1. Introduction
The use of Artificial Intelligence (AI) in education is rapidly evolving. Yet, there is a need for studies on the use of AI in education while teachers and students try out different AIs for different purposes. Attending a conference in 2024, we heard a scholar stating that “AI, being just 18 months old, is a new thing”. He is not unique in thinking this and in relating AI in education to OpenAI’s ChatGPT 4 (Generative Pre-training Transformer), which became publicly available in March 2023. However, various forms of generative AI (genAI) have been available for years.
Teachers typically exhibit one of three stances when it comes to embracing new technology: some resist incorporating it into their teaching, others exhibit enthusiasm and a drive to discover viable applications for the innovation, and a third group assume a cautious ‘wait-and-see’ approach (Elstad, 2016). Possibly, a balanced approach is fruitful (Cao & Dede, 2023). In Norway, the context of the current study, there is no unified approach to AI in education. Accountability can become murky if feedback is fully automated, and there is always the risk of surveillance or data privacy issues if General Data Protection Regulation (GDPR) standards are not strictly adhered to. Some municipalities, regions, and universities in Norway have banned AI, while others have enthusiastically embraced its use with clear guidelines for students and technologically safe environments (GDPR). Thus, the present study is based on an innovative approach in which an integrated AI solution reads all the data from the students’ work using five different AI plug-ins and four different knowledge bases for the optimization of feedback to students. The AI tools were used in tandem at 50 secondary schools to provide guidance and feedback to the students during their work on assignments. The present article reports the findings from the qualitative part of the study, in which five of the schools participated. The purpose of the qualitative part of the research was to understand in what ways AI can enhance or hinder formative classroom assessments at five of the participating schools. With this background in mind, two research questions were posed:
How do students perceive the use of AI for formative feedback practices?
How do teachers experience formative feedback practices when they use AI?
Theoretical Framework
One of the most important ingredients for successful formative assessment is formative feedback (Hattie & Timperley, 2007; Shute, 2008). This is feedback that informs the learner about the quality of a process or product, how the process or product can improve, and how to go about improving the process or product (Black & Wiliam, 1998, 2009). Formative feedback is a powerful part of formative assessment (Hattie & Timperley, 2007). It is feedback that is received during a learning process, not after. Feedback loops have been suggested (Hounsell et al., 2008) so that students need to act on the feedback through support and supervision. However, certain hindrances to successful formative feedback have been identified. These hindrances include time and resources, for example, teachers’ workload and a lack of assessment literacy (Brooks et al., 2019; Wiliam, 2011), making high-quality individualized formative feedback a challenge for teachers (Hopfenbeck et al., 2023).
The use of AI technologies in formative assessment can have several advantages. It can make the assessment more personalized and provide immediate feedback to students. However, challenges include the fact that AI sometimes provides less accurate assessment since AI does not know the students, that teachers and students may lack AI literacy—understood as a lack of understanding and competency (Ng et al., 2021)—that teachers could feel that their professional expertise is being sidelined, and that accountability may be black-boxed in the sense that no humans may take responsibility for assessments, and there may also be an issue of surveillance (Swiecki et al., 2022) if GDPR is not addressed. One may also question whether AI can assess holistically rather than assess a limited amount of knowledge (Swiecki et al., 2022).
Overviews of research on educational AI have taken several directions. Ma et al. (2014) wrote a meta-analysis of intelligent tutoring systems (ITSs), measuring their effect sizes regarding student outcomes. The use of ITSs led to significantly greater student outcomes across 107 effect sizes involving 14,321 participants, independent of education level and subject. Shute and Rahimi (2017) reviewed a computer-based assessment for learning in primary and secondary education. The main findings were that computer-based assessment generally enhances learning across subject areas, makes assessment more efficient and personalized, and clarifies misconceptions early in the process of learning. Knight et al. (2020) used AcaWriter, a learning analytics tool using AI, to provide automated feedback. Students were more likely to improve drafts, and AI provided additional writing support.
