Using Large Language Models for the Enforcement of Consumer Rights in Germany

Eurosymposium 2023

At Eurosymposium 2023 in Sopot, I virtually presented our paper, “Using Large Language Models for the Enforcement of Consumer Rights in Germany,” a part of the KIVEDU project. This work delves into the potential of Artificial Intelligence, specifically Large Language Models (LLMs), to automate the enforcement of consumer rights in the European legal context.

Our primary focus is on automating the examination of cease-and-desist declarations to ensure compliance with European consumer law. The traditional approach to this task has been manual, time-consuming, and prone to errors, leading to a risk of ongoing unlawful practices by companies. To address this, we propose an AI-driven system capable of monitoring, analyzing, and documenting potential violations of these declarations on the Internet.

The proposed system is designed to operate in the background, scanning websites for potential infringements, and using AI models to detect violations. This interdisciplinary project combines expertise from both technical and legal professionals, and is bolstered by our partnership with a consumer protection center, providing a substantial dataset for training the AI.

We discuss the software architecture of the system, designed around service-oriented and event-driven principles, with a focus on cloud-native components. The architecture ensures reusability, modularity, scalability, and flexibility, with components like a web application, APIs, NoSQL databases, and serverless functions for various tasks including text extraction and violation archiving.

The paper also outlines specific use cases, such as food labeling, compliance with legal obligations, data protection, and trademark usage, showcasing the broad applicability of the system. Additionally, we address the legal and technical challenges in developing such a system, emphasizing GDPR compliance, AI regulation, and intellectual property law.

We conducted preliminary research on suitable LLMs, balancing performance with privacy considerations. Our findings highlighted proprietary models like ChatGPT for performance, and open-source models for data privacy, with a particular interest in the Luminous and MPT models.

In conclusion, the KIVEDU project is a pioneering effort in using AI to automate the monitoring and documentation of cease-and-desist declaration violations. We are committed to advancing this endeavor, with an emphasis on strengthening consumer rights in the EU and Germany. I enjoyed sharing our work at the Eurosymposium 2023 and engaging with the community on these critical issues.