Traditionally Artificial Intelligence (AI) was primarily focused on building adaptive tutoring systems that mimicked the role of teachers to deliver personalized instruction. Over time, AI applications have expanded to other domains, such as drop-out prediction and performance analytics, with a central goal of understanding and enhancing learning. This expansion has driven the growth of research fields like Educational data mining, Learning Analytics, and AI in Education. These fields have illustrated the potential of AI harnessing data from learning platforms and even from physical classroom spaces. Thus, AI can help teachers to eficiently observe and understand what is happening in their classrooms, augmenting the teacher's ability to maximize positive impact on learning. One emerging approach to achieving this synergy between humans and AI is hybrid intelligence, which emphasizes the collaboration and co-evolution of humans and AI. In this paper, we present our ongoing research eforts to design and develop educational technologies with an ability to evolve and adapt from their interactions with teachers and students, and align with human values and norms.
Technological advancements particularly in high-performance computing and big data, have enabled the generation and processing of vast amounts of learning traces -from student interactions with learning environments- with the purpose of enhancing student learning. Consequently, research fields such as Learning Analytics (LA), education data mining, and artificial intelligence (AI) in education emerged, demonstrating AI capabilities to harness learning data to extract patterns and insights from learning data. However, learning occurs within contexts, and without a holistic understanding, the full potential of those patterns and insights may remain unrealized.
As Akata et al. [
Emerging research has increasingly focused on the design of efective Teacher-AI partnerships in
education, emphasizing its potential to enhance student outcomes by complementing teacher expertise
with AI-driven insights and tools. Earlier studies such as Holstein and Aleven [
CEUR
ceur-ws.org
AI collaboration, which not only requires the development of sophisticated AI systems but also a deep understanding of teachers’ needs, perceptions, and practices to ensure meaningful collaboration.
Teacher-AI collaboration can be viewed through the lens of complementarity, where teachers and AI
systems bring distinct, yet synergistic strengths to the learning environment. However, research by
Kim [
Earlier research has shown that to fully support teacher-AI complementarity, it is essential to design
AI tools that make pedagogical models transparent for teachers, enabling them to trust and efectively
use these tools in practice [
We present three use cases of teacher-AI complementarity, which are currently being developed by our research group, to improve teachers’ understanding of how students’ complex learning processes in mathematics or science can be supported.
This case is based on the premise that each student learns at a diferent pace, requiring a learning
environment that supports personalized learning experiences. We argue that the design of
instructional practices should account for students’ individual characteristics - such as prior knowledge,
self-regulation, and other skills to direct their knowledge in a flexible learning environment. Our
approach [
The domain model for a specific subject is developed following the Estonian national curriculum. In
this process, subject experts encode their domain knowledge in a machine-readable format, enabling
integration with technologies to support flexible instructional trajectories. The instructional trajectories
resulting from the domain model comprise a series of episodes, tasks, hints, and additional learning
resources. These trajectories are aligned with the principles of the 4C/ID model [
The tools were designed over three years, following a Design-Based Research (DBR) methodology in multiple phases. Each DBR phase involved evaluations with teachers and students. The evaluations with the teachers happened in the first phase, where they used ready-made learning resources and adapted them on a small scale. Despite their limited experience, they still found the tools beneficial for adding flexibility to the learning process. The students had mixed experiences: some benefitted from its ease of use, while others faced cognitive overload, reducing its eficiency.
The Teacher-AI complementarity is evident through a balanced partnership where technology supports teachers in understanding how students develop their competence. AI handles tasks such as tracking student progress, identifying learning patterns, and ofering actionable insights by analyzing the relationships between skill acquisition and knowledge development. This allows teachers to focus on interpreting these insights to identify areas where students may need additional support or challenges, guiding their learning more efectively.