Farrokhnia et al. (2023) analyzed the opportunities and threats that ChatGPT brings to the field of education through a SWOT analysis (strengths, weaknesses, opportunities, and threats). Among the opportunities, they mention an increased accessibility of information, personalized learning through personalized support and feedback to students, and the facilitation of complex learning by challenging students at their aptitude level to develop critical arguments and by decreasing teachers’ workload. The threats they mention are AI’s lack of understanding of the educational context, the risk to security and of cheating, the perpetuation of existing biases in education, and a possible decline in higher-order thinking skills such as creativity and problem-solving due to the simplification of answers given to students. Lee (2023) examined the use of AI-generated feedback in writing in higher education. By scoring 2500 essays, the study showed that AI-generated feedback makes assessment more effective and that the students find the feedback meaningful.
Abdelghani et al. (2024) developed an experiment using a GPT-3-powered chatbot. They used prompt-based methods to train GPT-3 to ask closed and open-ended curious questions, which were tested in an experiment with Year 4 elementary school children. The researchers found that large language models can be relevant tools to support children in generating more curious questions. Albadarin et al. (2024) conducted a systematic review of 14 empirical studies in which teachers and students use AI (limited to ChatGPT) in various educational settings. The findings revealed that students have used ChatGPT to receive instant feedback, explain complex topics, enhance writing skills, and translate and summarize texts. The review also indicates that the overuse of ChatGPT may negatively impact collaboration and innovation. Teachers have used ChatGPT to make lesson plans and quizzes, helping them become more effective when it comes to both materials and time. Most of the reviewed studies recommend structured guidelines, training, and support.
The only relevant study to our knowledge, published in the same context as our study, is the one by Elstad and Eriksen (2024). They contend that AI is used in varying degrees in schools in Norway and that there is no unified approach to generative AI in Norwegian education. Elstad and Eriksen (2024) examined how 236 teachers at five high schools in Oslo experienced the use of AI during early implementation (6–8 weeks after) through a survey. One main finding was that individual factors have greater influence than organizational ones when it comes to the utility of AI. Moreover, they found positive relationships between professional communities (discussions with colleagues) and teacher efficacy.
2. Materials and Methods
2.1. The Tool Used in This Study
The AI-powered educational platform used in this study has the potential to enhance learning by providing personalized feedback and leverages four key knowledge bases: the general GPT model, teacher planning, student work, and expert fine-tuning. This integrated approach bridges the gap between theory and practice and may foster deeper understanding, critical thinking, and meaningful engagement among students. In the following sections, the four key knowledge bases will be explained.
2.1.1. General Knowledge Base (GPT Model)
The feedback system is built on AI models like GPT (Generative Pre-training Transformer), which provide a broad base of general knowledge across various subjects. The AI can answer questions, offer explanations, and provide initial feedback to students. For instance, a student working on a writing project can receive suggestions on grammar, structure, and storytelling techniques. The AI’s ability to tap into this vast knowledge base helps students refine their work and think critically. A custom version of Azure OpenAI, which includes four other tools from Azure Cognitive Services, is used to interpret images and other document formats.
2.1.2. Teachers’ Planning and Expertise
Teachers’ subject expertise and curriculum planning shape how an AI interacts with students. In the AI-powered educational platform used in this study, teachers input their learning goals and key concepts, allowing the AI to tailor its feedback to these objectives. For example, in a science lesson about ecosystems, a teacher can outline key concepts like biodiversity, and the AI will generate feedback and questions directly aligned with these topics, ensuring that students receive guidance that supports their learning goals. The systems use predefined prompts that include data for the users of the platform. The teacher has some agency in being able to define certain factors in the prompt, including which curriculum goals are applicable, and which themes and concepts are especially important. The prompt in all cases is derived from a predefined and thoroughly tested template that can only be modified by the teacher in specific and limited ways.
2.1.3. Student Work and Progress
The AI can also analyze students’ work in real time, offering personalized feedback based on each student’s current progress. Tools such as Idealab allow students to brainstorm and organize ideas, and the AI can track their thought processes, suggesting new approaches or deeper exploration. This dynamic interaction ensures that feedback evolves as the student progresses, supporting continuous learning and reflection.
2.1.4. Expert Fine-Tuning
The final knowledge base is expert fine-tuning. Teachers and educational experts can review AI-generated feedback and refine it to suit classroom dynamics or individual needs. For instance, in a history class, using a collective interaction tool, teachers can adjust the AI’s feedback to emphasize critical thinking about historical events, guiding students toward a deeper analysis of cause and effect rather than merely recalling facts.
