While teachers are considered to have a low risk of being automated (Frey and Osborne, 2017), augmenting teachers' pedagogical abilities (An et al., 2020) seems like a more likely scenario than complete automation of teaching. These technologies, potentially shaping the future of the teaching profession, are evolving rapidly and include, e.g., learning analytics applications (Viberg et al., 2020), intelligent tutoring systems (Mousavinasab et al., 2021), adaptive learning technologies (Molenaar et al., 2019), and educational chatbots (Winkler et al., 2020).

The division of control between the teacher and technology varies between different technologies. Molenaar (2021) has presented the six levels of automation of personalized learning to demonstrate the fluctuating nature of control between the teacher and technology: the control may be entirely on the teacher (level one) or technology (level six) or flexibly shared between the teacher and technology (levels from two to five). For example, on level four, the teacher is not expected to monitor the learning activities continuously, since technology is expected to signal when teacher control is needed (Molenaar, 2021).

Selwyn (2019) envisions two different scenarios regarding teaching augmentation. In the first scenario, technology frees up teachers to engage in meaningful acts of leading, arranging, explaining, and inspiring while technology takes care of routines and duties (Selwyn, 2019). In the second, more pessimistic scenario, teachers end up losing their autonomy while fulfilling the expectations of the technology (e.g., encouraging students to write in ways favored by automatic grading systems) (Selwyn, 2019). There is already some evidence of development toward this scenario: In a UK survey carried out during the COVID-19 pandemic, it was found that higher education teachers thought that their pedagogical practice had been “reduced to the fulfilment of rudimentary technical functions” (Watermeyer et al., 2021).

In both scenarios, teachers will increasingly be involved in designing digital environments and making meaning of the data students produce (Lodge et al., 2018). Pedagogical knowledge is required to define what type of data is needed and what is valuable to analyze to support learning (Viberg et al., 2018). Moreover, while some aspects of the learning process are more straightforward to quantify and analyze than others, using specific learning technologies may enforce particular pedagogies (Williamson et al., 2020). It is crucial that teachers are engaged in the development of teaching augmentation technologies and that these systems are studied in real-world contexts (Holstein and Aleven, 2022). Arantes and Buchanan (2022) suggest that the teachers who are active in the co-development of learning technologies should be recognized as educational data advocates, experts who understand the implications of using data-driven educational technologies.

Development of learning skills such as self-regulated learning (SRL) and self-directed learning (SDL) have globally become important aims of education (European Commission and Directorate-General for Education, Youth, Sport and Culture, 2019; OECD, 2019; ILO, 2021). Both concepts describe the learner's active role in setting goals, choosing learning strategies, and evaluating learning outcomes and are sometimes used interchangeably—however, SDL describes the more general approach to learning, whereas SRL focuses on the learning processes in a certain context (Gandomkar and Sandars, 2018).

In a traditional classroom setting, the teacher can support learners' self-regulation and then gradually decrease the amount of support, promoting the learning of self-regulation skills. One of the long-term goals for learning technologies is to take up the role of supporting and promoting students' SRL. However, in the current state-of-art learning technologies, the primary focus is on cognitive aspects of learning (i.e., student's knowledge) instead of a broader view of the student, including emotion, motivation, and self-regulation (Molenaar, 2021).

It has been suggested that learning technologies can support students' SRL indirectly (i.e., nudges, prompts) and directly (i.e., targeted suggestions specific to the student and the lesson) (Lodge et al., 2018) and adjust external regulation based on insights gained from data (Molenaar et al., 2019). Developing such a technology requires collecting and analyzing data about a learner's self-regulated learning behavior (Lim et al., 2021; Fan et al., 2022a,b) and using the generated insights to provide personalized feedback for the learner (Pardo et al., 2019, 2022). Currently, such systems are at their preliminary stages, and most solutions rely on a human teacher who is in control (Molenaar, 2021).

