Artelt and Kunter (2019) provide a comprehensive summary of typical use cases and according sub goals involved in teaching and learning. They address the design of lessons or lectures, focusing on the “delivery of teaching content (Peng et al., 2021)” detail this as the cognitive processes of “providing knowledge” and “receiving knowledge,” which (Tsankova and Manolova, 2022) contextualize within education.

A broader perspective on related essential use cases and connected goals as part of an AIS-supported “interactive process of teaching and learning” is given by Munna and Kalam (2021). A very basic and early definition of “learning” emphasizes that “Many, but not all, forms of learning have to do with acquiring knowledge—either knowledge that something is the case or knowledge how to do something (White, 1998).” A more recent understanding of related “competencies” describes them as “knowledge, skills, attitudes, values, motivations, and beliefs people need in order to be successful in a job (Selvi, 2010).” Additionally, Musial et al. (2012) connect the terms “teaching” and “skills” with higher granularity as the “practice implemented by a teacher aimed at transmitting skills (knowledge, know-how, and interpersonal skills) to a learner, a student, or any other audience in the context of an educational institution.” Moreover, they state that “teaching is closely related to learning, the student's activity of appropriating this knowledge.” Also, Bialik et al. (2015) further reflect on skills and competencies, whereas they suggest that “skills relate to the use of knowledge and engage in a feedback loop with knowledge,” whereas “competency, on the other hand, is defined as the set of knowledge, skills, and experience necessary for future, which manifests in activities.” Complementary, Zamora and Zamora (2022) mention that “for some, skills are a combination of the knowledge, abilities and experience they have obtained both before entering the profession and during their employment.” Also, they reflect on an interplay between teaching and “facilitation of learning,” while in this context a recent work of Rone et al. (2023) strictly emphasize the importance of engaging motivation and classroom participation of learners. In education the interactive process of teaching and learning, particularly with the support of AIS, lead to a significant paradigm shift in recent years (Gentile et al., 2023). A more modern understanding of interactive teaching processes and required skills and competencies is introduced by Albrahim (2020); Blane (2021); Garcia (2024) and Ng et al. (2023b). Accordingly, still vital learning activities (Jakavonytė-Staškuvienė and Mereckaitė-Kušleikė, 2023; Gericke et al., 2023), but also future (digital) learning processes and their required skills and competencies are discussed by Abendan et al. (2023). Thus, important future-proof goals, as part of interactive blended teaching and learning, particularly involve the development of skills and competencies in the digital age (Marr, 2022; Thornhill-Miller et al., 2023) also including Extended Reality and Metaverse applications (Jagatheesaperumal et al., 2024; Pregowska et al., 2024).

Consequently, with the aim of offering more precise and practical definitions, “Education” is selected as a specific field, encompassing one essential goal of “developing skills and competencies.” In this field, the most apparent process to achieve the development of skills and competencies is defined as “the interactive process of teaching and learning,” including human actors, a blended teaching/learning system (Bhadri and Patil, 2022; Janse van Rensburg and Oguttu, 2022; Venkateswari, 2024), AIS and a wide array of corresponding use cases (Figure 2). Emphasizing an UML-based approach, any actor is in a specific role, whereas remarkably any use case can be accomplished by exhibiting certain behavior. Ethical behavior is also a well-known concept in digital AI-driven contexts, since “electronic interactions encapsulate both human-machine and machine-machine interactions (Buytendijk, 2014; Ashok et al., 2022).” These studies emphasize the importance of transparent and ethical behavior in processes where natural humans interact with machines. Therefore, ethical behavior is considered an essential additional low-level property of the proposed conceptual framework.

In summary, the proposed framework aligns with the previously discussed top-down definitions. Closely connected to the UML, it identifies “Education” as the field, “development of skills and competencies” as one essential overarching goal, and the AIS-supported “interactive process of teaching and learning (lessons or lectures)” with its various use cases. Furthermore, it involves human educators, human learners, and AIS as actors in specific roles which must exhibit ethical behavior while accomplishing designated use cases. Figure 2 gives an overview on most essential use cases for typical interactions of human educators, learners, and AIS with a blended teaching/learning system in school education. The objective is to support a systematic and model-based concept, refined to a granular level, that is implementable. Thus, ancillary to a risk-based assessment of any AIS at a higher level, this framework may facilitate the concrete ethical justification of related low-level use cases.

In the previous section, essential preliminary definitions of the conceptual framework were identified to frame a concrete process. However, the framework still lacks detailed characterization of the involved actors. In response, the low-level human-sensing taxonomy (HST), introduced by Teixeira et al. (2010), is incorporated to facilitate an ethical use of AIS in the presence of natural human actors. In the context of typical educational processes, it is extended to a more detailed set of low-level properties. With the types of actors in their according roles as a starting point, it is possible to even distinguish between natural human educators, learners, and AIS. This distinction is crucial, as future AIS may be fully capable of behaving like natural humans, potentially leading to unethical mimicking or replacement. Whilst supporting or augmenting is a role widely accepted by the educational community, it is important that AIS do not replace natural human actors (Flores-Vivar and Garćıa-Peñalvo, 2023). Consequently, the first UML-related key property emerges, describing different “roles of actors,” also depicted in Figure 2. Further, in alignment with the HST, the set adopts the “number of people in an environment (count)” as the UML-based “mulitplicty” (Figures 1, 2). Hence, as an intermediate result, the framework defines three different roles of actors with the according multiplicity:

