1 Introduction

1.1 AI technologies for education, challenges, and ethical considerations

In recent years, Artificial Intelligence in Education (AIED) has made significant strides, with proponents asserting the potential for personalised learning experiences for students, increased efficiency for teachers, and automation of administrative tasks in schools (Wang et al., 2023). Adaptive learning systems and Intelligent tutoring systems (ITS) are often touted for providing tailored support and interactive learning experiences, ultimately enhancing overall learning outcomes (Mousavinasab et al., 2018). Platforms can integrate teachers' assessment standards, while AI agents assist in pre-service training with subject-specific scenarios and pedagogical tips tailored to future educators' proficiency levels (Liu, 2023). AI assistants can also facilitate group projects, assigning roles, guiding discussions, tracking progress, and promoting equal participation. Virtual facilitators based on Natural Language Processing (NLP) can be used to develop professional learning opportunities for teachers (Copur-Gencturk et al., 2024). Additionally, AI-powered chatbots, like Andy English Chatbot, are proving to be valuable tools in language learning, providing learners with a low-pressure environment to practice speaking and receive feedback without fear of judgment, promoting fluency and self-confidence (Fathi et al., 2024). AI-driven language learning platforms like Duolingo have demonstrated their significance in lifelong learning of a second language, supported by evidence from both written and oral proficiency assessments characterised by high validity and practicality measures (Sudina & Plonsky, 2024).

Albeit the potential benefits these technologies offer, concerns have surfaced regarding their potential drawbacks. For instance, Learning Analytics platforms can provide insights into student engagement, performance trends, and learning trajectories, with descriptive and predictive value. Nevertheless, the Council of Europe (2022) raises ethical concerns regarding the potential impact of these technologies on student and teacher agency. In fact, research suggests that interactions with automated artificial systems can diminish human operators' sense of control (Berberian et al., 2012; Obhi and Hall, 2011). Additionally, ethical considerations surround the collection of private and sensitive data such as mood analysis and activity logs concerning political views, ethnic identity, and sexual orientation (Tundrea, 2020).

NLP tools, while providing feedback on grammar, syntax, and coherence, may prioritise quantity over content quality and exhibit unreliability in error detection (Liu & Kunnan, 2016; Miao et al., 2021). Other AI tools for mathematics, like Photomath, offer opportunities for students to engage with mathematical concepts. However, they can also pose challenges in interpreting the information provided. This difficulty can stem from users' varying levels of prior mathematical knowledge (Gaona et al., 2022), potentially impacting their self-concept in relation to mathematics. Ensuring transparency and explainability in AI systems is key to uphold ethical principles in education. If educators and students cannot understand how these systems reach decisions, it can detach them from the learning experience.

Moreover, ITS have faced scrutiny concerning learners with diverse backgrounds or special needs (Alrakhawi et al., 2023). Indeed, there are apprehensions that these technologies may exacerbate existing educational disparities and expose students to bias and cultural barriers (Miao et al., 2021), impacting their learning processes and assessment. Pedagogical agents, for example, while aiming to enhance learning experiences by simulating human-like interactions and positive emotions, raise concerns about affective privacy and virtual relationships (Hudlicka, 2016). Ultimately, ethical challenges may encompass competing interests and extend beyond individual users to encompass their families, and discussions on these matters often reflect the perspectives of stakeholders, which can vary depending on their affiliations with private organisations, government agencies, research centres, universities, or schools (Popenici & Kerr, 2017).

Drawing on research, a comprehensive examination of AIED studies from 1970 to 2020 highlights the insufficient focus on ethics in the implementation of these technologies and the overlooking of viewpoints from teachers, parents, and students, who could provide valuable insights for ethical considerations (Bozkurt et al., 2021). This is consistent with the findings of a UNESCO report, which revealed that although 11 Member States have endorsed AI curricula, some either do not engage or engage only to a limited extent with AI ethics and its social implications (UNESCO, 2022).

Utilising Biesta's (2020) principles of subjectification, socialisation, and qualification, which are key to delineating educational endeavours, it becomes crucial to understand the educational objectives advocated by AIED and how they balance these principles. At a pedagogical level, certain AI applications may not align with socio-constructivist learning frameworks, resulting in a disjointed integration of technology into formal education settings. While some AIED resources are marketed as personalised learning tools, they often provide individualised pathways that ultimately converge toward a common goal, not considering the variance of the impact they may have on singularities and their ecosystems. Both scenarios can be argued to be a deliberate preference for mass education influenced by political and economic considerations, undermining the processes of subjectification and qualification. Are these AI-driven resources designed and used in a way that upholds the underlying foundation of qualifications, which includes self-efficacy? How can they be used to support the core properties of human agency: forethought, self-reactiveness and self-reflection (Bandura, 2023)? Regarding socialisation, a study by Hrastinski et al. (2019) highlights a crucial ethical concern surrounding AIED – its potential impact on socialisation within the classroom. The authors discussed concerns among teachers, researchers, and pedagogical developers related to shifts in the teacher's role and teacher-student relationships due to AI, largely emphasising the importance of teacher professional development for a relevant implementation of AIED in K-12 classrooms.

1.2 Ethical and meaningful pedagogical standards through continuing professional development

The significance of Continuing Professional Development (CPD) for teachers in enriching pedagogical practices is well-established (Abakah et al., 2022; Darling-Hammond et al., 2017; UNESCO, 2015). CPD tends to foster heightened expectations that indirectly influence student achievement (Rubie-Davies, 2015) and holistic development by enhancing teachers' capacity for utilising evidence-based pedagogical methods, designing intentional and meaningful learning experiences, and implementing relevant assessment techniques. It can lead to experiences of rupture with methodological normativity and reaffirm the inevitability of pedagogical mediation (Trindade, 2012). Teachers involved in CPD can also serve as role models for their students and peers, promoting the value of learning, knowledge sharing, and critical thinking within learning communities. It may also enhance teachers’ influence in decision-making processes, improving their decisional capital (Hargreaves & Fullan, 2012). Additionally, CPD may equip teachers with the ability to adapt to changing contexts, address the evolving needs of their students more confidently, meet societal expectations, and be accountable to the entire school community (Learning Forward, 2022). This seems particularly relevant for teachers considering technological advancements and their influence on students' lives.

Following an Ellulian perspective, CPD can serve as a catalyst for questioning the underlying mindset that governs our society and for the critical examination of the purposes and dynamics of technique, assisting educators in comprehending and navigating the intricate interplay between technology, pedagogy, and the broader societal context. Only an approach responsive to the challenges posed not only by AI technologies, but by a world driven by technique, can reveal how its dominance has permeated all layers of society and institutions, extending even into education. Such an approach can redirect discussions on educational ethics far beyond superficial debates centred solely on AI ethics.