As the sophistication of such systems grows over time, they may affect how we see the role of a teacher. However, it is important to note that the effects on the teaching profession are not predetermined. As Selwyn (2019) points out, integration of any technology into society should be approached as a choice: “it is crucial that we consider the possibility of alternative technological pathways and different digital futures for education”. Therefore discussing these topics within the community of education practitioners is extremely important.

Understanding teachers' beliefs is important, as beliefs are related to their teaching practices and student outcomes (Fives and Buehl, 2012). Generally, teachers' perspectives toward automation of teaching have been critical (Tondeur et al., 2013). One reason might be, that novel technologies may challenge teachers' intuitive epistemic theories, and they need to rethink what learning is and how it should be promoted (Lammassaari et al., 2022).

Optimally, technology use should evolve simultaneously with novel learning and teaching practices (Hakkarainen, 2009). However, the rapid uptake of learning technologies during COVID-19 pandemic seems to have caused a situation where educational technologies are partly redefining and reducing the concepts of teaching and learning (Teräs et al., 2020). Teachers may be ill-prepared to face this situation, resulting in (perhaps healthy) distrust with learning technologies. Nazaretsky et al. (2022) argue that the key to gaining a teacher's trust is teacher agency: technologies should not restrict teachers to follow specific pedagogical scenarios and should allow teachers modify or override recommendations given by the technology. Moreover, Luckin et al. (2022) suggest AI readiness training for educators to be better equipped in leveraging AI to the benefit of learners. Finally, Lammassaari et al. (2022) found that if teachers' epistemic theory was in harmony with the digital reforms, there is a positive association with work engagement and negative association with burnout.

With the massive increase in the use of digital tools in teaching and learning during the pandemic, it is essential to investigate the implications to educational arrangements in general and the teaching profession in particular. Based on the literature (see, e.g., González-Calatayud et al., 2021; Molenaar, 2021), machines can handle routine teacher tasks such as grading. However, supporting learners' self-regulation has been an exclusively human endeavor and central to the teaching profession. As digital systems get more sophisticated, it is essential to broadly discuss a machine's possibilities and limitations to guide, encourage and support students and promote their self-regulated learning skills. This kind of technology potentially has far-reaching effects on teacher's role, workload, and profession in general. So far, there has been little discussion about the topic among practitioners, and little is known about practitioners' views on the topic.

In April 2021, an opinion piece by a Finnish elementary school teacher was published in the largest daily newspaper in Finland (Luoto, 2021). In the opinion piece, the author describes how learning new content is increasingly guided by adaptive learning technologies instead of a human teacher and how this kind of shift would free teachers' time to focus on tasks where humans are best. The piece sparked a lively discussion among Finnish education practitioners in several social media platforms. While such discussion spontaneously occurred, we decided to collect the material and analyze the discussion in depth.

In the present study, we analyze social media posts commenting on the scenario where some teacher tasks are controlled by technology instead of the teacher. Our research questions are the following:

In Finland, all schools shifted to remote learning because of the COVID-19 pandemic for 8 weeks during spring 2020. Afterward, schools have altered between contact teaching and remote learning depending on the development of the epidemic in each region.

During the pandemic, there have been active discussions in Finnish media about learning with digital technologies. One contribution to this discussion is an opinion piece (Luoto, 2021) written by a Helsinki-based elementary school teacher and published on April 19th, 2021 by Helsingin Sanomat, the largest daily newspaper in Finland. The author suggested introducing weekly remote learning, where elementary school pupils would study remotely 1 day a week, leveraging adaptive learning technologies. He describes how this shift would free teachers' time to focus on things where humans are best, such as contextual interaction.

The piece generated lots of comments on social media, especially in two Facebook groups: a general forum for discussion among Finnish education professionals (ca. 17,500 members) and a group specializing in the teaching profession (ca. 13,000 members). Comments were posted during April 2021.

