Periodic Reporting for period 3 - HHAIR (Hybrid Human-AI Regulation: Supporting Young Learners' Self-Regulated Learning)
Reporting period: 2024-02-01 to 2025-07-31
Summary of the context and overall objective of the project
The HHAIR project aims to develop a new adaptive learning system to help young learners develop self-regulated learning skills. Self-regulated learning is a critical skill set that allows learners to monitor and control their own learning. This combination of cognitive and metacognitive skills starts to develop when learners are in primary education, and research consistently shows that most learners need help to properly develop and apply these skills. Self-regulated learning skills support acquiring and storing new knowledge and facilitate lifelong learning.
In this project, we work on an adaptive learning system that combines human and artificial intelligence so that young learners receive the “right” level of support to develop self-regulated learning skills. In the Netherlands, young learners often use adaptive learning technologies to learn arithmetic, mathematics, grammar, and spelling. Depending on the learners ' performance, these systems use an algorithm to select the right exercise or problem. In a similar way, HHAIR develops an algorithm to support learners in how to monitor and control their own learning. This algorithm will work on top of existing adaptive learning technologies already used in schools, such as Gynzy, through a newly developed data infrastructure.
Agency over your own learning process is an important element of self-regulated learning. Therefore, HHAIR aims to transfer agency from the adaptive learning technology to the learner as these self-regulated learning skills develop. This means that with the development of these skills, decisions that are first made by adaptive learning technology are later made by the learner. The first step in this transfer is for learners to become aware of how they are learning. In the HHAIR project, we will develop personalized visualizations in the so-called learning path app that show learners their learning trajectories over time. The combination of awareness of how the adaptive learning technology is regulating your learning and the development of your own skills is expected to allow young learners to develop these important self-regulated learning skills. In this way, the project aims to optimize learning and support future and lifelong learning.
This innovative and unique project proposes a hybrid intelligence approach for training self-regulated learning skills with AI. Hybrid intelligence aims to research and develop intelligent systems that augment rather than replace human intelligence. These systems are developed to leverage human strengths and compensate for human weaknesses with AI. The Hybrid Human-AI Regulation developed in this project aims to apply this notion in the context of human learning to support self-regulated learning.
The HHAIR project aims to develop a new adaptive learning system to help young learners develop self-regulated learning skills. Self-regulated learning is a critical skill set that allows learners to monitor and control their own learning. This combination of cognitive and metacognitive skills starts to develop when learners are in primary education, and research consistently shows that most learners need help to properly develop and apply these skills. Self-regulated learning skills support acquiring and storing new knowledge and facilitate lifelong learning.
In this project, we work on an adaptive learning system that combines human and artificial intelligence so that young learners receive the “right” level of support to develop self-regulated learning skills. In the Netherlands, young learners often use adaptive learning technologies to learn arithmetic, mathematics, grammar, and spelling. Depending on the learners ' performance, these systems use an algorithm to select the right exercise or problem. In a similar way, HHAIR develops an algorithm to support learners in how to monitor and control their own learning. This algorithm will work on top of existing adaptive learning technologies already used in schools, such as Gynzy, through a newly developed data infrastructure.
Agency over your own learning process is an important element of self-regulated learning. Therefore, HHAIR aims to transfer agency from the adaptive learning technology to the learner as these self-regulated learning skills develop. This means that with the development of these skills, decisions that are first made by adaptive learning technology are later made by the learner. The first step in this transfer is for learners to become aware of how they are learning. In the HHAIR project, we will develop personalized visualizations in the so-called learning path app that show learners their learning trajectories over time. The combination of awareness of how the adaptive learning technology is regulating your learning and the development of your own skills is expected to allow young learners to develop these important self-regulated learning skills. In this way, the project aims to optimize learning and support future and lifelong learning.
This innovative and unique project proposes a hybrid intelligence approach for training self-regulated learning skills with AI. Hybrid intelligence aims to research and develop intelligent systems that augment rather than replace human intelligence. These systems are developed to leverage human strengths and compensate for human weaknesses with AI. The Hybrid Human-AI Regulation developed in this project aims to apply this notion in the context of human learning to support self-regulated learning.
Work performed and main results
The research, design and development of the HHAIR systems resolves around four scientific challenges:
i) identify individual learner’s self-regulated learning during learning;
ii) design degrees of hybrid human-AI regulation;
iii) short-term effects of HHAIR on deep learning; and
iv) long-term effects of HHAIR on SRL skills for future learning.
The first and second challenges have been the focus of our work during the first year and a half. The first challenge was finding an appropriate way to measure how learners regulate their learning in an adaptive learning technology. We developed a new algorithm to assess learners’ regulation support needs while they are learning in adaptive technology Gynzy. We have used individual learners' learning trajectories to train this algorithm and found nine different groups of learners. Using learning process and outcome data, we interpreted the groups, who are these students, and what do they need to improve their learning trajectory. Currently, we are in the process of determining how much data we need to accurately predict which group a learner belongs to so that we can start supporting the learner in the right way early during learning.
The second challenge is the design of the different degrees of hybrid human-AI regulation. Each degree consists of AI and human regulation. AI regulation consists of the decisions made by the adaptive learning technology, while human regulation concerns the decisions made by the learner. We have developed a co-regulation degree in which the learner can follow how the adaptive learning technology selects the different problems and how the learner performs over time. By showing this information, we help learners to better monitor their learning with personalized dashboards. We found that learners in this co-regulating degree improve the regulation of their practices and behaviour: they make more problems, solve more problems correctly, and show fewer complex trajectories over time. The effects on learning outcomes are not yet conclusive.