Kowalczuk-Waledziak et al., (2019) also assert that, given the intersection of ethical considerations with pedagogical approaches, there is a need to reconsider teacher education. This could be done through a participatory approach involving educators, to make meaningful progress in pedagogical terms (Chichekian & Benteux, 2022). While UNESCO (2023a) recommends governmental agencies to build capacity for the proper use of Generative AI (GenAI) in education and research, a systematic literature review exploring the ethical dimensions of AIED technology implementation in education from 2011 to 2022 revealed a noticeable absence of emphasis on ethical considerations, despite some training projects being introduced in schools (Mouta et al., 2023a). However, more recently, some initiatives are currently underway to address these concerns and requirements. Set to launch in 2024, UNESCO's AI competency framework for educators takes a step further by outlining a three-stage progression that covers areas such as fostering a human-centric approach, addressing AI ethics, understanding AI fundamentals and applications, incorporating AI into pedagogy, and utilising AI for professional growth (UNESCO, 2023b). A training practical example comes from AI4T (2024), an Erasmus + K3 initiative designed by France, Slovenia, Italy, Ireland, and Luxembourg, which emerged to facilitate AI education specifically for teachers and school leaders. Understanding the ethical implications of employing AI in educational environments was included as a component of the shared learning objectives across all countries' professional development pathways.

1.3 Teacher agency

The dominance of a world of technique (the pre-eminence of rationality and effectiveness), the increasing diversity of students, the necessity to prepare learners for environmental transitions and to critically examine what development might mean, and the attempt to enhance learning and assessment through digital means greatly influence the direction of education. Thus, it becomes central to incorporate the needs of students and society into an andragogic narrative that specifically caters to teachers, recognising them as active contributors in co-authoring the educational plot. Intelligent professionalism suggests educators to be consistently involved in the processes that educational systems claim are enhancing education (Thompson, 2021). This is expected to come alongside the commitment to change within schools and among teachers (Kukulska-Hulme et al., 2020) and the recognition of families and community members as vital partners (European Union 2018). Being a teacher involves taking a public stance on important educational issues and public policy development, elevating the professional status across all systems (Thompson, 2021; UNESCO, 2021). Such an approach to professional development can strengthen teachers’ shared epistemic agency (Damşa, 2014) and reinforce ethical practices related to AIED knowledge. Moreover, it can promote teachers’ relational and collective agency, reassuring the structural context supports and facilitates the interdependence, while understanding how agency unfolds over time (Damşa, 2014). This perspective aligns more closely with an ecological approach advocated by Priestley et al. (2015), which suggests that teacher agency isn't an inherent attribute of individuals but rather a result of the interaction between personal capabilities and the environmental factors that influence their actions.

Ultimately, teacher agency is recognised as pivotal, as educators possess the ability to actively approve, adapt, or resist policies and programmes (Severance et al., 2016). While it is true that numerous decisions regarding educational technologies are often made from the top-down, involving government bodies, big tech companies, national research, education networks, and appointed staff from learning institutions, the ultimate realisation of these technological initiatives heavily relies on teachers themselves. Their awareness, knowledge, and, most importantly, their attitudes and level of engagement play a key role. To inform these aspects, training may be significant. Once educators are active agents in designing, implementing, evaluating, advocating for, and experiencing quality professional learning and the systems that support it (Cochran-Smith et al., 2022; Learning Forward, 2022), training initiatives built upon their knowledge and deliberation may yield meaningful investments for schools. Teachers are also expected to undergo training centred on the same principles of agency they are to instil in their students, enabling learners to exercise it when interacting with autonomous AI agents.

In this context, this paper presents a specific segment of a participatory research aimed at strengthening teachers' agency in training related to ethics and AIED. It is rooted in the theoretical underpinning of dialogic ethics, which acknowledges the multifaceted nature of ethical concerns and the need for a thorough examination involving various stakeholders (Santos, 2012). While existing research has explored the theoretical aspects of AIED ethics, it often overlooks the perspectives of educators themselves. This study distinguishes itself by focusing on educators' voices through the lens of Continuing Professional Development. Specifically, the gap in participatory studies in this field is addressed by utilising focus groups with those responsible for shaping teacher education.

2 Method

This component is part of a comprehensive Educational Design Research, which began with a systematic literature review (SLR) to pinpoint major gaps in AIED and ethics (Mouta et al., 2023a). The findings from the SLR guided subsequent research steps, including the recognition of the need for the involvement of stakeholders from diverse educational backgrounds in research and for capacity building in AIED and education ethics, especially among teachers. As a result, the Delphi method was employed to investigate educational experts’ ethical concerns regarding AI in education (Mouta et al., 2023b). With the goal of encouraging comprehensive debates on the development of professional learning in AIED, education ethics, and agency, these concerns informed the design of futures scenarios to facilitate in-depth focus groups discussions. Focus groups are commonly employed in educational research because of their usefulness for programme development (Nagle & Williams, 2013) and their ability to yield valuable data to shape policies and practices. This section delves into the initial inputs arising from those focus group discussions among professors, teachers, and trainers regarding the attributes of such a training. While the development of the professional learning materials is planned for later stages, this phase provides insights and discussion around the course’s objectives, implementation strategies, and design aspects. This study’s third stage is the main focus of the upcoming sections and is structured according to the framework proposed by Silva et al. (2014), which includes planning, preparation, moderation (referred to here as "implementation"), data analysis, and results communication.

2.1 Planning

The planning phase of focus group research is key to ensuring that clear objectives guide the moderation process.

2.1.1 Research guiding questions

To inform the design of a meaningful training programme for K-12 educators on AIED, ethics, and agency, this stage of the study delves into the following key research questions (RQ):

  • RQ 1: Which structuring dimensions do educators consider important for a training programme on AIED, ethics, and agency targeting K-12 pre-service and in-service teachers?

  • RQ 2: What objectives do educators prioritise for such a training programme?

  • RQ3: What specific aspects should be taken into account when developing a training programme in this field?

  • RQ 4: What strategies can be used to update teachers' knowledge and skills on AIED while also fostering agency and ethical thinking about its potential impact on education?

2.1.2 The moderation scripts

To facilitate participants' reflections and deepen their responses, it is recommended to use a moderation script with a "funnel" strategy, where questions become increasingly specific throughout the session (Borges & Santos, 2005). While the main objective of these focus groups was to identify the goals, topics, and delivery methods for a training programme on ethical AIED use for teachers, some initial questions on AIED ethics were included to familiarise participants with the topic under discussion. In advance of the session, each participant was provided with two hypothetical scenarios that had been generated during the Delphi phase of this research (Mouta et al., 2023b). The aim was to motivate the participants for the discussions by providing these narratives. It is worth noting that these scenarios resulted from a three-stage iterative process, leading to their creation in 2021 for discussions held in 2022. AI applications to the field of education have progressed significantly since then, as emphasised in the introduction. Despite this, the scenarios remain relevant for the purpose of this research. However, their dystopian aspects and the potential shift towards vignettes illustrating current experimental cases during the training will be investigated in the discussion section.