By combining these four knowledge bases—AI knowledge, teacher planning, student work, and expert fine-tuning—the AI in the present study delivers feedback that is both personalized and aligned with educational objectives. The AI supports teachers by offering real-time insights based on student performance, encouraging critical thinking and a deeper understanding of concepts.
2.2. Sample, Data Collection, and Data Analysis
The sample consisted of 26 students and 13 teachers at five secondary schools. Focus-group interviews, one-to-one interviews, and unstructured observations were used to collect data. The focus-group interviews with students had 2–5 students in each group with the exception of one interview in which one of the students was ill and the other student had to do the interview alone. The student interviews lasted up to 20 min. The interviews with teachers were mainly one-to-one interviews with the exception of two interviews in which two teachers attended due to practicalities. The teacher interviews lasted up to 30 min.
The first interview was conducted collectively, i.e., including three interviewers where two of them acted as observers. This was carried out to increase the reliability of the interviews by modeling and sharing the use of the interview guides (see Appendix A and Appendix B). This also gave us the opportunity to debrief after the interview. The rest of the interviews were conducted by one interviewer and with no observers. All the interviews were conducted and recorded in separate quiet rooms. The interviews were transcribed and printed out during the phase of analysis.
In order to be more specific rather than general regarding the use of AI, the students were asked to focus on the use of AI in one school subject during the interviews. Table 1 shows the distribution of the students at the five different schools.
The 13 teachers have varied teaching experience and educational backgrounds (see Table 2). Each teacher was also observed during classroom teaching 1–3 times. The observations were unsystematic and part of a larger school development project. However, they provided important input on how teachers work when using AI to provide formative feedback to students. The purpose of the unsystematic observations, conducted by several members of the development project, was to see how the implementation of AI in classroom teaching and assessment progresses. The purpose was not to collect data for the current study (hence, there was no observation protocol), but the project members who had been in the classrooms were eager to share their experiences related to the assessment and feedback part relevant to this study. In other words, the unstructured observations can be regarded as secondary sources of informal data (i.e., field notes) used to inform the primary source of data (the interviews).
Due to the large amount of interview transcripts, NVivo was used to systematize them into meaningful units. The interview transcripts were analyzed using the constant comparative method of analysis (Strauss & Corbin, 1998). The three authors of this article read through each transcript a couple of times to get a sense of it, before coming together and coding each meaningful unit in NVivo. Following an open coding process, the codes were grouped into sub-categories. Some sub-categories were collapsed into larger sub-categories by comparing and contrasting different parts of the interviews. Finally, the main categories were agreed upon. For the student interviews, the main categories were effective time management, feedback, lack of AI use, and writing improvement. For the teacher interviews, the main categories were content improvement, planning and organizing, and student improvement.
3. Results
The first part of the Results Section provides student responses to answer the first research question in this article: “How do students perceive the use of AI for formative feedback practices?” The response categories are shown in Figure 1 below.
This section is divided into two: the results from the students and the results from the teachers.
3.1. Effective Time Management
The students generally agreed that the use of AI had led to more effective formative feedback practices. They reported that they could ask AI for feedback as much as they wanted and whenever they wanted: “You may ask for feedback as many times you want on what you have updated. Then you avoid waiting for the teacher to come and read through the whole notebook, which can take time”. Comparing the situation with AI with receiving feedback from the teacher was common: “It can take a week almost if he [the teacher] is going to correct everything…”. In the classroom, “You have to wait for a long time when raising your hand, since others need help too”. According to the students, more effective time management had two spin-offs: more time to spend on projects and more feedback.
3.2. Feedback
This main category was divided into four sub-categories: “AI-generated feedback from the teacher”, “subject-related feedback from AI”, “potential for improvement of feedback from AI”, and “positive feedback from AI”.
3.2.1. AI-Generated Feedback from the Teacher
One form of formative feedback was AI-generated feedback that teachers reviewed, edited when necessary, and forwarded to the students. The students had also received direct feedback from AI, and some of them recognized the way the feedback was provided: “It [the feedback] resembled a lot what AI would have said…”. One strategy the students used to distinguish between AI-generated feedback and teacher-generated feedback was to compare feedback given to several students. The AI-generated feedback was very similar in tone, wording, and structure. The AI-generated feedback was described as “easily explained”, and the students agreed that it should be used “As long as he [the teacher] look at it [the feedback] and agrees…”. The idea that the teacher should have the last word was also put forward by the teachers, which we will return to later.
