This is the source code for Website (https://cedanl.github.io/AI-Alignment/)
about AI alignment.
The website is part of a project to support the Dutch Education community via CEDA. Its main aim is to stimulate conversation about comparing the alignment of AI models and systems against what we consider important.
- Information Dashboard - An information dashboard about AI Alignment. The menu option Scorecard/Visualization is a mockup of a potential community process.
At the end of 2025 the Npuls LA team came to the following conclusion about the state of Generative AI in Dutch education:
The importance of LA in (Gen)AI was best summarized by the “Confused Expert” analogy. This argument posits that GenAI is a powerful but unguided medium, not a solution. The LA cycle, grounded in pedagogical data, clear metrics, and a feedback loop, is an essential framework for guiding AI responsibly and effectively in an educational context.
Effective AI will need better data and guiding practices, not the other way around.
The solution argues for Benchmarking and training AI models on pedagogical data, and using the LA cycle to guide the development and deployment of AI in education. This approach ensures that AI systems are aligned with educational goals and can be effectively integrated into teaching and learning processes.
Before taking time efort and gold to train models based on Learning Analytics data we need to be able to measure the qualities of the models and AI systems against our values. This project is an initial review of the theme with the goal of developing a framework for benchmarking and scorecarding AI systems in education, with a focus on alignment with educational values and goals. The project will involve identifying relevant metrics, developing benchmarking methodologies, and creating scorecards to evaluate AI systems in the context of education.
The LA cycle, grounded in pedagogical data, clear metrics, and a feedback loop, is an essential framework for guiding AI responsibly and effectively in an educational context.