The scenarios presented to the participants were the same for each element within a group, but they differed between groups. The script utilised during the session featured open-ended questions on participant attitudes towards AIED, ethical concerns, student agency, and the design of an educator training programme within the context of AIED (see Appendix 1).

2.1.3 The participants recruitment

To prevent dispersion, a relatively uniform group was established, selecting participants according to criteria that corresponded to the research objectives (Silva et al., 2014). For this study, participants with professional certification for teaching and experience in the field of education at any level were eligible for selection. Additionally, they were required to hold responsibilities in K-12 teacher pre-service or in-service education. Participants also had to be either native Spanish speakers or highly proficient in the language. This requirement was based in the primary language of the research centre where the study was conducted, facilitating the initial recruitment of participants through convenience sampling due to time constraints at the start of the research phase. However, to ensure diversity in cultural perspectives, a subsequent snowball sampling approach was employed. This involved efforts to purposively recruit participants from various geographical regions, thereby improving the generalisability of the study's findings. This approach balanced the need for active communication in the research language with the goal of capturing a broad range of cultural insights and experiences related to the study objectives. Invitations were sent via email. A larger group than required was invited to tackle the challenge posed by potential refusals (Ogbeifun et al., 2016). Among those who declined the invitation, the explanation provided was either difficulty in participating due to professional commitments or a lack of confidence in engaging in discussions on such a topic. So, 19 participants from a pool of 40 invitations agreed to participate and were subsequently divided into four groups. Participants provided their informed consent, and the moderator explained the research objectives and ensured participants' positive responses (Silva & Fortunato, 2021). The mean age of the participants is 44 years old (SD 7.46) and 47.4% identify as women and 52.6% as men. The group comprises individuals from five countries, all of whom hold responsibilities in preparing pre-service or in-service teachers in the K-12 education sector. They possess pedagogical experience in either K-12 or higher education. Participants general sociodemographic and occupational profiles are presented in Table 1.

Table 1 Teachers’ sociodemographic profile
Full size table

In each group, there was an in-depth discussion of two scenarios selected from the set of hypothetical scenarios generated in the previous Delphi phase, making a total of eight vignettes analysed across the four groups. Hennink et al. (2019) proposed that the appropriate sample size to reach saturation in focus group research depends on the study purpose, code characteristics, group composition, desired type/degree of saturation. These focus groups usually consist of three to six sessions, with a larger number of sessions when accommodating a diverse study population (Coenen et al., 2012; Guest et al., 2016). Therefore, in this research code and meaning saturation were achieved through four discussions, aligning with the desired saturation levels for focus groups not stratified by demographic characteristics.

2.2 Preparation

Participants were notified through emails, which contained details about the research objectives, session duration, group specifics, focus group dynamics, session recording, and informed consent. The two hypothetical scenarios were dispatched 48 h prior to the session. Since the invitees originated from diverse regions and the world still faced pandemic restrictions, it was deemed more appropriate to conduct virtual meetings, through the Zoom platform. Despite some limitations, research has shown that virtual focus groups can be just as productive as in-person focus groups in generating rich and meaningful data (Krueger & Casey, 2015).

2.3 Implementation

Four focus groups were implemented in the first semester of 2022, with varying numbers of participants (6, 4, 5, and 4) to accommodate their availability.

The participants were unfamiliar with each other, except for two individuals in the first and in the second groups. In both cases, their professional roles were independent from each other. Across all sessions, there was a notable absence of dominant speakers. All participants tried to ensure everyone had an opportunity to share their thoughts, mitigating power differentials and ensuring that diverse viewpoints were thoroughly considered. The duration of the focus group discussions varied: the first session lasted 80 min, the second and fourth sessions lasted 64 min each, and the third session lasted 62 min.

2.4 Data analysis

In accordance with Roberts et al.'s (2019) proposal, the deductive phase of this study included creating a preliminary codebook, aiming to ensure credibility through three key elements: content (describing the sampling frame and creating research instruments), criterion-relatedness (testing research tools for consistency), and construct validity (ensuring inferences match the research question) (Long & Johnson, 2000). The subsequent sections detail the methodology, which includes iterative coding, category refinement, and reliability and validity testing.

All sessions were recorded using the recording option in Zoom, after obtaining email and verbal consent from all participants. The recordings facilitated the verbatim transcription of the session's content using an AI-powered software (Trint). All transcriptions were then thoroughly reviewed and cross-checked against the video recordings by the principal researcher. The analysis was based on the research question being asked ("How should a training programme targeting ethics in AIED for K-12 teachers or aspiring teachers be structured and organised?"), the initial analysis of the literature, the Delphi method approach undertaken as part of the project, and a preliminary scan of the raw data. Additionally, an inductive approach was employed, which allowed for the identification of any unexpected themes that emerged during the coding process. So, the preliminary data analysis entailed the researcher's close examination of the raw data, selectively marking segments for coding. The entire dataset, transcribed from audio recordings, was analysed using NVivo software. Moderator contributions were transcribed but not included in the analysis. The research team agreed on unit separation criteria upon importing the data into the software. A thematic approach was selected, taking into account the specific content associated with each element of the text. The structural matrix, employed for the coding of the data, was derived from both the extant literature and the aforementioned initial reading and identification of preliminary themes from the raw data. In cases where a significant proportion of the codable units could not be accurately captured by the existing codes, further codes were added to the analytical framework. This process was repeated multiple times until no new codes were identified, thereby confirming the matrix as a valid representation of the data. The identified codes were added to NVivo as nodes, and the coded text was matched to the nodes in a systematic way. As detailed in Fig. 1, the combination of deductive and inductive approaches enabled the study to identify patterns both within and outside of the predetermined codes, resulting in a comprehensive and nuanced analysis of the research question.

