The “Generation AI” project analyses how these tools affect students’ learning and digital well-being, and under what conditions they enhance or hinder inclusive STEM education across Europe.

GenAI applies a longitudinal, mixed-method design across 8 countries and 2 age groups (12-13, 16-17). Quantitative and qualitative data – including tests, AI literacy diagnostics, interaction logs, and interviews – are linked to usage, learner profiles and school context.

A novel benchmark of AI systems in STEM education links instructional quality and subject accuracy to student outcomes, supporting responsible, future-oriented AI use. GenAI evaluates AI-based interventions using a three-arm cluster-randomised design contrasting supported AI use, unsupported AI use, and a control group. This reveals conditions for effective, ethical integration in real classrooms.

Educators and policymakers are engaged through co-creation processes – including validation workshops, stakeholder meetings and targeted outputs such as teaching guidelines and policy recommendations – to ensure usability and uptake.

“Generation AI” supports HORIZON-CL2-2025-01-TRANSFO-07 by addressing the role of digital tools in everyday life and schools. It delivers interdisciplinary evidence, models, and resources to build inclusive, ethical, future-ready education systems across Europe.

To ensure cross-context comparability, GenAI applies a harmonised mixed-methods design across

• eight European countries, representing a range of school systems and digital maturity levels
• two key age groups (12–13 and 16–17), covering early and late adolescence at secondary school
• different STEM subjects (mathematics, science, technology & engineering)
• multiple student profiles, capturing differences in socio-economic status (SES), gender, prior achievement

and digital experience, thereby promoting diversity and equity

• combining quantitative and qualitative measurement for robust, in-depth evidence

Achievement of this objective will be measured by:

• A validated, multidimensional instrument for assessing digital well-being, STEM competences and AI literacy,
  developed through piloting and expert review.
• Large-scale quantitative data from ~8,000 students across eight countries (12–13 and 16–17 age groups).
• Qualitative interview data from 5 teachers and 30 students per country (total: 40 teachers and 240 students).
• A mixed-method research report synthesising findings across countries, profiles, and usage patterns.

More in detail, the project will focus on three core dimensions:

1. STEM proficiency (age- and subjectspecific problem-solving capacity)
2. Instructional quality (curricular alignment, depth, and clarity)
3. Adaptivity (responsiveness to misconceptions, feedback quality, and scaffold design)

Each dimension is operationalised through structured tasks and rubrics aligned with international standards (e.g. PISA). The benchmark will be applied to at least ten widely used AI systems (e.g. GPT-4, Gemini). Each of this 10 AI models will be tested on a common item pool through 100+ runs. This will produce the first systematic, crossplatform comparison of AI’s tutoring capacity, including pedagogical strengths, risks, and alignment with European educational values.

• Germany, PH Freiburg
• Netherlands, Utrecht University, Freudenthal Institute
• Spain, Jaen University
• Cyprus, Edex
• Austria, Univ. of Klagenfurt
• Portugal, Lisbon University
• Turkey, Haceteppe University
• Norway, Norges Tech.Nat. University
• Italy, Promoscience

• Brave, R., Russo, F., Uzovic, O. and Wagemans, J. (2022). Can an AI Analyze Arguments? Argument-Checking and the Challenges of Assessing the Quality of Online Information (PDF). In (pp. 267–281): Chapman and Hall/CRC.
• Meulenbroeks, R. and Van Joolingen, W. R. (2022). Students’ self-reported well-being under corona measures, lessons for the future (PDF) Heliyon, 8(1), e08733 doi:10.1016/j.heliyon.2022.e08733.
• Russo, F., Schliesser, E. and Wagemans, J. (2024). Connecting ethics and epistemology of AI (PDF) AI & SOCIETY, 39(4), 1585–1603 doi:10.1007/s00146-022-01617-6.
• Thurm, D., Vandervieren, E., Moons, F., Drijvers, P., Barzel, B., Klinger, M., Van Der Ree, H. and Doorman, M. (2023). Distance mathematics education in Flanders, Germany, and the Netherlands during the COVID 19 lockdown—the student perspective (PDF) ZDM – Mathematics Education, 55(1), 79–93 doi:10.1007/s11858-022-01409-8.
• Van der Lubbe, L., Van Borculo, S., Boon, P., Van Velthoven, W. and Jeuring, J. (2023). Bridging the Computer Science Teacher Shortage with a Digital Learning Platform (PDF), 15th International Conference on Computer Supported Education, CSEDU (pp. 289–296): SciTePress.