3.2.2. Subject-Related Feedback from AI
Regarding subject-related feedback, AI helped students to produce better content by, for example, suggesting new words and concepts, asking for clarifications, and providing elaborations of content, including adding pictures. The students agreed that the subject-related feedback was useful and concrete, and that the AI was “very clever”. One main difference between AI and teacher feedback, according to the high-performing students, was that AI continues to provide formative feedback on the content even when the student is satisfied: “It never ends when it comes to what you can add”; “I may think I’m very satisfied with this, but then it [AI] comes and in a way says that I should go deeper into this…”; “It’s [AI] never satisfied, it’s [the student product] never good enough”.
3.2.3. Potential for Improvement of Feedback from AI
However, the use of AI was not unproblematic. Potential improvements, as pointed out by the students, concerned “sometimes irrelevant information [from AI] that didn’t have anything to do with the task”, AI sometimes providing different feedback on the same text, sometimes being too slow, sometimes not providing enough examples of how to improve, and sometimes being too unspecific. There were also examples of feedback that students thought was too hard or too long.
3.2.4. Positive Feedback from AI
Nonetheless, most of the formative feedback practices were perceived positively by the students. They thought the AI provided good, helpful, useful, “more fun”, and motivating feedback. One student said “… I was very happy that someone [AI] motivated me, so that you kind of didn’t feel stupid when you sit there alone. Cause you can kind of not talk to anyone else. […] You can, like, not always have sit and have the teacher next to you all the time during an assessment and ask all the time”.
3.3. Lack of AI Use
The five participating schools were part of a larger development project in which AI is used in teaching and assessment at 50 schools. Despite most of the students at the five participating schools having tried to use AI in formative feedback practices, we found examples where AI was seldom used or the students were not aware of its use: “We have not used it [AI] much”, “We almost only use OneNote”, “It may actually be that we haven’t used it [AI]”, and “I don’t remember much of it [the AI providing feedback]” are some examples from the interview materials.
3.4. Writing Improvement
Several examples of AI use revolved around student writing. AI gave examples of how to edit texts and make additions to improve the quality of the content—“When I wrote about non-renewable energy sources, it [AI] wanted me to write, like, about climate footprints etc.”—and to make texts more exciting by, for example, adding pictures.
The second part of the Results Section provides teacher responses to answer the second research question in this article: “How do teachers experience formative feedback practices when they use AI?” The response categories are visualized in Figure 2 below.
3.5. Content Improvement
This main category was divided into two sub-categories, “AI improving writing” and “subject-related use of AI”, which correspond to the categories based on student interviews.
3.5.1. AI Improving Writing
The teachers mentioned high-performing students who struggle with writing: “And then it [AI] has given them a ‘Oh yes, of course, I can think in that direction, yes’ […] maybe you can come up with some concrete examples you know about […] ‘Oh, I’ve only defined the term’ [re-iterating what her students say when using AI]”. In German (a foreign language subject in Norway), “… you can insert the curriculum objectives about grammar and language structures… so there were lots of concrete feedback to the students, ‘I see that you have written… try writing in the perfect past tense’… very concrete and very good”. The teachers expressed that AI did the “boring job with nit-picking, such as punctuation […] and then we can, like, focus on the big picture and help [students] with content”. In general, teachers reported that students wrote more lengthy texts: “Some [students] managed to write three pages in an hour, who would usually write half a page”. In tandem with what the students said, teachers also found that AI formative feedback was sometimes formulated using “a bit big words for some [students] to understand what to do”.
3.5.2. Subject-Related Use of AI
In one of the foreign language subjects, students who could not write “My name is…” started asking the teacher “How can I edit my text, how can I write about my hobbies?”, underlining the fact that AI formative feedback pushed students and motivated them to write more. The feedback was more at the sentence and word levels in language subjects, which leaves time for the teachers to focus more on content. Sometimes, teachers had to explain words that may have been too complicated for the students, for example, “You have to explain the ideology” in social science, where students did not understand “ideology”.