Fig. 1
figure 1

Process of code creation and testing. Note. The process of code creation and testing was adapted from Roberts, K., Dowell, A., & Nie, J. B. (2019) Attempting rigour and replicability in the thematic analysis of qualitative research data; a case study of codebook development. BMC Medical Research Methodology, 19, 66. https://doi.org/https://doi.org/10.1186/s12874-019-0707-y

Full size image

The final selection of nodes resulted from two reliability tests: a test–retest reliability and an inter-rater/coder analysis (cf. https://drive.google.com/drive/folders/13fF4h6fGbrwf9woDKApJ7NulfoeNzLEr?usp=drive_link), conducted by two researchers. The Pearson correlation coefficient between the initial coding and the coding performed three weeks later was 0.86, indicating high consistency in the coding process over time. The percentage agreement between the coders was 81%, demonstrating the extent to which they agreed on coding decisions. The Cohen’s Kappa score was 0.83, signifying substantial agreement beyond chance across different coders. This analysis allowed for the examination of information using a framework matrix consisting of four main categories with its own set of subcategories: attitudes towards AIED, benefits and opportunities, ethical challenges, teacher initial and CPD; plus, one category labelled as "other" was added to account for text segments that were tangential and not directly related to the ongoing discussion (cf. Table 2). The description of each node and respective examples from categorisation in NVivo is also available (see Appendix 2).

Table 2 NVivo framework matrix
Full size table

Instances arose where information exhibited overlapping characteristics, being assigned to multiple categories simultaneously, or left uncoded if it did not align with a specific category. For this analysis, categories encompassing all the gathered information were established, although only specifically studying the data directly relevant to the main research question. In fact, the initial open-ended questions were designed to prepare participants for the discussion. Since the topic under consideration is relatively new or often overlooked in current educational discourse, these questions aimed to provide the necessary context for discussing training methods. The need to code such information stems from the fact that coding ensures that no data is overlooked and that all related aspects are taken into account during the analysis process. However, for the purpose of this analysis, a deliberate decision was made to analyse and present the data that directly addresses teacher training in ethics in AIED.

2.5 Results communication

This section presents the key findings from the focus group discussions in relation to the research questions guiding this stage of the project.

2.5.1 RQ1: Structuring dimensions for the training programme

Concerning the findings on teacher initial and Continuing Professional Development, the relationships between the coded text excerpts and their corresponding nodes are examined. This analysis uses the data shown in Fig. 2, which identifies the topics participants discussed the most. This figure provides an overview of the percentage frequency of each category mentioned across the entire dataset, highlighting its prominence in the overall discussions.

Fig. 2
figure 2

Number of references per item

Full size image

These discussions were further categorised based on different structuring dimensions of the training course. By analysing the arrangement of the nodes, it is possible to identify the following key structuring dimensions: training aims and objectives, design considerations, and implementation strategies. Table 3 provides the percentage of discussion dedicated to each category within these specific planning dimensions, offering insights into their relative prominence among them. For instance, while “broadened ethical thinking” constitutes 31% of the overall discussion dataset (cf. Figure 2), it comprises 51.06% of the references associated with the “aims/objectives” category in Table 3. This “aims/objectives” category was indeed the primary focus of the participants' discussions, centring on the purpose of the training. With 60% of the discussions dedicated to appraising and explicitly articulating its objectives, it is clear that participants were committed to ensuring the course aligned with teachers’ needs.

Table 3 Optimising teacher training: Organising items based on planning dimensions
Full size table

To further analyse the relationships between these discussion topics, NVivo's cluster similarity function was employed. This function measures similarity between nodes, considering the presence or absence of shared words between each pair of nodes. The results of this cluster analysis are displayed as a horizontal dendrogram (cf. Figure 3). The following paragraphs will discuss categories and subcategories, as well as the connections between them, highlighting those revealed by the analysis.

Fig. 3
figure 3

Items clustered by coding similarity

Full size image

2.5.2 RQ2: Aims and objectives prioritised by educators

Broadened ethical thinking

The aims and objectives category accounted for a total of 47 references across five items. The emphasis on this category reveals a significant concern among the participants regarding the purpose and direction of an AIED ethics training course. Their main concern revolved around the necessity to foster broad ethical thinking; a topic extensively discussed in approximately 51% of the discourse segments. In this context, one of the participants expressed, "when I think about teacher training, I don't see it from a technicist point of view (…) [it] is the space for reflection and construction, joint construction about what the challenges, the difficulties, the opportunities might be”. This highlights the understanding that dealing with the ethical aspects of technological progress necessitates a holistic strategy that encompasses all disciplines. By cultivating a climate of ethical consciousness and accountability regarding technological advancements, individuals are given the opportunity to question the very ethical nature of the educational project.

Update of knowledge and skills

Moreover, by recognising the significance of keeping knowledge and skills up to date, as evidenced by the subcategory that garnered 17% of the references, participants suggested that training is expected to question what is technically known, ethically debatable, and required to create a theoretical and conceptual rationale. The following excerpt illustrates this perspective: “we must perhaps have a significant ability to develop competencies for knowledge, to know the advances of science and their relationship with the benefits and problems that ethical aspects in sciences and technologies can generate for us”. Participants also underscored the ever-changing nature of the field and the imperative for trainees to remain abreast of the latest advancements. In the context of CPD, offering training that aligns with teachers' current needs while also anticipating and addressing upcoming changes and advancements may help them stay ahead of the curve. This dimension may also support educators to respond to any emotional discomfort that may arise from global shifts affecting their learning environments, whether due to emerging resources, technologies, or pedagogical trends. It may also encourage them to explore alternative approaches, experiment with emerging resources and help students engage meaningfully with the learning experience, keeping teachers motivated and satisfied in the long run.

The cluster similarity analysis in Fig. 3 highlights a connection between these two subcategories: broadened ethical thinking and knowledge/skills update. This association underscores that extensive ethical reasoning in AIED leads to the recognition of the importance of updating knowledge and skills in this field. Conversely, a strong knowledge base is crucial for informing ethical deliberations accurately.

Questioning educational paradigms

As a result of this mindset and skills update, teachers will be instigated to question current educational paradigms, a subcategory that garnered 13% of the references. In this case, participants acknowledged the significance of motivating trainees to challenge the established norms and embrace a more comprehensive and ethical educational approach, facilitated by the dilemmas posed by this recent technological endeavour. As one of the participants stated, "It has to come accompanied (…) by a significant innovation (…) within the educational system. For example, it's about ending the concept of exams, the concept of a teacher, as we see it today".

New role requirements

Once teachers begin to challenge current educational paradigms, their role in the classroom will also be challenged, as stated 13% of the times and as the cluster similarity analysis underscores (cf. Figure 3). This is exemplified by an excerpt from one participant: “becoming guides means positioning ourselves horizontally in the processes of development, in learning, in the construction process, in existence”. So, training is expected to encourage teachers to re-evaluate the adequacy of their current roles and practices, critically question what it means to be a “guide” when integrating AI technologies, and understand data-driven insights in a holistic way.

Compliance with a policy framework

Finally, participants recommended preparing teachers to comply with a policy framework, which accounted for approximately 6% of the recommendations, as expressed in this discourse: “teachers should indeed receive training (…) on the current legislation and how to apply it in the context of using applications that collect personal information, such as (…) artificial intelligence systems”. This highlights their concern for achieving conformity once legal and regulatory guidelines are established. Furthermore, it underscores the significance of enhancing teachers' professional credibility and safeguarding students' rights by ensuring consistent practices across contexts.