We gathered social media posts related to the opinion piece from a Finnish education professionals group (N = 48) and a teaching profession group (N = 81) on Facebook, Twitter (N = 12), and a newspaper comment board (N = 10). Typically one person started the discussion by posting a link to the opinion piece, and others commented on the discussion thread. We collected all comments posted by the end of April 2021. We removed posts that did not include any argumentation (e.g., posts with only one word or emoticon) from the dataset, ending up with 114 posts. In addition to the post content, we collected the amount of social media reactions (i.e., likes) for each post. The original language of the opinion piece, as well as all the comments, is Finnish. We translated and paraphrased the excerpts presented here.

Townsend and Wallace (2016) suggest that when using data from social media, the researcher should consider the terms and conditions of the platforms, social media users' possible vulnerability, their expectation to be observed by strangers, subject matter sensitivity, and anonymization of research outputs. In the ethical considerations of the current study, we addressed these themes. First, we checked that the platforms used allowed their data to be used for research and obtained necessary approvals (e.g., Twitter Academic Account). Second, we considered the users in question (education practitioners) not particularly vulnerable. Third, since the forums of the discussion were either really large or completely open, we considered that users should reasonably expect to be observed by strangers. Fourth, we did not consider educational practitioners views and opinions about adaptive learning technologies as sensitive data. Finally, we anonymized all the data and paraphrased all the excerpts used in this research output.

To answer our first research question, we conducted thematic and epistemic network analyses on social media posts by Finnish education professionals. In thematic analysis, we found six themes: Educational arrangements (found in 77 posts), human presence and interaction (44 posts), technological capability (43 posts), teacher workload and efficiency (42 posts), pupil diversity and equality (28 posts) and learning skills (27 posts). The number of posts, including each code, and the sum of those posts' likes are presented in Table 2. Out of 114 posts analyzed, 32 (28%) were supportive of the proposed scenario, whereas 30 (26%) were ambivalent and 52 (46%) critical. While this presents the perspectives of professionals taking part in the discussion, the sum of social media likes shows the perspectives of the larger community. Out of 1251 likes, 103 (8%) were connected with posts showing a supportive perspective, whereas 192 (15%) were connected with posts showing ambivalent and 956 (76%) with posts showing a critical perspective.

The epistemic network analysis graphs are presented in Figure 1. The most central code was Educational arrangements (ARR), having strong connections with all other codes. When comparing the networks of supportive and critical posts, it is clear that different themes are highlighted. In the supportive network, the focus is on connections with Technological capability (TEC), and in the critical network on connections with Human presence and interaction (PRE).

Moreover, comparing what connections are missing in each graph is interesting. In the supportive network, the connection between Pupil diversity and equality (DIV) and Human presence and interaction (PRE) is missing, as well as the connection between Technological capability (TEC) and Human presence and interaction (PRE). Respectively in the critical network, the connection between Technological capability (TEC) and Teacher workload and efficiency (WOR) is missing, as well as the connection between Technological capability (TEC) and Learning skills (SKI).

To answer our second research question, we conducted qualitative content analysis on posts about learning skills. Learning skills were recognized as a theme in 27 social media posts, of which 14 were critical (390 likes), 10 supportive (36 likes), and 3 ambivalent (123 likes). The theme of learning skills was connected with both educational arrangements and technological capability in the supportive posts, but only with the theme of educational arrangements in the critical and the ambivalent posts.