Another aspect of AI use in the subjects mentioned by the teachers was the possibility for students with learning disabilities, for example, dyslexia, to listen to the feedback. Students could choose between listening to and/or reading the AI feedback. According to the teachers, AI also helped them to see which students needed more feedback, including subject-related challenges. The threshold for asking AI for formative feedback was described by the teachers as being very low. Everyone could ask as much as they wanted in order to improve their performance in the subjects. This helped teachers to differentiate their feedback: “The most high performing students can be challenged by it [AI] and I can help a bit the most low performing ones”.
However, the teachers unanimously agreed that no matter how effective, concrete, and good AI is in providing formative feedback, it can never substitute the teacher: “AI is good, it provides very good feedback, but that human way of thinking… there I think the teacher should be present”. In sum, they believed that they help provide better formative feedback, since “I don’t do this alone; there are two of us [teacher and AI] looking at the student texts”.
3.6. Planning and Organizing
This main category was divided into three sub-categories: “effective use of time”, “familiarity with AI”, and “periodic planning for use of AI”. The first sub-category corresponds to effective time management in relation to the student responses reported earlier.
3.6.1. Effective Use of Time
The teachers expressed that they saved time by using AI to generate feedback for the students. One of them said “This round tour in the classroom, you know, which one does routinely… it becomes less of a coincidence. When you [as a student] discover that you are done and you are actually waiting for an answer”, emphasizing that AI provides feedback whenever students feel the need for it. Many other teachers repeated the advantage of immediate and rapid feedback when AI was used to provide formative feedback to the students. In addition, they mentioned that AI can help students in their writing process when they are stuck (quality), and they can receive more feedback (quantity) whenever they want to. Specifically, the “big” subjects (often referring to Norwegian, Mathematics, and English) were mentioned by the teachers when it came to the effectiveness of using AI. In these subjects, teachers normally have lots of text to read and provide feedback on. Some teachers used the time they saved to spend more time moving around in the classroom and to be more available to the students.
3.6.2. Familiarity with AI
Part of conducting, planning, and organizing the use of AI for formative feedback purposes is the extent to which teachers and students are familiar with AI. Some teachers reported that their students were not even used to personal computers (PC), despite schools in Norway being thoroughly digitalized: “I thought they possessed more digital skills and knowledge; cause they have had iPads since elementary school and even earlier, so in a way it’s their first time using PC in Year 8”. In the beginning, students asked “Is there a robot coming?”, and teachers explained “No, it is built into the PC”. Other teachers expressed that their students did not need any introduction: “It was like self-taught”. The teachers themselves appreciated the introduction to AI tools and expressed the need to use them over time. One teacher said “We should not be like a dinosaur, totally different than the life they actually live outside of school”, emphasizing that she likes to try out new methods and technologies. In general, the interviews indicate great variation when it comes to familiarity with AI.
3.6.3. Periodic Planning for Use of AI
The efficient use of AI for formative feedback purposes presumes that teachers plan the assessments for a whole school term (the fall term is from August to December and the spring term is from January to June). As was succinctly put by one of the teachers, “I have to be more specific when it comes to tasks I give [the students] during the whole period”. Some find this challenging, while others cope well with planning ahead: “It [AI] provides feedback a bit on behalf of me, since I have done the pre-work first, with curriculum objectives, concepts, periodic planning… I think this is important… but I see that the result in the other end is much more fruitful both for them [students] and me”. In addition, good periodic planning promotes cross-curricular activities: “… much neater to make cross-curricular tasks. Relate Norwegian and social science and maybe religion together. All in one task is possible, there will be fewer assessments for the students, at the same time as they view the world in a bigger picture”. Periodic planning has, moreover, helped teachers create more diverse tasks and see the curriculum from a wider perspective, as opposed to earlier, when they “… followed a chapter—then assessment—then new chapter”.
3.7. Student Perspective
The teachers’ responses in the interviews reflected positive student experiences with AI. They described the formative feedback that students received as “good”, “alright”, “fun”, and flexible in the sense that students could ask for more feedback whenever they wanted and whenever they were ready. “You almost think that it’s a bit scary how detailed and specific the feedback [from AI] is. I think that’s very very fun” is a quote that indicates the element of surprise that was present when teachers evaluated the quality of AI formative feedback. Moreover, they said that it was easier to use when the feedback was continuously provided in small portions during students’ work with their texts.