2.5.3 RQ3: Design considerations for the training programme

The design category was the less mentioned, receiving a total of 11 mentions. However, the existence of six subcategories related to this topic suggests that the participants did differentiate their insights into the strategic and organisational aspects of the training programme.

Dealing with resistance and transition

In 36% of the discourses, the subcategory dealing with resistance and transition was mentioned. It suggests a concern about how teachers might respond to the integration of these technologies in learning environments, due to their cognitive and affective attitudes or lack of familiarity with the theme of AIED. It encompasses resistance to change and managing transitions comprehensively. As one participant emphasised: “I see what I call the transition risk, meaning, in the initial stages, the first teachers may not adopt these new technologies. There will be a transitional phase”. Therefore, it is proposed that the training tackles concerns related to a smooth and contextually meaningful transition. The dendrogram's analysis offers insights on this matter, indicating the design category's proximity to two implementation categories (cf. Figure 3). It suggests that addressing concerns within the design category may necessitate practical interventions to gain teachers' support. Hands-on strategies and innovative pedagogical methods aligned with a major purpose could prove instrumental in building teachers' confidence.

Collaboration between teachers and with families

The subcategory that pertains to the collaboration among teachers and between teachers and families was covered by 18% of the data, indicating a strong concern in engaging with key stakeholders who play a proximal role in shaping students' ethical understanding and behaviour, as highlighted by this participant: “But it is here where we, especially teachers, need to take a stand, collaborating closely with families to enhance the quality of students' learning”. This also recognises that these actors hold the power to either support or undermine an educational project.

Pointless without AI technologies

The subcategory that questions the relevance of such a training without AI technologies available at a school level was covered by 18% of the references, shedding light on the belief that ethics training in AIED holds significance when it is integrated with and informed by AI technology. Participants acknowledged that educational endeavours should be purposeful and aligned with the actual needs and practices of stakeholders, grounding ethics training in the available current technological landscape, as demonstrated by this participant: “It wouldn't make sense to do it too far ahead of the time when teachers will have the technologies to work with”. The cluster similarity analysis reveals an interconnection between this subcategory and the one that raises a thought-provoking question on setting priorities (cf. Figure 3).

Setting priorities

In 9% of the discourses, the importance of adding more training specifically focused on AI was questioned. Other shortcomings in teacher education related to inclusion and gender perspective were mentioned as priorities. The participant expressing this concern highlighted the following: “Are we prepared to take on more training, specifically in the field of AI, when we still have so much pending training related to inclusion, gender perspective, and many other areas in which we are (…) failing in teacher education?”. Once again, the dendrogram provides insights on this matter, particularly highlighting the farthest neighbour clustering between the categories of training's relevance without AI technology and the necessity for debates regarding new role requirements or educational paradigms (cf. Figure 3). According to this pragmatic approach, without setting priorities for training focuses and ensuring teachers have access to AIED, this discussion is spurious.

Pre-service training

9% of the references had also been made for equipping educators with the necessary knowledge and skills related to AIED ethics during their pre-service training, as demonstrated by this quote from one of the participants: “However, in many instances, the discussion of ethics needs to start from a more fundamental standpoint. This particularly applies to the generations now entering universities”. This integration has the potential to cultivate a culture of ethical consciousness and responsibility among future educators. Additionally, when teachers with different levels of experience and diverse knowledge exchange their perspectives in educational settings, positive outcomes can result.

Informal and self-directed continuing professional development

9% of the discourses have emphasised the importance of maintaining an informal and self-directed Continuing Professional Development (e.g., “And the positive awareness that comes with this type of tool, without the need for other types of courses.”), revealing the recognition that professional growth in ethical practices should extend beyond formal training programmes. This can be fostered through reflective practice, learning communities, diverse resources, and collaborative learning. Such subcategory and the one that pertains collaboration among educational actors appear connected in the cluster similarity analysis (cf. Figure 3). This suggests that teachers and families who actively engage in collaboration are more likely to be the ones that take ownership of their professional development and seek out opportunities for learning and growth independently. This proactive approach to professional learning and to forming alliances reflects a collective agency and commitment to improving educational outcomes.

Finally, the cluster similarity analysis reveals that the nodes "compliance with a policy framework" and "pre-service training" share some characteristics, despite being represented by different colours, indicating a less close association between them (cf. Figure 3). Nevertheless, this near neighbour clustering highlights the importance that some participants place on establishing a strong foundation for the integration of AIED, suggesting it be formalised through policy and incorporated into the curriculum for pre-service educators. This concern calls the attention to the efforts that are expected to accompany teacher professional learning and that relate to policy development and curriculum design.

2.5.4 RQ4: Implementation strategies

The implementation category received 20 references, indicating that the participants placed importance on the delivery methodologies of the training programme. These references likely stem from the participants' experiences as trainers, suggesting the value of hands-on, skill-based training and innovative pedagogical practices in the context of AIED ethics.

Innovative and purposeful pedagogical practices

The subcategory centred around innovative and purposeful pedagogical practices has received 45% of the references, indicating a strong desire for intentional and creative practices. In this regard, one of the participants stated:

“There, regarding the assessment aspect, there needs to be a modification. I still don't understand why teachers are reluctant to (…) all kinds of exploration, advancement, and the construction of students' knowledge with open-book assessments. For instance, presenting a case like the one you've just raised and, around it, initiating a discussion or debate. If knowledge and experiences are meant to be shared among everyone, to discuss common points, identify areas of differing thoughts, and establish consensus—which is the most challenging part. (…) I mean, the focus should be on the knowledge-building processes, in utilising what is being generated through artificial intelligence."

This passage illustrates the assumptions participants hold regarding the impact of traditional practices, particularly assessment, in education. To address this concern, they advocate for creativity coupled with purpose. Aligning intentionality with innovative practices reflects a deliberate desire to challenge existing paradigms of meaning-making in education and stimulate fresh approaches to thinking and action. From this viewpoint, participants propose that trainers not only consider how to assess students but also question whether the objectives and nature of what is being assessed remain static. These practices may initially disrupt conventional norms, but they hold the potential to spark meaningful shifts in the educational context, shifting from learning better to an overall experience of better learning.

Hands-on and skill-based training

The subcategory concerning hands-on and skill-based training received 40% of the references, indicating a strong emphasis on experiential training and the desire to provide opportunities for applying knowledge in real-world scenarios. This perspective is exemplified by a participant who stated: “I believe that courses for professionals who are currently working should be very practical. They should not be awareness courses”. As highlighted earlier, insights gleaned from the cluster similarity analysis, illustrated in Fig. 3, suggest a connection between these two subcategories and indicate that both approaches can address potential resistance and ease transitions by instilling confidence in participants regarding their existing level of awareness in their field.