In many comments, it was emphasized that pupils had limited learning skills and that supporting self-regulation remotely is challenging:

Some of the critical comments directly addressed the shortcomings of technology in guiding and supporting pupils:

Other comments did not mention the technology but emphasized that guiding and supporting should be carried out by the teacher:

On the other hand, there were many comments which described how technology would be able to support pupils :

As described in the Introduction, a skilled teacher gradually decreases the amount of support for learners to learn to regulate themselves. It is interesting whether this kind of fine-grained adaptive regulation support is expected from a machine. One comment discussed developing students' learning skills. Still, it was left a bit unclear whether this development is seen as a side effect of individual studying with technology or as a consequence of scaffolding by either teacher or technology:

In another comment, it was suggested that pupils could practice self-regulatory skills with technology, but there should be a human backup:

A recurring theme in the posts was pupil diversity. Some pupils were seen as capable of studying with learning technologies while others were not:

In some comments, the professionals suggested that technology could help by allowing teachers to focus on the pupils who need support:

In summary, in the supportive posts, technology was seen as capable of guiding and supporting students' self-regulation. On the other hand, the critical posts mostly, rather than criticizing the technology, emphasized the role of the teacher in supporting self-regulation. Overall, the role of the teacher is seen as necessary even with adaptive learning technologies available.

To answer our third research question, we conducted qualitative content analysis specifically on the posts about considering teachers' workload. Teacher's workload was recognized as a theme in 42 social media posts, of which 17 were critical (309 likes), 13 ambivalent (42 likes), and 12 supportive (47 likes). The workload theme was connected with educational arrangements, technological capability, and human presence in the supportive and ambivalent posts. In the critical posts, the workload theme connected with educational arrangements and human presence themes.

Especially the word efficient in the original opinion piece (“Teacher's working hours could be used more efficiently”) sparked many comments. There was some confusion about how the presented scenario would increase efficiency and whether that is even a meaningful goal:

In many posts it was highlighted that teachers are currently overloaded, and working with learning technologies was seen as additional work that would require additional resources:

Some of the comments about technology increasing workload were related to negative experiences with current systems rather than future possibilities of learning technologies:

On the other hand, in many posts, technology was seen as increasing efficiency:

In some posts, technology was seen as an enabler for more meaningful interactions in the classroom:

In summary, the views on adaptive learning technologies' effects on workload are polarized. Some professionals are already overloaded and fear that the new technologies will increase workload rather than decrease it. On the other hand, some professionals see that the latest technologies could enable more meaningful classroom interaction by automating routine work.

In line with previous work in different industries (Brougham and Haar, 2018), teachers were not concerned about being replaced by novel technologies. The general sentiment toward adaptive learning technologies was critical: 76% of social media reactions (i.e. “likes”) were connected with posts showing a critical attitude. The themes found in the discussion mostly replicated the themes of the original opinion piece. However, the most prevalent themes differed between critical and supportive posts.

In the critical posts, the professionals emphasized the role of the teacher's presence in facing student diversity and varying levels of self-directedness and self-regulation. This emphasis on human presence or “human touch” is in line with previous work on employees' perceptions among different industries (Bhargava et al., 2021). The theme of technological capability was mentioned in only 29% of the critical posts, whereas in the supportive posts, it was one of the main themes (66% of posts). In a sense, the critique on learning with adaptive learning technologies focused more on whether it “should be done” than whether it “could be done”. On the other hand, in the supportive posts, the theme of technological capability was connected to the themes of teacher workload and self-directed or self-regulated learning. These connections were absent in the critical argumentation, and we will discuss this further in the following.

The views regarding the capability of technology to support students' self-regulated learning varied. In some supportive posts, technology was seen as capable of guiding students, e.g., providing direct feedback. However, in many posts, this vision was recognized, but there was skepticism about the actual capability of technology to guide students. Overall, the professionals admitted the possibilities of technology in guiding and supporting students to some extent, but technological scaffolding alone was not considered sufficient. Compared to six levels of automation of personalized learning (Molenaar, 2021), most comments about technological capability considered automation on levels two or three. Moreover, the possibilities of learning analytics were mentioned only once, and none of the posts mentioned an option that technology could signal the teacher when teacher control is needed.

