems? Detecting Punishable Hate Speech

The assessment of legal problems requires the consideration of a specific legal system and its levels of abstraction, from constitutional law to statutory law to case law. The extent to which Large Language Models (LLMs) internalize such legal systems is unknown. In this paper, we propose and investigate different approaches to condition LLMs at different levels of abstraction in legal systems. This paper examines different approaches to conditioning LLMs at multiple levels of abstraction in legal systems to detect potentially punishable hate speech. We focus on the task of classifying whether a specific social media posts falls under the criminal offense of incitement to hatred as prescribed by the German Criminal Code. The results show that there is still a significant performance gap between models and legal experts in the legal assessment of hate speech, regardless of the level of abstraction with which the models were conditioned. Our analysis revealed, that models conditioned on abstract legal knowledge lacked deep task understanding, often contradicting themselves and hallucinating answers, while models using concrete legal knowledge performed reasonably well in identifying relevant target groups, but struggled with classifying target conducts.

cs.CL↗

arXiv2024

Operationalizing content moderation "accuracy" in the Digital Services Act

The Digital Services Act, recently adopted by the EU, requires social media platforms to report the "accuracy" of their automated content moderation systems. The colloquial term is vague, or open-textured -- the literal accuracy (number of correct predictions divided by the total) is not suitable for problems with large class imbalance, and the ground truth and dataset to measure accuracy against is unspecified. Without further specification, the regulatory requirement allows for deficient reporting. In this interdisciplinary work, we operationalize "accuracy" reporting by refining legal concepts and relating them to technical implementation. We start by elucidating the legislative purpose of the Act to legally justify an interpretation of "accuracy" as precision and recall. These metrics remain informative in class imbalanced settings, and reflect the proportional balancing of Fundamental Rights of the EU Charter. We then focus on the estimation of recall, as its naive estimation can incur extremely high annotation costs and disproportionately interfere with the platform's right to conduct business. Through a simulation study, we show that recall can be efficiently estimated using stratified sampling with trained classifiers, and provide concrete recommendations for its application. Finally, we present a case study of recall reporting for a subset of Reddit under the Act. Based on the language in the Act, we identify a number of ways recall could be reported due to underspecification. We report on one possibility using our improved estimator, and discuss the implications and areas for further legal clarification.

cs.SI↗

arXiv2021

A Legal Approach to Hate Speech: Operationalizing the EU's Legal Framework against the Expression of Hatred as an NLP Task

We propose a 'legal approach' to hate speech detection by operationalization of the decision as to whether a post is subject to criminal law into an NLP task. Comparing existing regulatory regimes for hate speech, we base our investigation on the European Union's framework as it provides a widely applicable legal minimum standard. Accurately judging whether a post is punishable or not usually requires legal training. We show that, by breaking the legal assessment down into a series of simpler sub-decisions, even laypersons can annotate consistently. Based on a newly annotated dataset, our experiments show that directly learning an automated model of punishable content is challenging. However, learning the two sub-tasks of `target group' and `targeting conduct' instead of an end-to-end approach to punishability yields better results. Overall, our method also provides decisions that are more transparent than those of end-to-end models, which is a crucial point in legal decision-making.

cs.CL↗