Learning by example

The topic of learning through examples arose in approximately 15% of discussions, highlighting the significance of practical illustrations and real-life scenarios, through which participants can gain insights into ethical dilemmas and develop skills to navigate them proficiently, as demonstrated by this quote: “The only mechanisms that come to mind for training in that area would be to show examples of what not to do so that they can learn how to use them. (…) Teaching small cases tailored to the user and having them identify if they share that use and if they see potential risks in using it. And then, they are presented with half of the case and the consequences are observed”. This subcategory, as indicated in Fig. 3, is linked with the promotion of broadened ethical thinking and the update of knowledge and skills. It implies a recognition of the importance of examples, role-playing, and role-taking in stimulating perspective taking and the affective-cognitive dissonance that leads to the emergence of ethical dilemmas and to a deeper understanding of ethical decision-making processes. Furthermore, these examples serve as valuable learning tools by offering direction on which areas to prioritise for additional learning. This is especially important when considering the informed allocation of resources needed for continuing professional development.

3 Discussion

The findings of this research contribute to shaping the next steps in developing an AIED and education ethics training course for K-12 in-service and pre-service teacher programmes. This evolving framework aims to foster dialogic practices that allow teachers to reflect on the ethics of education in the presence of AI technologies. The discussion examines the findings concerning the research questions to guide the next steps in developing the training course. It integrates theoretical and conceptual elements from the current draft of the AI Competency Framework for Teachers' Development (UNESCO, 2023c), emphasising how the findings align with ongoing work by experts in this field. Additionally, it interprets the results through the lens of the ik-model domains (Mouta et al., 2015). The ik-model builds upon the TPACK framework by introducing diachronic dimensions to enhance the understanding of technological integration in learning environments. These dimensions focus on relationships and processes, adding depth to the technological and content domains. They provide a perspective on how relationships between different educational stakeholders are shaped and change, and how pedagogical processes are initiated, evolve, and adapt contextually to foster meaningful learning experiences. Additionally, participants' responses naturally encompassed the dimensions of the ik-model in their narratives, thereby demonstrating the model's relevance and applicability for discussing the results. The forthcoming discussion on each research question will utilise the findings to explore its practical implications for teacher training on these matters.

Regarding the first question, concerning the structuring dimensions of a training programme on AIED, ethics, and agency it is possible to acknowledge that participants expressed a nuanced perspective on what is expected to be implemented, far from a positivist approach to designing training experiences. In fact, teachers spontaneously covered all the ik-model’s domains as relevant parts to be considered in such a training: (a) technological domain – AIED techniques and technologies; (b) content domain – ethical thinking, educational paradigms, teachers and students’ roles, policies frameworks, best practices; (c) relational domain – roles and collaboration between key educational stakeholders; and (d) processes domain – dealing with resistance and transition, experiential practices, learning by example, self-directed learning. What is worth noting on establishing this parallel on the discussion, is that the relational and processes dimensions are often sidelined in training programmes on AIED for education, which predominantly focus on content knowledge and pedagogical content knowledge. The categories addressed within the processes domain also demonstrate participants' proficiency in andragogical principles, possibly stemming from their roles as teacher trainers, indicating a deep understanding of the dynamics involved in designing and facilitating adult learning.

On the second question, regarding objectives, participants covered the course technological and content domains. The first implication for preparing this course regarding its aims is the following:

  1. (1)

    Consider it an ethical decision in itself to discuss AIED frameworks as either supporting tools or constraints to meaningful practices and pertinent innovation.

This concern and recommendation regarding teacher competency align with the UNESCO draft, which proposes the inclusion of “ethics of AI” as an asset. This asset encompasses an understanding of fundamental ethical principles related to AI, as well as participation in communities for the iterative development of institutional and societal regulatory environments, bridging the gap from understanding to active engagement. In these focus groups, participants put emphasis on the importance of discussing AIED-oriented policy frameworks during training. The adoption of such frameworks is deemed suitable solely when enabling a comprehensive approach tailored to local interests and school microsystems. As these frameworks are typically developed through extensive research, consultations, and expert input, with the aim of reflecting societal values, legal requirements, and best practices in the field, they can provide reassurance to teachers during transition stages. For instance, the Beijing Consensus on Artificial Intelligence and Education (UNESCO, 2019), The Ethical Framework for AI in Education (The Institute for Ethical Al in Education, 2021), and the draft AI Competency Frameworks for school students and teachers (UNESCO, 2023c) can serve as valuable guidelines in this regard. These frameworks are scalable and allow for some contextualisation and progression, serving as an initial step in addressing the ethical challenges posed by AIED and ensuring its alignment with pedagogical principles.

On the one hand, the decision to follow these guidelines in the context of AIED can be seen as an ethical decision in itself. In fact, it can help reduce excessive reliance on singular experiences and personal biases. On the other hand, it may prompt considerations as to whether these frameworks might unintentionally impede meaningful innovation, creativity, or more dynamic and fully context-driven approaches. Nevertheless, exploring these frameworks may provide a basis for discussing a set of principles, considering the perspectives and interests of various stakeholders.

Regarding the third question about particular considerations for designing a training programme in this field, participants mainly reflected on the course processes domain. The second implication for preparing this training is as follows:

  1. (2)

    Provide opportunities for addressing teachers' concerns, needs, and uncertainty during transition processes throughout course's design and implementation phases.

The utopian and dystopian narratives surrounding AI, the current knowledge, and conditions for its integration, combined with past experiences of integrating digital technologies into learning environments, make it key to address the cognitive, affective, and behavioural aspects related to AI integration during training. In fact, participants underscored the necessity of assisting teachers in handling resistance and facilitating transitions to foster positive training effects. Optimising attention to change, uncertainty and transition processes can also be strategically used to measure the course's impact on educators' cognitive, affective, and behavioural dimensions of attitude.

Cukurova et al. (2023) devised a reliable instrument for assessing the comprehensive factors that impact teachers' adoption of adaptive learning platforms in educational settings. According to their study, it is imperative that these technologies do not impose an additional workload on teachers, requiring them to switch between different tools, or navigate various pedagogical practices during implementation, while ensuring the minimisation of ethical concerns. Certain frameworks concentrate on the emotional and cognitive phases that teachers undergo when faced with change, including endings, neutral zones, and new beginnings (Bridges, 1991; Hall & Hord, 1987). Frameworks like The Diffusion of Innovations Theory (Rogers, 2003) and Change Management Models highlight the importance of communication, social networks, stakeholder involvement, leadership support, and the perceived advantages of the innovation in influencing teachers' acceptance and positive implementation. The selection and combination of frameworks may vary depending on the context, nature of the change, and the specific requirements of teachers and educational institutions, recognising and exploring transition pathways throughout the implementation phase.