The levels of automation model (Molenaar, 2021) is based on the idea that some teacher tasks are automatized and others controlled by the teacher. However, this kind of thinking was largely absent in the posts by education professionals. Instead, the focus was on differences between students: which students need a human teacher and which students can learn with adaptive learning technologies. In many posts, there was an idea that part of the group could study independently with adaptive learning technologies, and the teacher could focus on the pupils who need individual guidance and support. This kind of teacher-led grouping of students based on their learning skills could be a first step toward the vision of fluctuating SRL support between human teachers and technology.

The views on adaptive learning technologies' impact on teachers' workload were mixed. Comparing the epistemic networks, the connection between technological capability and teacher workload is relatively strong in the positive network and non-existing in the critical network. The comments where technology was seen as increasing efficiency and decreasing workload were mostly about automatic assessment and provision of exercises. Additionally, one comment mentioned learning analytics as a way to increase efficiency. The remarks on technology increasing workload were related either to dissatisfaction with current information systems or the effort to design remote teaching.

When compared to Selwyn (2019) two scenarios of how technology may change the teaching profession, many posts, as well as the original opinion piece, resemble Selwyn's first scenario: technology frees up teachers' time for more meaningful activities. Again, these comments emphasized the differences between students. While some students can study with adaptive learning technologies, the teacher can focus on the students who need the human teacher most. The fear about teachers ending up fulfilling expectations of technology, as in Selwyn's second scenario, was not found in any of the comments. This may be related to Finnish teachers' high autonomy (Niemi et al., 2018)—if a teacher does not see the value of a particular technological tool, they will most likely stop using it.

The present study has some limitations. First, the sample size is small, with only 114 social media posts. This is related to the method of data collection: we collected all the posts in the discussions that spontaneously occurred in the social media after the publication of the opinion piece. Spontaneous discussions considering such a specific topic are scarce, and other data collection methods would be needed to obtain a larger sample size.

Second is the question of representativeness. While we do not have any background data about the social media users, we cannot tell how well the views presented in the analyzed posts represent the views of Finnish education practitioners. For example, we do not objectively know the amount of teaching experience the users had—if any. However, the contents of the posts certainly reflected personal experiences from a school environment. Moreover, it is possible that the strongest opinions were loudest in the discussion. There is a well-accepted view that social media platforms act as echo chambers that encourage the adoption of more extreme ideological positions, although the empirical evidence is still inconclusive (Kitchens et al., 2020). On the other hand, the forums in question were relatively large and general by nature, instead of groups of only like-minded people.

Third limitation is related to the authors' role in interpreting the qualitative data. The backgrounds of the authors influenced on both the creation of themes as well as the coding of individual posts. To increase the objectivity of our analysis, coding was carried out separately by both authors, and inter-rater reliability metrics for different codes were calculated. The agreement was lowest in the themes of pupil diversity, technological capability and educational arrangements. Some repeated disagreements were whether the use of video conferencing tools was about educational arrangements or technological capability, and whether posts mentioning differences in learning skills should be coded with both SKI and DIV codes.

In future work, alternative data gathering methods such as surveys and interviews should be considered. Furthermore, future work should also target possible differences in views, e.g., between different professional groups, geographical areas, teaching subjects, gender, and age, a few to mention.

Based on our findings, the discussion about adaptive learning technologies is polarized and includes many misconceptions. The supportive and critical argumentation are focused on different themes, limiting the opportunities for constructive dialogue. To enhance this dialogue, supporters of adaptive learning technologies could focus more on the effects of technologies regarding pupil diversity and equality. On the other hand, the critique of adaptive learning technologies could be broadened by analyzing the limitations in the technological capabilities of such systems.

Furthermore, when adaptive learning technologies are used, it is essential to collect and analyze the experiences of teachers and pupils. Do the teachers feel that the technology allows them to focus on more meaningful things, or do they feel like being servants of the technology like in Selwyn's second scenario? Do the pupils find the methods of studying meaningful? This is especially important should the technologies be more responsible for supporting and promoting pupils' self-regulation. Technological capability is an important but only one of many aspects to consider when using adaptive learning technologies.