Human-Centred, Applied and Responsible AI in Education (HARAI-Ed)

Title: Human-Centred, Applied and Responsible AI in Education (HARAI-Ed)

Project duration: Ongoing

Research Area: Responsible AI

HARAI-Ed investigates how AI can support teaching, learning, and educational decision-making while remaining human-centred, inclusive, adaptive, and context-sensitive. The research connects technical AI methods with educational theory and practice, responsible AI principles, and human-centred design. Current work addresses learner-centred academic pathway guidance, explainable and interactive learning analytics, agentic AI and human-AI collaboration, and post-training AI intervention and control. A particular focus is the Kenyan Competency-Based Education (CBE) context, where research prototypes are being developed and evaluated in collaboration with Kenyan students, researchers, and educational stakeholders.

Aims

The research aims to develop educational AI systems that augment rather than replace human judgement and agency. It investigates approaches that enable learners and educators to understand, influence, question, and intervene in AI-supported processes while accounting for diverse learner needs, pedagogical contexts, and institutional requirements.

Problems

Educational AI is often developed primarily around prediction, automation, or system performance. HARAI-Ed addresses the challenge of developing AI systems that remain responsive to human goals and contextual needs, while being steerable, contestable, and correctable. This includes creating meaningful opportunities for human participation, oversight, intervention and control in educational processes and decision-making.

Practical example created during the project (if applicable)

Current work includes a human-in-the-loop AI system for learner-centred adaptive academic pathway guidance and an agentic AI – Teacher Teaming approach for formative assessment in Kenya’s CBE. These projects explore how AI can analyse learner information,  support educational decisions and guidance while retaining meaningful teacher and learner involvement.

A further research strand investigates machine unlearning and post-training model intervention for responsible educational AI. Current projects explore how trained models can be selectively modified or unlearn information, including for privacy protection and bias mitigation. Together, these approaches address not only how AI can support educational processes, but also how its behaviour can be revised when requirements, data, or circumstances change.

Technology

The research combines Reinforcement Learning, Explainable AI, Human-in-the-Loop approaches, learning analytics, agentic and multi-agent AI, and machine unlearning. These methods are integrated with educational frameworks to develop AI systems that can respond to diverse educational contexts while supporting:

• transparency,
• human agency and intervention, and
• responsible use.

Outlook

HARAI-Ed will continue to develop and empirically validate its existing research prototypes, translate mature projects into scholarly publications, and explore emerging directions including responsible agentic AI, trustworthy educational decision support and guidance,  and post-training intervention. Further research will strengthen collaboration between Germany, Kenya and wider Africa and support the development of new research, funding, and knowledge-transfer initiatives.

Publications

• Machine Unlearning for Responsible and Adaptive AI, ESORICS 2025 – STMUS.

• From “Where Do I Fit?” to “Where Do I Want to Go?”: Learner-Centred, Adaptive Aspiration-Oriented Academic Pathway Guidance with Reinforcement Learning, submitted to CHI 2027.

• Further manuscripts on learner modelling, academic pathway guidance, machine unlearning, and responsible educational AI are in preparation.

Team

Lead

Dr Betty Mayeku

Partners

• Kibabii University, Kenya
• other academic and educational partners