Nevertheless, in this regard, understanding uncertainty as the movement through which agency can be restored is crucial for enabling educational actors to lead implementation. The process of integrating new technologies is inherently non-linear and open-ended. Within this space, imagination, experimentation, and the recombination of old and new elements can thrive, fostering meaningful decision-making and agency. Becoming is not only about moving towards new possibilities but also about letting go of what will no longer be or other alternative modes of being. Understanding and incorporating this dynamic movement into the strategic training axis is indeed particularly rich for engaging educational actors in practices of meaning and collective agency.

In relation to the fourth research question, focused on implementation strategies, participants covered the course technological, relational and processes domains. With respect to the technological domain, they proposed the following:

  1. (3)

    Use AI not only to support course design, delivery, and assessment but also to redefine and concretise AI tools within dialogical practices.

Participants highlighted the need for direct access to AI tools in order to comprehend their techniques and subsequently assess their potentials and pitfalls. They suggest complementing theoretical discussions of ethical concerns with practical training using various AI resources to support course delivery and assessment. Methods for implementation include offering guidance in course forums through AI teaching assistants, fostering collaboration with AI facilitators, and evaluating course activities using learning analytics. Additionally, GenAI can be trained to provide models for teacher assistance, incorporating school and national educational guidelines. This includes addressing several topics which can be chosen by the trainees, such as managing challenging classroom behaviours or offering pedagogical tips to improve student engagement with a given subject. The suggestion aligns with the competency outlined in the UNESCO draft, which advocates for teachers to progress from understanding fundamental AI concepts and functions to comprehensively mastering the adaptation of AI tools to develop solutions tailored to their own educational contexts.

This strategy should not be perceived solely through an instrumental lens. It goes beyond mere tool usage to understanding its functionality, enhancing learning experiences, and addressing the ethical challenges these resources may pose. A perspective aligned with Bardone et al.’s (2024) concept of tinkering would be more dynamic, exploratory, and diachronic for all educational actors. Through this approach, they may participate in the process of concretisation of these tools, as defined by Simondon (2008). A new iteration of the technological object would facilitate the concretisation of its overabundant functions, potentially leading to new technological possibilities within a technogeographical space situated between nature and technics. This process involves the integration of the technological object with its surroundings, demonstrating a level of internal compatibility that produces its external adaptability. Importantly, such an approach holds dual significance: it pertains to the concretisation of pedagogical practices themselves, reshaping evolving roles among all actors; and it also transforms power dynamics in shaping narratives surrounding AI for education.

The relational domain was also covered with respect to the fourth research question on implementation strategies, with participants advocating for:

  1. (4)

    Fostering individual and collective agency on AIED ethical issues through informal and self-directed learning channels.

In addition to engaging in critical reflection, prioritising training initiatives that enhance collective agency is crucial. In fact, teachers are expected to translate insights into meaningful actions aligned with the specific demands of their educational environments. So, another insight derived from this study is that any training fostering a reflective and collective agency journey in AIED and education ethics must persist through various channels, including professional learning communities, communities of practice, online professional learning networks, teacher associations and unions, faculty meetings and committees, school leadership teams, and action research projects. The UNESCO draft on AI for Professional Development also encompasses this perspective, viewing AI as a facilitator of learning for teachers. It emphasises the importance of teachers critically adjusting or altering AI tools to better meet the evolving needs of professional communities and contexts. This makes it clear that in the context of AI, professional learning is key, namely through informal communities and self-directed endeavours.

Collective agency can also be fostered in educational cultures that allow for contextualisation and autonomy, through curriculum flexibility, adapting to the diverse needs of student populations and broader community requirements in feasible timeframes (OECD, 2018). This work can support the building of alliances that advocate for the pertinence and sustainability of bottom-up decisions. According to Crary (2022), these bottom-up strategies are critical to collectively reshape the landscape of technology in education, contributing to a social movement that brings about collective benefits and shared risks. This assertion gains heightened significance in light of the prospect of AIED automating pedagogical practices that lack contextual meaningfulness, coupled with the risk of potentially disempowering teachers in the long run. In a more positive approach, by incorporating AI techniques thoughtfully and ethically, educators can amplify collective agency. This can be achieved by leveraging data in a contextualised and insightful manner to inform decision-making and collaborative planning processes. Moreover, collective agency can be fostered when educators are not only motivated to ethically integrate AI into educational practices (educating with AI) but also to develop students' understanding and critical thinking about AI itself (education for AI). Initiatives such as the European Commission's Code Week (European Schoolnet, 2021) and the MIT RAISE programme (RAISE, 2024) exemplify this approach by supporting or providing K-12 curricula to expose students to AI at an early age. As Gertz (2016) argues, technology not only mediates our nihilism but also mediates our responsibility, serving as a means to prompt reflection on the human values that shape and nurture AI today.

A fifth implication for the implementation dimension regarding the processes domain has to do with the need of:

  1. (5)

    Employing narratives to contextualise these AIED technologies, considering ecosystemic factors.

The subcategories that emerged from participants' discussions, focusing on dealing with resistance and facilitating change, are associated with the implementation of training through innovative and purposeful andragogic strategies. In line with this objective, the ethical scenarios derived from the initial stages of this educational design research were considered relevant by participants for knowledge acquisition processes, aiming for consensus. Therefore, another recommendation is to employ narratives to contextualise these AIED technologies, considering ecosystemic factors, implementation strategies, and societal dynamics. Consequently, in the upcoming phase of this study, these dilemmas will undergo review to incorporate specific instances reflecting advancements in AIED technologies concerning the same ethical dilemmas. These scenarios, combining both dystopian and utopian elements, will function as heuristic tools for reimagining education from an ethical and agentic perspective. This can aid in developing the UNESCO-recommended competency focused on a human-centred mindset, ranging from recognising opportunities and risks to demonstrating a profound understanding of societal impact and the ability to engage in transformative actions to address related challenges.

Another implication in terms of implementation strategies related to the processes domain concerns the following aspect:

  1. (6)

    Incorporate meaningful experiential learning as a means to stimulate dialogic ethics in AIED-related discussions.

