We live in an age characterized by disagreements about artificial intelligence and chatbotsʼ epistemological and ethical aspects. The debate occurs within several areas of society and is characterized by the uncertainty that affects our traditional understanding of knowledge, learning and education. That chatbot etc., will play a role in several areas in schools in the future, and most agree. However, there is still uncertainty about how this will play out more specifically in teaching, assessment, and pupilsʼ learning in primary school. Additionally, of course, there are no simple answers to such complex questions when the area is, characterically, a “moving target”, where technology is developed at high speed. It is nevertheless essential to take a closer look at some of the knowledge base regarding the use of adaptive learning technology and artificial intelligence in school subjects from the last eight years in Norwegian schools. Because what happens when adaptive learning technology is woven in, as part of the many teaching aids, digital platforms and methods a teacher uses in mathematics teaching?

This topic is what we in the research group Digitale Læringsfelleskap (UiB) have addressed, and where we have already looked at some of the issues that have emerged in the wake of the great interest in ChatGPT. Especially when it comes to how artificial intelligence plays into studentsʼ daily learning work, inside and outside of school. Through three sub-studies, we have looked at one of the adaptive learning tools used in mathematics in Norwegian primary schools. This is developed on the basis of Knewton technology platform and their algorithm-based, adaptive learning tools based on artificial intelligence, machine learning and learning analytics (and used by over 40 million learners and students worldwide). In the first two studies, the focus is more generally on technology implementation and digital platforms, while in the third study, we look more specifically at this adaptive learning tool.

The backdrop for our first study was a literature review on learning analytics and adaptive learning tools (Krumsvik & Røkenes 2016), and where it was noted that little had been done in this area in primary schools in this country. So, in the study, “Everyday Digital Schooling – implementing tablets in Norwegian primary Schools” (Krumsvik et al. 2018), we investigated, among other things, the implementation of tablets from 2014 onwards, and teachersʼ complex everyday teaching with a variety of analogue, digital, and adaptive learning tools in mathematics, based on various types of register data (national tests etc.) and qualitative observational data. It was found, for example, that for boys in third grade, there were significant negative effects in mathematics, while for boys in fifth grade, there were significant positive effects in mathematics. However, since the study had an overall perspective, we cannot say whether this is due to the adaptive learning technology, teachersʼ general teaching, other teaching aids, or other conditions. Nevertheless, the observational data from the qualitative part showed that this adaptive learning tool provided several new opportunities for students to practice factual knowledge, skills, basic conceptual understanding and rote learning in mathematics.

Our second study, “Implementing Tablets in Norwegian Primary Schools. Examining Outcome Measures in the Second Cohort” (Krumsvik et al. 2021), shows, for example, significant positive effects for English for boys in fifth grade, but mixed results in mathematics for both girls and boys. There were also no unequivocal findings regarding the educational role adaptive learning technology had across the primary and secondary schools in mathematics. However, one result from the qualitative part of this study was that the way adaptive learning technology was used in mathematics was linked to a teacherʼs digital competence and the tool provided new opportunities for rote learning in mathematics in the home arena.

In our third study, “Glimpses Into Real-Life Introduction of Adaptive Learning” (Moltudal et al., 2020), we investigated more specifically the adaptive learning tool and homework in collaboration with teachers (grades 5-7) in mathematics. Fifteen minutes of homework with adaptive learning tools each day, or a total of 60 minutes per week, were used to free up time for practical mathematics and in-depth learning with the teacher at school. The studentsʼ level of competence, learning, motivation and basic psychological needs were measured quantitatively before and after the four-week intervention, and the intervention was followed up with basic data collection. The findings in the study show that adaptive learning tools could help ensure quantity training and rote learning in mathematics in homework, thereby freeing up time for practical mathematics and in-depth learning with the teacher in the teaching itself. However, the study indicates an interwoven connection between learning, motivation and rote learning, as well as some pitfalls teachers should be aware of in their classroom management. The study shows that using adaptive learning technologies for homework at the intermediate level contributed significantly positively to studentsʼ basic learning in mathematics (ES= 0.39, P = 0.001). However, the studentsʼ self-reporting show a discrepancy between their perceived learning outcomes (subjective learning outcomes) and their actual learning outcomes (objective learning outcomes) from the intervention. In other words, the students thought they learned less, while they actually learned more during this intervention period. The study also shows the importance of a teacherʼs class management when adaptive teaching aids are used by students across the school and the home arena (Krumsvik 2023). Here, as elsewhere in the school, it is digitally competent teachersʼ orchestration and broad repertoire that best safeguard the potential that adaptive learning technology can have for pupilsʼ learning and formative assessment processes in mathematics.

Across the three sub-studies, it can be seen that adaptive learning technology has an interesting, but for the time being, somewhat unfulfilled potential in mathematics, but also in pure research methodology. This adaptive learning technology can help overcome known reliability and validity problems, that exist in, for example, the self-reporting of homework, as it has been shown that there are many sources of error in this method. Internationally, there is a call for more attention to be paid to the quality of homework and the pupilsʼ effort when doing homework rather than self-reported time spent. For example, research findings (Rawson et al., 2017) show that time spent on tasks (the time spent on a homework assignment) is positively correlated with grades, and here, adaptive learning tools have contributed towards avoiding some of the methodological pitfalls that previous homework research has had to deal with. Hattie (2008) found an average effect size of d = 0.38 in his review of four meta-analyses, in which he examined 135 experimental studies on time spent on tasks. There is much evidence that the time spent on the tasks one does during homework is essential, and not the total time spent on homework, which through self-reporting has often been shown to include several activities other than the homework itself (Rawson et al., 2017). Roschelle et al. (2016) found in their study that 2,850 mathematics students who used adaptive learning tools and homework as central parts of the intervention group had higher scores in a standardized end-of-year assessment in mathematics than a control group who continued with traditional homework. Pupils with the lowest previous mathematics achievements benefited the most from the intervention.

Although there are many possibilities with artificial intelligence, such adaptive learning tools can also be a purely ethical “minefield”, and raise many ethical and privacy-related questions. Several researchers express that the philosopher Jeremy Bentham (1787) and Michel Foucaultʼs (1977) use of the term panopticon can be given a different meaning in such adaptive learning contexts. It can go beyond both ethics and privacy that we get access to so much information about students who use such adaptive learning tools, and therefore GDPR must “sit in the driverʼs seat” when artificial intelligence is implemented in schools.

In the future, it is essential to highlight the distinction between generic language technology tools such as ChatGPT and digital artefacts such as adaptive learning tools developed explicitly for school subjects in schools. The latter is an artefact designed intentionally and contextually regarding the curriculum, competence aims, teaching and assessment, while the former has yet to be. Such digital adaptive artefacts thus become “carriers” of insights from educators (the artefact is tuned to the schoolʼs context, curriculum, and competence goals), from lawyers (the artefact must safeguard privacy and GDPR), from computer scientists (the school learning tool is built on top of a generic technological platform), and from publishers (the adaptive learning tool has been developed together with a physical textbook in mathematics). It does not necessarily mean that they always work better than other digital tools – it is entirely up to the teacherʼs pedagogical repertoire and the pupilsʼ use of the artefact. From our studies, we can see that this is often linked to teachersʼ and studentsʼ digital competence, and one, therefore, will need theoretical “lenses” in the future to be able to understand what it is that constitutes digital competence in school and teacher education (Krumsvik 2012) when digital adaptive artefacts are used.