Finally, the unsystematic classroom observations corroborate the data from the interviews to a great degree. It is also worth mentioning that Schools 4 and 5 have used AI for a few months longer than the other schools, and also in more subjects and with more students. The teachers at these schools seemed to be more active and engaged. The leadership at Schools 1 and 3 did not follow up the teachers’ use of AI well, thus resulting in a weak sense of ownership of the AI trials.
4. Discussion and Conclusions
In this study, the following two research questions were posed: “How do students perceive the use of AI for formative feedback practices?” and “How do teachers experience formative feedback practices when they use AI?”
The students in the present study were generally satisfied with the AI-generated feedback. The findings indicate that AI has the potential to significantly enhance the quality of feedback in educational settings. It is noteworthy that the students referred to the feedback from teachers as “correcting”, whereas AI was said to provide “feedback”. By providing personalized, real-time feedback, AI bridged the gap between student needs and teacher capacity, particularly when workloads were high (Brooks et al., 2019). Combining tools like the ones used in the present study led to the utilization of AI to offer tailored guidance based on four key knowledge sources: general AI knowledge, teacher planning, student work, and expert fine-tuning. This multi-layered approach helps to create feedback that is specific, relevant, and aligned with curriculum goals (Abdelghani et al., 2024; Farrokhnia et al., 2023; Knight et al., 2020).
However, AI may pose some challenges. One of the primary concerns is the potential for less accurate assessments if AI tools misinterpret student inputs or fail to fully grasp the educational context. For instance, AI-generated feedback might be overly generic or miss out on the deeper cognitive challenges a student is facing. Some students in this study found the AI-generated feedback to be too general or complicated. This might be due to inadequate prompts given by the students. Nonetheless, AI helped the high-performing students who struggled with writing with ideas that simplified the writing process. Additionally, AI showed a surprisingly good capacity in providing subject-specific feedback to the students.
One of the key advantages of AI in feedback, as indicated in this study, is its ability to save time (Abdelghani et al., 2024; Hopfenbeck et al., 2023). Teachers often struggle with balancing their instructional duties and providing individualized feedback. With AI, much of the time-consuming task of reviewing student work can be automated, giving teachers more time to focus on tasks that require human interaction—like building relationships, offering emotional support, and fostering classroom discussion. For example, the AI tools in this study offered feedback that encouraged students to explore different angles or delve deeper into their work, while the teacher could focus on more nuanced, student-specific issues. This is where the human touch in teaching becomes irreplaceable: AI can analyze data, but only a teacher can fully understand the emotional, social, and motivational contexts of each student.
Research indicates that a lack of AI literacy among teachers and students can hinder the effective use of AI. Teachers may feel that their professional expertise is being sidelined (Swiecki et al., 2022). In this study, teachers agreed that they needed to have the last word, despite the students regarding the teacher as the one “correcting” versus regarding the AI as the one “providing feedback” and despite the teachers’ positivity toward the quality of the feedback provided by AI. Teacher education and support structures must adapt to equip educators with the necessary competencies to harness AI effectively. The adoption of generative AI by high school teachers can be a multifaceted process involving several stages, from initial awareness to full integration into instructional practices. This is an avenue for us to conduct further research (Elstad & Eriksen, 2024). Furthermore, periodic planning proved to be crucial for the teachers in the present study. If teachers do not manage to plan their assessments for a whole term at a time, they might miss out on the benefits of AI-generated feedback. Finally, we believe teachers and school leaders should spend less time and energy on cheating aspects and, rather, should use AI-generated feedback, with the teacher as the subject expert, and evaluate its efficiency like the teachers did in our study.
To maximize the benefits of AI in educational feedback, further research is needed in several areas. One key area is the development of AI systems that better understand individual student contexts and needs, making feedback more personal and accurate. Another area of focus should be teacher education: empowering educators with the skills to integrate AI effectively into their feedback practices without feeling replaced, including ethical and emotional aspects. Finally, research should examine the refinement of AI algorithms to enhance their ability to assess higher-order thinking, creativity, and problem-solving—skills that are crucial for students’ long-term success.
