Realising Lawful and Trustworthy AI: An Analysis of the Requirements of Trustworthy AI and EU AI Act
Author
Andersen, Frederik Olsen
Term
4. semester
Education
Publication year
2022
Submitted on
2022-06-10
Abstract
This thesis investigates how the European Union’s initiatives for lawful and trustworthy artificial intelligence can be put into practice. It analyzes the EU High-Level Expert Group’s Guidelines for Trustworthy AI and the proposed EU Artificial Intelligence Act, focusing on how their requirements can be implemented in the development and deployment of AI systems, particularly those deemed high-risk. Through a review and analysis of the guidelines, legal obligations, and relevant literature (including explainable AI), the work compiles practical recommendations—methods, techniques, and practices—to meet the identified requirements and synthesizes them into an operational model with a prototype to support organizations in achieving legality, ethics, and robustness in AI. The excerpt does not report an empirical evaluation, but the main contribution is a consolidated set of implementation suggestions and an actionable model intended to help practitioners engineer AI systems that are trustworthy and aligned with the EU framework.
Dette speciale undersøger, hvordan EU’s initiativer for at fremme lovlig og tillidsvækkende kunstig intelligens kan omsættes til praksis. Det analyserer EU’s High-Level Expert Group on AI’s retningslinjer for Trustworthy AI og det foreslåede EU-regulativ om AI (AI Act) med fokus på, hvordan kravene kan realiseres i udvikling og anvendelse af især højrisiko-AI-systemer. Gennem en gennemgang og analyse af retningslinjer, lovkrav og relevant faglitteratur (herunder forklarlig AI) samler arbejdet konkrete anbefalinger til metoder, teknikker og praksisser for at opfylde de identificerede krav og omsætter dem til en operationel model med prototype, der kan støtte organisationer i at etablere lovlighed, etik og robusthed i AI. Uddraget rapporterer ikke en empirisk evaluering, men hovedbidraget er et konsolideret katalog af implementeringsforslag og en anvendelsesorienteret model, der kan hjælpe praktikere med at bygge AI-systemer, der er tillidsvækkende og i overensstemmelse med EU’s ramme.
[This apstract has been generated with the help of AI directly from the project full text]