Navigation auf uzh.ch
Navigation auf uzh.ch
Note: These are preliminary considerations based on our current understanding of the EU AI Act. For specific legal compliance requirements, consultation with legal experts (e.g., UZH RUD) is necessary. For detailed requirements and specific provisions of the EU AI Act, refer to the official documentation at digital-strategy.ec.europa.eu.
The EU AI Act explicitly classifies AI systems used for student assessment and evaluation of learning outcomes as high-risk systems. This classification affects our work in AI-assisted exam grading because these systems can significantly impact educational outcomes and student opportunities. The Act recognizes the potential influence these systems have on fundamental rights in education and future career paths.
The Act mandates several key requirements for high-risk AI systems. Data quality and governance form the foundation, requiring training data to be representative and carefully documented. Human oversight must be built into the system's design, ensuring that graders can effectively monitor and intervene in the AI's operation. The system must also maintain comprehensive documentation and provide transparency about its operations and decisions.
The EU AI Act enters into force on August 1, 2024, with the first provisions being enforced from February 2025. This timeline gives educational institutions and technology providers a transition period to adapt their systems and processes. The gradual implementation allows for careful consideration of compliance requirements while continuing to improve assessment technologies.
While Switzerland currently has no specific AI legislation, the EU AI Act will likely influence Swiss practices significantly through both direct and indirect channels. Swiss educational institutions developing or using AI-assisted grading systems should monitor these developments closely. The Federal Council is preparing regulatory approaches by early 2025, and Switzerland's close ties with the EU suggest eventual alignment with EU standards. For Swiss universities, proactively following EU AI Act guidelines represents a prudent approach to future-proof AI-assisted assessment systems.
For our project, the Act's requirements align well with our human-in-the-loop approach to AI-assisted grading. The system must enable human oversight of grading decisions while maintaining detailed documentation of the process. This approach ensures that AI assists rather than replaces human judgment in assessment, sup-porting our framework's distinction between assisted, semi-automated, and automated grading approaches.
Risk management forms a crucial part of compliance with the Act. This includes regular assessment of system accuracy, monitoring for potential biases, and maintaining clear documentation of mitigation strategies. The focus should be on ensuring that the AI system enhances rather than compromises the fairness and accuracy of student assessment.
