DKK 7.1 million for ITU research into a new generation of educational AI
Backed by Independent Research Fund Denmark (IRFD), a new interdisciplinary project will investigate how AI chatbots can encourage students to think more critically by introducing carefully designed ambiguity rather than providing instant answers.
Christian HardmeierJichen ZhuPawel GrabarczykResearchalgorithmsartificial intelligencegrants
Written 9 November, 2026 07:56 by Theis Duelund Jensen
Generative AI tools such as ChatGPT are rapidly transforming education. While many students use large language models (LLMs) to complete assignments more efficiently, teachers and researchers are increasingly concerned that convenience comes at a cost: students may achieve good results without developing the underlying skills and critical thinking that learning is meant to foster.
A new five-year research project at the IT University of Copenhagen (ITU) aims to tackle that challenge by turning a common assumption about AI’s efficiency and accuracy on its head. Rather than making tasks easier and faster, the project will investigate how AI can introduce carefully designed "positive friction" in the learning process that encourages students to reflect, question, and reason for themselves.
Led by Professor Jichen Zhu, with Associate Professor Christian Hardmeier as co-PI, as well as Associate Professor Pawel Grabarczyk from ITU’s Center for Computing Education Research (CCER), who contributes domain expertise on ethics and responsible AI, the interdisciplinary project brings together expertise from human-computer interaction (HCI), artificial intelligence, and learning sciences. The project focuses specifically on ethics education and zooms in on how AI can support students in engaging with complex ethical questions rather than simply providing ready-made answers.
“Much of today's AI in education is optimised for outcomes and efficiency. But learning is not only about getting the right answer. It is about the process of thinking, questioning, and developing understanding,” says Jichen Zhu, Professor at ITU and principal investigator of the project.
“We want to explore how AI can help students engage more deeply with difficult questions by giving them space and productive obstacles to overcome.”
Using ambiguity as a learning tool
The project, Friction by Ambiguity: Improving Engagement and Critical Thinking in LLM-Supported Learning, is built around a novel idea: controlled ambiguity.
Instead of delivering straightforward answers, AI systems may sometimes provide responses that are intentionally open-ended or incomplete. Combined with appropriate pedagogical support, this ambiguity can prompt students to evaluate different interpretations, weigh arguments, and develop their own positions.
The researchers argue that this approach may help address growing concerns about overreliance on AI tools and the cognitive offloading that jeopardizes important learning skills.
A central outcome of the project will be the development of Diffrakt, an AI-powered educational chatbot designed to support the teaching of AI ethics. Drawing on the Socratic method, the chatbot will use guided ambiguity and questioning techniques to stimulate discussion and reflection rather than delivering definitive answers.
The project will also investigate new technical methods for generating responses with calibrated levels of ambiguity based on a model's uncertainty. This aspect of the work is led by Christian Hardmeier, whose research focuses on uncertainty communication and language technologies.
“Current general-purpose AI systems were not designed with educational goals in mind,” says Jichen Zhu. “Some students use them productively, while others use them to offload the intellectual work. We are interested in a different vision of AI for learning, where the technology acts as a sounding board that helps learners find their own path.”
Bringing together AI, ethics and human-centred design
The project combines technical AI research with human-centred design methods. Researchers will work closely with students and educators, observing how learners respond to AI systems that incorporate guided ambiguity and iteratively refining both the technology and its design.
While ambiguity has long been recognised as an important aspect of human communication, relatively little research has explored how it can be used productively in AI-supported learning environments.
By combining insights from HCI, AI, and ethics education, the researchers hope to develop new design principles for educational technologies and new technical approaches for controlling ambiguity in language models. Teachers and educators will also play an active advisory role throughout the project to ensure that the research addresses real educational needs and classroom practices.
The research will initially be tested with students at ITU, with the longer-term ambition of informing future educational technologies and practices both within and beyond higher education.
Theis Duelund Jensen, Press Officer, phone +45 2555 0447, email thej@itu.dk