In conclusion, while AI holds great promise in streamlining subject-specific feedback and enhancing learning, it should complement rather than replace human interaction. This study showed several interesting ways in which AI complements teachers’ assessments when it is used to generate formative feedback for students. The study was limited to five schools, and the primary source was interviews. The classroom observations had the limitation of being unstructured and unsystematic. In a follow-up study, it would be interesting to use an observation protocol based on the interview categories to investigate the interactions between students and AI-generated feedback, among the students whenever they interact based on AI-generated feedback, and between the teacher and the students when students interpret and use the AI-generated feedback. The goal should always be to free up teachers’ time for tasks that AI cannot perform—building relationships, offering mentorship, and fostering a rich, collaborative learning environment. By focusing on these strengths, AI can be a valuable tool for improving educational outcomes without diminishing the role of the teacher.
Conceptualization, all authors; methodology, Burner; software, Lindvig & Wærness; validation, all authors; formal analysis, all authors; investigation, Burner; resources, all authors; data curation, all authors; writing—original draft preparation, Burner; writing—review and editing, Burner & Lindvig; visualization, Burner; supervision, Burner; project administration, Burner; funding acquisition, Lindvig & Wærness. All authors have read and agreed to the published version of the manuscript.
The study was conducted in accordance with the guidelines by the Norwegian Agency for Shared Services in Education and Research, Approval Code: 598382 Approval Date: 16 November 2023.
Informed consent was obtained from all subjects involved in the study.
Data available in Nvivo.
No conflicts of interest.
Footnotes
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Overview of the students in this study.
| School 1 | Focus-group interview, 3 boys, Year 9 |
| School 2 | One-to-one interview, 1 girl, Year 13 |
| School 3 | Focus-group interview, 3 boys, Year 8 |
| School 4 | Focus-group interview, 1 boy and 2 girls, Year 9 |
| School 5 | Focus-group interview, 3 boys, Year 9 |
Overview of the teachers in the study.
| School 1 |
|
| School 2 |
|
| School 3 |
|
| School 4 |
|
| School 5 |
|
Appendix A. Interview Guide: Teachers
-
Background questions: gender, position, seniority, teaching level, subject(s)
-
Has your work with formative assessment changed after using AI? If yes, how?
-
Keywords:
-
Practical organization: use of tests, assessment free, grading, use of points, assessment at different aptitude levels, time usage, documenting assessment, use of LMS, whole-day tests
-
Quality work: the quality of student work, students’ development, relations between curriculum, subject and assessment, how do students showcase their competence, differentiation, student involvement, inclusion, feedback to all students that is useful for their further learning
-
Methodology: the AI platform—how is it used in teaching, students’ use of AI, teachers’ use of AI, periodic planning, the AI model with its plug-ins: advantages and challenges
Appendix B. Interview Guide: Students
-
Background questions: gender, class, subject, motivation
-
Describe the ways you receive feedback in the chosen subject.
-
How are assessments conducted?
-
Quality work: how does the student show his/her competency? Student involvement? Inclusion? Useful feedback?
-
Methodology: the AI platform—how is it used in teaching, students’ use of AI, teachers’ use of AI, periodic planning
-
Practical organization (if necessary): use of tests, grading, use of points, assessment at different aptitude levels, use of LMS, whole-day tests?
References
Abdelghani, R.; Wang, Y.-H.; Yuan, X.; Wang, T.; Lucas, P.; Sauzéon, H.; Oudeyer, P.-Y. GPT-3-driven pedagogical agents to train children’s curious question-asking skills. International Journal of Artificial Intelligence in Education; 2024; 34,
Albadarin, Y.; Saqr, M.; Pope, N.; Turkiainen, M. A systematic literature review of empirical research on ChatGPT in education. Discover Education; 2024; 3, 60. [DOI: https://dx.doi.org/10.1007/s44217-024-00138-2]
Black, P.; Wiliam, D. Assessment and classroom learning. Assessment in Education: Principles, Policy & Practice; 1998; 5,
Black, P.; Wiliam, D. Developing the theory of formative assessment. Educational Assessment, Evaluation and Accountability; 2009; 21,
Brooks, C.; Carroll, A.; Gillies, R. M.; Hattie, J. A matrix of feedback for learning. Australian Journal of Teacher Education; 2019; 44, pp. 14-32. [DOI: https://dx.doi.org/10.14221/ajte.2018v44n4.2]
Cao, L.; Dede, C. Navigating a world of generative AI: Suggestions for educators; Harvard Graduate School of Education: 2023; Available online: https://bpb-us-e1.wpmucdn.com/websites.harvard.edu/dist/a/108/files/2023/08/Cao_Dede_final_8.4.23.pdf (accessed on 1 January 2025).
