LSP3 – Education: Trustworthy AI for Deeper Learning in STEM Education

The Education LSP, one of the seven Large Scale Pilots (LSPs) of the FAITH project, focuses on assessing the trustworthiness of novel, AI-based assessment methods of students’ complex skills as problem-solving. This is realized through inquiry-based activities for STEM subjects, ensuring that the AI tools shaping the next generation of learners are not only effective, but genuinely trustworthy.

The Educational LSP AI Solution, available through trust-ai-lab.eu, offers AI-enhanced personalized learning and formative student assessment. It also allows students to pursue goals that require extended engagement or persistence across multiple contexts and learning opportunities, furthermore enabling teachers to track student progress. Since the launch of the 1st Pilot on May 2024, the solution developed by IMU/ICCS has been deployed extensively. Inquiry-based scenarios have been implemented repeatedly at Ellinogermaniki Agogi and in other schools across Greece and other European countries, serving both to validate the pedagogical approach and to train the AI models of the platform.

Alongside student implementations, a series of workshops and teacher training activities have been delivered. During these sessions, teachers explored the platform, engaged with the technical team on the AI methodology behind the solution, and took part in discussions on trustworthy AI in education. Participants then completed questionnaires evaluating the trustworthiness of the AI tools in the context of inquiry-based learning, directly feeding into the project’s impact assessment.

From the outset, the solution has been built around a clear vision: that trustworthy AI and AI should be synonymous. This means that every design and implementation decision is guided by the core trustworthiness dimensions most relevant to educational AI:

• Validity, thus ensuring AI-generated content is scientifically accurate and pedagogically sound, directly impacting student engagement and learning outcomes.
• Explainability, thus making the AI’s reasoning transparent and understandable to both teachers and students, fostering trust and informed use of the system.
• Fairness & non-discrimination, which is designing assessments and recommendations that are unbiased and equitable across all learners.
• Security & Data Sovereignty, which is protecting student data and ensuring compliance with privacy and security standards throughout the platform.

A rigorous in-isolation trustworthiness assessment of the Educational LSP AI Solution has been completed across all AI lifecycle phases, using the FAITH-AI – TAF tools. The impact assessment identified validity and explainability as the most critical dimensions of trustworthiness for the education LSP. To conduct the trustworthiness assessment, ICSS and EA participated in dedicated training workshops on the use of the FAITH-AI-TAF tools. The Replication Pilot is now underway, with plans to conduct an in-context trustworthiness assessment of the solution in real classroom environments across Europe. The Educational LSP AI Solution represents a new standard for responsible AI in education, one that places student learning, teacher empowerment, and trustworthiness at its heart.