Union of Intelligences. Augmenting European Policy-making with Trustworthy AI

Resumen

The integration of Artificial Intelligence (AI) into policy-making processes can lead to enhancing the efficiency of public administration.
This thesis, "Union of Intelligences: Augmenting European Policy-Making with Reliable AI," explores the implementation of AI within the European Commission (EC), focusing on Large Language Model (LLMs)- based services that are trustworthy and robust. The thesis explores practical approaches in the technical, social, and design domains.
The technical domain proposes Retrieval Augmented Generation (RAG) and agentic workflows to anchor the generated text in high quality and up-to-date documents and datasets. These techniques allow for a way to refer back to the official documents that generated the response making the responses more trustworthy. The technical chapter presents our experiments with processing and managing documents using open-source tools like KNIME and Python, highlighting the importance of secure internal developments and the role of metadata standards such as DCAT-AP in a RAG pipeline. This experimental component demonstrates the preprocessing of documents, vectorization techniques and LLM prompting; emphasizing the importance of associating metadata with document chunks for efficient retrieval and context preservation
The second chapter focuses on people. The thesis presents inclusive and participatory approaches in AI development and implementation. It suggests employing methods like Open Space Technology, World Café, and Ritual Dissent to foster collaborative discussions and co-create AI services that are user-centric and fit seamlessly in existing processes. Customized training programs tailored to the specific needs of EC staff are proposed to facilitate the adoption of AI technologies.
The design solutions focus on user engagement and service design principles to ensure AI tools are practical and effective. By incorporating systems thinking and co-creation workshops, the study emphasizes the need to align AI services with administrative processes and user needs, ensuring that AI supports rather than replaces human decision-making. The chapter presents design tools to be used in these processes.
By adopting a holistic approach spanning technology, people, and design, the EU can harness the power of LLMs to augment policy-making while maintaining the trustworthiness and human-centricity essential for public sector applications of AI.