Our mission

Artificial intelligence (AI) has demonstrated significant potential in addressing various challenges within the education sector. Despite its potential, AI in education is often criticized for neglecting key learning processes such as motivation, emotion, and (meta)cognition. Many systems are built with little input from educators or stakeholders, function largely as black boxes, and overlook ethical issues like data quality and bias. Current approaches also prioritize automated, individualized instruction over fostering metacognitive and self-regulated learning. Even with these shortcomings, AI technologies are steadily being integrated into public education.

Our mission is to shift the focus from AI in education to AI for education by developing responsible, transparent, and co-designed AI methods that truly serve learners and educators. We emphasise both the responsible development and the responsible use of AI, building hybrid human–AI systems that support human judgement, and empower stakeholders at every stage of learning.

From AI in education to AI for education

At our research group, we make a clear distinction:

• AI in education often means applying generic AI methods to educational contexts, such as tutoring systems with purely data-driven models, developed without meaningful input from stakeholders (e.g., educators, learners, researchers) or validated educational data.

• AI for education, in contrast, means hybrid human–AI collaboration at two stages:

  • Responsible development – AI methods and systems are co-developed with stakeholders and developers.

  • Responsible use – AI systems facilitate collaboration between humans and AI in practice.

We believe this paradigm shift from AI in to AI for education is urgently needed. Crucially, responsible use of AI in classrooms depends on responsible development from the start. While technical fixes—such as fine-tuning data-driven models to reduce hallucinations or mitigate bias—may address surface problems, they cannot resolve the deeper structural flaws that arise when systems are built without meaningful human collaboration.

What we mean by Responsible AI?

We subscribe to definition of Goellner and colleagues (2024), viewing responsible AI as a human-centred approach that builds user trust through ethical decision-making, explainable outcomes, and privacy-preserving implementation.

Responsible development through Hybrid Human-AI 

In the development stage, we employ hybrid AI methods that combine data-driven learning with stakeholders’ explicit knowledge. Examples include probabilistic graphical models (e.g., Bayesian networks) and neural-symbolic computing. 

Stakeholders—such as teachers, psychologists, and domain experts—can actively participate by injecting their practical and theoretical knowledge into these models. This symbolic knowledge constrains and guides data-driven learning, making models more context-aware and educationally meaningful. Examples of such knowledge include:

• Causal relationships among variables 

• Theoretical models linking learning constructs 

• Interactions between variables affecting learning outcomes

• Ethical constraints, such as fairness principles that prevent biased treatment of specific learner groups 

• Privacy-preserving rules, such as limiting reliance on sensitive personal data while still enabling accurate modelling

Benefits of Knowledge-Enhanced Models

This approach directly promotes responsible AI by design: 

• Human-centredness through co-development with stakeholders

• Ethical decision-making by avoiding spurious correlations and addressing data quality issues

• Interpretability by using both injected knowledge and learned patterns to make model reasoning transparent

• Privacy preservation by reducing reliance on sensitive student data

• User trust through stakeholder co-development, and through ethical, transparent, and privacy-aware decision-making in development

Does this mean the models merely follow stakeholder instructions and lose their data-driven power?

No. They still freely learn patterns from data—both with and without stakeholder knowledge—ensuring a balance between pattern recognition and educationally grounded guidance.

Responsible use through Hybrid Human-AI collaboration

Responsible use of AI in education means going beyond automation to ensure that systems support human judgement, learner agency, and meaningful learning processes. Many of the existing AI-based systems (e.g., large language models and purely data-driven systems) cannot reliably track skill mastery, capture motivation or emotion, or inform learners and teachers about a student’s full learning trajectory. They often overemphasize individualized instruction while neglecting the development of metacognitive and self-regulated learning skills, risking learners becoming passive recipients. By contrast, responsible use emphasizes hybrid human–AI collaboration, where AI tools act as partners—providing transparent insights, supporting reflection, and empowering both students and teachers—while keeping critical educational decisions under human oversight.

Does this mean we should not use large language models or other fully data-driven systems in education?

Not at all. These systems can be highly valuable and should be used—but always within a thoughtful, hybrid framework where human oversight, stakeholder knowledge, and ethical considerations guide their application.