We continued to develop the shared-regulation degree in which learners are supported in monitoring and receiving some more control opportunities to enact agency. In this degree, learners can control the difficulty level of the problems they receive from the adaptive learning technology. We are now examining how learners make these control decisions and how these decisions interact with their knowledge, self-regulated learning skills, motivation, and perceptions of competencies. Our findings suggest a complex relationship between control enactment and learners’ motivation. Based on these insights, we are further advancing this shared-regulation degree.
The research, design and development of the HHAIR systems resolves around four scientific challenges:
i) identify individual learner’s self-regulated learning during learning;
ii) design degrees of hybrid human-AI regulation;
iii) short-term effects of HHAIR on deep learning; and
iv) long-term effects of HHAIR on SRL skills for future learning.
The first and second challenges have been the focus of our work during the first year and a half. The first challenge was finding an appropriate way to measure how learners regulate their learning in an adaptive learning technology. We developed a new algorithm to assess learners’ regulation support needs while they are learning in adaptive technology Gynzy. We have used individual learners' learning trajectories to train this algorithm and found nine different groups of learners. Using learning process and outcome data, we interpreted the groups, who are these students, and what do they need to improve their learning trajectory. Currently, we are in the process of determining how much data we need to accurately predict which group a learner belongs to so that we can start supporting the learner in the right way early during learning.
The second challenge is the design of the different degrees of hybrid human-AI regulation. Each degree consists of AI and human regulation. AI regulation consists of the decisions made by the adaptive learning technology, while human regulation concerns the decisions made by the learner. We have developed a co-regulation degree in which the learner can follow how the adaptive learning technology selects the different problems and how the learner performs over time. By showing this information, we help learners to better monitor their learning with personalized dashboards. We found that learners in this co-regulating degree improve the regulation of their practices and behaviour: they make more problems, solve more problems correctly, and show fewer complex trajectories over time. The effects on learning outcomes are not yet conclusive.
We continued to develop the shared-regulation degree in which learners are supported in monitoring and receiving some more control opportunities to enact agency. In this degree, learners can control the difficulty level of the problems they receive from the adaptive learning technology. We are now examining how learners make these control decisions and how these decisions interact with their knowledge, self-regulated learning skills, motivation, and perceptions of competencies. Our findings suggest a complex relationship between control enactment and learners’ motivation. Based on these insights, we are further advancing this shared-regulation degree.
Progress beyond the state of the art and expected results until the end of the project
We are currently designing and investigating the different degrees of Hybrid Human-AI regulations and implementing them into the learning path app that is connected to the adaptive learning technology Gynzy. Next to the co- and shared-regulation degrees, we are preparing for the AI-regulation degree. We found that some learners continue to struggle with their learning despite the adaptive support. In the AI-regulation degree, we aim to advance the adaptive learning technology algorithm with insights from our newly developed algorithms. In this way, they can adjust better to the needs of individual learners. Finally, we will design a self-regulated learning degree for learners who have acquired the skills to perform without AI support.
An important future challenge is to develop a novel algorithm to help us select the “most effective” degree of support for different learners that optimizes both learning as well as the development of self-regulated learning skills. We will perform numerous experimental studies to understand these two objectives' trade-offs and the best level of support. For this development, the remaining two design challenges are crucial: investigating immediate effects on learning in short-term field studies and self-regulated learning skills for future learning in long-term field studies. This will be done in existing and widely used adaptive learning technologies to investigate the effects within the regular school curriculum. The AI@EDU infrastructure is developed for this purpose to connect our hybrid human AI regulation solution to ALTs used daily in schools across the Netherlands.
In this way, the project will develop advanced measurements of self-regulated learning and novel algorithms to drive hybrid regulation for developing self-regulated learning skills while learning in adaptive learning technologies.
We are currently designing and investigating the different degrees of Hybrid Human-AI regulations and implementing them into the learning path app that is connected to the adaptive learning technology Gynzy. Next to the co- and shared-regulation degrees, we are preparing for the AI-regulation degree. We found that some learners continue to struggle with their learning despite the adaptive support. In the AI-regulation degree, we aim to advance the adaptive learning technology algorithm with insights from our newly developed algorithms. In this way, they can adjust better to the needs of individual learners. Finally, we will design a self-regulated learning degree for learners who have acquired the skills to perform without AI support.
An important future challenge is to develop a novel algorithm to help us select the “most effective” degree of support for different learners that optimizes both learning as well as the development of self-regulated learning skills. We will perform numerous experimental studies to understand these two objectives' trade-offs and the best level of support. For this development, the remaining two design challenges are crucial: investigating immediate effects on learning in short-term field studies and self-regulated learning skills for future learning in long-term field studies. This will be done in existing and widely used adaptive learning technologies to investigate the effects within the regular school curriculum. The AI@EDU infrastructure is developed for this purpose to connect our hybrid human AI regulation solution to ALTs used daily in schools across the Netherlands.
In this way, the project will develop advanced measurements of self-regulated learning and novel algorithms to drive hybrid regulation for developing self-regulated learning skills while learning in adaptive learning technologies.