The previous case argued the importance of designing flexible learning trajectories based on a domain
model to support students’ individual learning needs. However, designing a trajectory is only one phase
as it is also important to formulate and sequence tasks and learning activities to foster deeper learning
experiences and sustain cognitive engagement. Here, the ICAP framework [
Addressing this need, we developed Kool-Kit, a suite of tools that uses large language models to provide a system grounded in the ICAP framework (figure 2). Kool-Kit supports teachers in designing lesson plans and creating H5P-based learning activities adjusted for their classrooms. The first tool in the suite is strongly based on ICAP to guide teachers in creating and refining lesson plans that promote diferent levels of engagement through chat-based interactions. The second tool ofers an intuitive interface for specifying learning activities. Teachers can define the activity’s summary, complexity level, and type (e.g., fill in the blanks, multiple-choice questions), and the tool generates H5P-based learning activities based on these inputs. These activities can then be downloaded and seamlessly integrated into any H5P-compatible learning environment (e.g., trajectory in Case 1). By simplifying the creation process, the tool hides the technical complexities of generating H5P elements while significantly speeding up the preparation of classroom activities. This makes it easier for teachers to design and implement engaging learning experiences that align with pedagogical principles and support student engagement. These tools are suitable for teachers from diferent subjects and school levels, as its foundational model focuses on the general principles of cognitive engagement. The initial prototype was developed collaboratively within the research group, addressing the practical need for supporting teachers in designing TEL practices that promote cognitive engagement. However, it has not yet been evaluated.
The teacher-AI complementarity in Kool-Kit lies in its ability to foster a mutual relationship between human and artificial intelligence. Kool-Kit harnesses AI to learn from teachers’ experiences (earlier lesson plans and examples), adapting to their expectations and preferences, while simultaneously enabling teachers to benefit from AI’s knowledge of learning sciences literature. This dynamic interaction allows teachers to advance their pedagogical knowledge, refine their teaching strategies, and design more efective learning experiences.
The third case extends the focus on designing learning experiences to support students’ complex thinking skills and understanding their progress, recognizing that learning is a dynamic, collaborative process. Collaboration is a multifaceted skill involving components such as argumentation and cooperation, which require deliberate practice and teacher awareness of student activities during collaborative tasks. Monitoring students’ interactions, particularly in blended settings that combine face-to-face collaboration with digital tools, presents significant challenges for teachers and underscores the need for robust tools to aid collaboration monitoring.
To address this, we developed CoTrack, a multimodal LA tool designed to support teachers in
monitoring and co-regulating collaboration in classrooms. CoTrack is a web-based application that
provides a collaborative writing space for group participants and captures audio, video, and log data
to monitor their interactions. One of its key features is the real-time analysis of collaboration quality,
using data from these sources to estimate and visualize group dynamics [
CoTrack was developed through DBR methodology involving three iterations with more than 50
in-service teachers over 5 years [
The teacher-AI complementarity in the third case is realized through the collaborative partnership between teachers and CoTrack, which supports teachers by providing real-time insights into students’ collaborative processes, analyzing complex multimodal data (audio, video, and logs) to estimate collaboration quality and identify potential issues. These AI-driven analytics enable teachers to focus on interpreting the data and making informed decisions about when and how to intervene efectively. At the same time, teachers bring their contextual expertise to the table, tailoring interventions to the unique needs of their students and classroom dynamics. This collaboration ensures that while CoTrack provides objective and data-driven monitoring, the teacher retains the essential role of adapting actions to foster meaningful learning experiences.
The three cases presented demonstrate how AI tools can complement the teacher’s role by influencing their skills, knowledge, attitudes, and tasks in distinct ways. The first case highlights the importance of integrating domain models enhancing teachers’ content and domain knowledge and their ability to design flexible learning trajectories. The second case, through the Kool-Kit suite, emphasizes the development of teachers’ pedagogical and technological skills, knowledge, and capacity to create and implement H5P-based activities, based on the ICAP framework. The third case, with CoTrack, showcases how AI can support teachers in monitoring complex collaborative processes, enhancing their situational awareness and decision-making skills in real-time classroom settings. Collectively, these cases demonstrate how teacher-AI collaboration can redefine instructional tasks by automating routine elements, ofering evidence-based insights, and enabling teachers to engage in higher-order activities, such as interpreting data within the framework of pedagogical concepts, adapting learning designs, and providing support to students. This highlights the potential of teacher-AI complementarity to shape teachers’ professional learning but also enriching the learning experiences for students.
This research was supported by the Estonian Ministry of Education and Research under the research project ÕLHAR1: ”Flexible Learning Paths for Supporting Learner-Centered Learning in Schools”, and by the Estonian Research Council’s Personal Research Grant (PRG) under grant number PRG1634