Elstad, E. Why is there a wedge between the promise of educational technology and the experiences in a technology-rich Pioneer School?. Digital expectations and experiences in education; Elstad, E. Sense Publishers: 2016; pp. 77-96. [DOI: https://dx.doi.org/10.1007/978-94-6300-648-4_5]
Elstad, E.; Eriksen, H. High school teachers’ adoption of generative AI: Antecedents of instructional AI utility in the early stages of school-specific chatbot implementation. Nordic Journal of Comparative and International Education; 2024; 8,
Farrokhnia, M.; Banihashem, S. K.; Noroozi, O.; Wals, A. A SWOT analysis of ChatGPT: Implications for educational practice and research. Innovations in Education and Teaching International; 2023; 61,
Hattie, J.; Timperley, H. The power of feedback. Review of Educational Research; 2007; 77,
Hopfenbeck, T. N.; Zhang, Z.; Sun, S. Z.; Robertson, P.; McGrane, J. A. Challenges and opportunities for classroom-based formative assessment and AI: A perspective article. Frontiers in Education; 2023; 8, [DOI: https://dx.doi.org/10.3389/feduc.2023.1270700]
Hounsell, D.; McCune, V.; Hounsell, J.; Litjens, J. The quality of guidance and feedback to students. Higher Education Research & Development; 2008; 27,
Knight, S.; Shibani, A.; Abel, S.; Gibson, A.; Ryan, P.; Sutton, N.; Wight, R.; Lucas, C.; Sándor, Á.; Kitto, K.; Liu, M.; Vijay Mogarkar, R.; Buckingham Shum, S. AcaWriter: A learning analytics tool for formative feedback on academic writing. Journal of Writing Research; 2020; 12,
Lee, A. V. Y. Supporting students’ generation of feedback in large-scale online course with artificial intelligence-enabled evaluation. Studies in Educational Evaluation; 2023; 77, [DOI: https://dx.doi.org/10.1016/j.stueduc.2023.101250]
Ma, W.; Adesope, O. O.; Nesbit, J. C.; Liu, Q. Intelligent tutoring systems and learning outcomes: A meta-analysis. Journal of Educational Psychology; 2014; 106, pp. 901-918. [DOI: https://dx.doi.org/10.1037/a0037123]
Ng, D. T. K.; Leung, J. K. L.; Chu, S. K. W.; Qiao, M. S. Conceptualizing AI literacy: An exploratory review. Computers and Education: Artificial Intelligence; 2021; 2, [DOI: https://dx.doi.org/10.1016/j.caeai.2021.100041]
Shute, V. J. Focus on formative feedback. Review of Educational Research; 2008; 78,
Shute, V. J.; Rahimi, S. Review of computer-based assessment for learning in elementary and secondary education. Journal of Computer Assisted Learning; 2017; 33,
Strauss, A.; Corbin, J. Basics of qualitative research: Techniques and procedures for developing grounded theory; Sage Publications, Inc.: 1998.
Swiecki, Z.; Khosravi, H.; Chen, G.; Martinez-Maldonado, R.; Lodge, J. M.; Milligan, S.; Selwyn, N.; Gašević, D. Assessment in the age of artificial intelligence. Computers and. Education; 2022; 3, 100075. [DOI: https://dx.doi.org/10.1016/j.caeai.2022.100075]
Wiliam, D. What is assessment for learning?. Studies in Educational Evaluation; 2011; 37,
© 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.
Details
; Lindvig, Yngve 2 ; Wærness, Jarl Inge 2 1 Department of Languages and Literature Studies, University of South-Eastern Norway, 3045 Drammen, Norway
2 Learnlab, 0164 Oslo, Norway;