The broad acceptance of GenAI has sparked concerns among educators across secondary to tertiary education levels. Despite ChatGPT not being crafted for educational settings, it has raised worries, particularly in scenarios where it can complete homework for students. Nevertheless, these platforms offer an opportunity to reassess and redefine the learning processes and corresponding pedagogical methods, potentially leading to more meaningful outcomes. This can entail students interacting with these platforms and then assuming the role of teachers to correct the AI responses. They may also position themselves towards the information, taking on roles such as researcher or opponent. Additionally, involving other community actors in the learning process to diversify information sources and fostering group and project-based learning are viable approaches. Indeed, teachers are the ones who understand the microgenesis of learning, the progression that unfolds throughout a learning session. So, using large language models can, in fact, be an opportunity to challenge tasks of assimilation, accommodation, and evocation of knowledge, while creating agentic opportunities to deal with this technological novelty within school. This can be purposefully achieved by engaging students in other high-order learning processes, such as perspective-taking (from both humans and machines), introspection (focusing on the uniquely human process of self-reflectiveness), imagination (inspired by the vast possibilities enabled by AI), affective-cognitive dissonance (introducing dilemmas brought by the interaction with AI that stimulate ethical reasoning), and exploration of engagement with action through balanced power and agency (teachers share decisions with their students about how to learn and be assessed with GenAI).

So, another suggestion arising from this study is that training is expected to integrate meaningful experiential learning as a means to fully embrace dialogic ethics in learning settings. This can be maximised through activities such as role-playing and role-taking, enabling educators to assess and refine their strategies directly in their classrooms during the training period. Subsequently, they can return to the secure training space to evaluate their practices alongside peers and trainers. This also enables creating a platform for debate on ethics among educators, between educators and trainers, among students, and between teachers and students, ensuring that inter-generational perspectives on AI challenges are not only debated but also infused into subjects, potentially leading to valuable outcomes. This strategy directly aligns with the UNESCO suggested competency of guiding teachers to progress from identifying the potential benefits of using AI systems to critically evaluating AI in pedagogical practice, ultimately moving towards transformative pedagogy.

4 Conclusion

Until recently, the ethical foundations of education in the context of AIED have not been thoroughly explored, and the corporate narrative continues to dominate public discourse on AI for education. Debates on this topic seem to be especially pertinent as we live in a world of technique, as Ellul (1964) defined it: a world embedded in the pursuit of methods derived through rational means and achieving optimal effectiveness across all areas of human endeavour, where the humanities have lost their pace. AI technologies have become more pervasive, and major multinational corporations, in cooperation with at least one local government, are attempting to influence educational practices. Non-governmental curricula are already incorporating AI learning outcomes at middle and high school levels (UNESCO, 2022). Furthermore, various research findings indicate that AI curricula only marginally address AI ethics, and there is a shortage of opportunities for teacher preparation to tackle contemporary challenges in this area. To bridge this gap and capture diverse stakeholders’ perspectives, this study focused on insights on education ethics in the AI era stemming from teachers and teacher trainers focus group discussions. Envisioning the subsequent step as the development of a training course in this area, this research phase contributed significant insights on course objectives, implementation strategies, and design stages. By prioritising transversal assets beyond AI techniques and technologies, participants highlighted the significance of continuing professional development as a platform where teachers actively contribute to shaping their professional growth and standing, safeguarding their agency against potential encroachments by tools or policies. They were also recognised as key figures for challenging established knowledge in their field, while also questioning the very nature of epistemology and the conditions under which it is nurtured. Therefore, participants articulated their perspectives regarding the revitalisation of educational paradigms and teachers' roles, not only as gatekeepers of high-quality content, but as vital agents of thought and action in the face of the polysemic landscape of AIED technologies.

While education remains subject to state regulation, opportunities exist for elevating the status of teachers across diverse educational systems and impacting public policy development, in the interplay between technology, pedagogy, and society. Additionally, increased investment in professional learning and participatory policy reforms, leading to policies that are more comprehensive, has the potential to significantly enrich the outcomes of national investments in AI for the educational sector. In this context, the present research has sought to provide educators with an avenue to explore relevant educational possibilities and to assume proactive roles as agents of meaningful transitions in their learning environments. Participants' insights shed light on the importance of a broad ethical exercise and strategic collaboration among educational actors. They also revealed a nuanced understanding of the cognitive and emotional challenges teachers face during the integration of AI technologies, where uncertainty plays a key role. This calls for a comprehensive consideration of meaning indeterminacy and underscores the importance of carefully considering signification processes as they unfold in pedagogical practices. While the recommendations arising from the study underline the critical role of teacher autonomy in professional development, especially in the era of AIED, realising this objective can be a complex undertaking due to constraints related to limited resources, hierarchical organisational structures, and cultural factors that may hinder teachers' capacity to exercise professional autonomy. This clearly exemplifies the complex interplay between structure and agency to which the current research and further training aim to offer actionable insights. The participatory methodology used in designing research and teacher training situates this study within the broader discourse of dialogic ethics. This approach is sought to be responsive to the challenges posed not only by AI technologies, but by a world driven by technique. Its only when uncovering the technique layer that is overarching education too, that a consistent dialogue on ethics for education can be established, going beyond mere debates on the ethics of AI.

Finally, while this research advances past studies that merely explore the ethical challenges of AIED by offering concrete recommendations for training programmes targeting teachers in this domain, it is necessary to acknowledge its challenges and limitations. One challenge arises from the disconnect between the primary examples of the scenarios toolkit and the rapid advancements in AI technologies. This underscores the imperative of integrating more relevant examples that align with the current state of the art, thereby enhancing the applicability of the toolkit for the upcoming training. The scenarios will be used with these updated examples, as they depict ethical dilemmas that remain relevant today. Other constraints in the study emerge from concerns about generalisability, given that the insights derive from a particular participant group, even though efforts were made to encompass participants from various geographical backgrounds. Moreover, the study primarily focused on providing recommendations for teacher training in AIED, agency, and dialogic education ethics, without exploring the actual implementation and effects of such a training programme. This research calls for the next step, which pertains an examination of its impact on agentic decision-making regarding the use of AIED in educational settings. Additionally, cultural and societal factors can significantly influence the level of autonomy granted to teachers, complicating efforts to introduce alternatives within existing structures, reform those structures, or challenge and resist them, as advocated by Wright (2010). In certain educational contexts, traditional norms and hierarchical power dynamics impose strict adherence to prescribed curricula and pedagogical methods, leaving limited room for individual autonomy or innovation. This lack of professional autonomy restricts teachers' ability to shape their own professional growth and may hamper their capacity to read their learning ecosystems in light of evolving educational demands, practices, and pressures. As a result, these contexts are less likely to play a critical role in challenging the predominant dominance of technique, reaffirming educational ethics and leveraging the potential of AIED technologies when integrated meaningfully. Consequently, they are more prone to encountering the ethical challenges associated with their use.

5 Appendix 1

Table 4

Table 4 Focus groups script
Full size table

6 Appendix 2

Table 5

Table 5 Table of node categorisation in NVivo
Full size table