Natural human educators, human learners, and AIS, as the principal actors, may interact with a dedicated teaching/learning system in their specific roles, contributing to the overarching goal of developing skills and competencies. The interactions of the AIS are modeled with either ethical or unethical use cases (Figure 2). Combined with a well-defined role and multiplicity, the sum of these properties are intended to significantly clarify a potential process in education already. However, at the same time they question whether the actors are interacting simultaneously within a particular system, or if a potential for misuse must be considered. Concerns may be negligible in traditional lessons or lectures, where actors are physically close, and their presence and role can be readily identified. Nevertheless, issues may surface in typical distance learning settings, where actors are interacting online, over wider distances. To address related issues, it is proposed that essential properties of an actor's presence be identified to foster ethical justification.

The full spectrum of natural human presence is a multifaceted, complex, and long investigated topic (Segal, 1996), certainly exceeding the boundaries of this article. Nevertheless, essentially the multi-sensory perception of human presence can be described by the process through which humans interpret sensory information to construct their individual experiences of the real world. Furthermore, in addition to the HST, Cornelio et al. (2021) present an extensive review of recent technological advances related to the main human senses: vision, audition, touch, olfaction, and gustation, whereas research states that vision is the “most valued” (Enoch et al., 2019) and “most important” (Hutmacher, 2019) sense, being the sense humans mostly rely on. Consequently, the most relevant spatial and temporal, along with the visually recognized strongest, properties of any involved actor, are synthesized in the remainder of this section, leading to the “visual representation.”

Reflecting on essential temporal and spatial properties in an educational context, the concept of blended teaching/learning provides widely accepted definitions. It distinguishes clearly between synchronous (online) and asynchronous (offline) educational settings (Cleveland-Innes and Wilton, 2018) leading to the property of “synchronicity.” In synchronous settings, educators and learners are virtually present simultaneously, whereas in asynchronous settings, they are not. Moreover, blended teaching/learning distinguishes between co-located (face-to-face) and distant (distance learning) settings representing the property of “location.” In a co-located setting, educators and learners are physically present, contributing to the interactive process of teaching and learning at close distances (d < < ). This may pertain to a typical school lesson in a classroom or a traditional lecture in a lecture hall. Conversely, in a distant setting, educators are typically connected over greater distances (d>>) through distance learning and, unlike face-to-face settings, are not present in the same physical space (Ayu, 2020).

In addition, the “embodiment” of the involved human actors and AIS may help to better define physical presence (Wainer et al., 2006; Mollahosseini et al., 2018), or, at least, its visual perception for distant settings (Bonfert et al., 2021). In contrast, any missing representation of embodiment, may lead to omnipresence, also causing issues with trustworthy AIS in similar settings (Kim et al., 2018; Sathikh et al., 2022). In the same way it may reduce transparency and trust in educational processes. Moreover, if the representation of embodiment is entirely missing, compared to any obviously (visually) represented actor, it may become significantly harder to distinguish between the roles of actors (HAO, ACC). Essentially, any required visual indication of the actual physical embodiment of the AIS could help to explain a more concrete spatially limited reference, better clarifying the boundaries of the AIS to improve TRA (Figure 3).

Under the assumption that any natural human actor's visual sense is the most important (Hutmacher, 2019), the resulting definition of a visual representation, combining the three properties synchronicity, location, and embodiment, requires more specific definitions from the field of visualization. Being particularly important for distance learning, the goal is to characterize if the visual representation of the involved actors is real, virtual or something in between, posing a strong indicator for their presence. A rather interdisciplinary perspective on physical real and visual virtual presence is given by Lee (2004), whereas a long established taxonomy emphasizes visual properties of an actor, ranging from real to virtual (Milgram et al., 1995a). An according widely accepted scale is the RV-continuum (Milgram et al., 1995b), “revisited” more recently by Skarbez et al. (2021). More recent work focuses on the definition of design spaces that better clarify what is real and what is virtual in settings that are similar to the presented framework, however, they do not focus on detailed ethical considerations in the context of education, but investigate on the social relevance of a virtual assistants embodiment in everyday life (Kim et al., 2018). Moreover, whilst Lee et al. (2023) present design patterns for situated augmentation of physical referents, the work of Suzuki et al. (2022) puts a stronger focus on augmenting interactions between natural humans and robotic AIS. Additionally, an extensive review of definitions of a digital twin, describing similar transitions between physical and virtual entities, is provided by Semeraro et al. (2021). Still, the unambiguous definition of a property with focus on ethical use in education seems to be missing. Specifically important for distant settings, the visual representation of any actor may be characterized by three distinct cases:

Remarkably, a classification as virtualization requires, above all, synchronicity and at least a visually true relative scale and accurate color of the embodiment (Table 1). The location is noted to more clearly define a real actor and distinguish between co-located and distant settings, however not relevant for a virtualization since it is typically displayed with wrong absolute scale and non-coherent location during distance learning. The main purpose of the virtualization is to preserve the presence of natural human actors, particularly regarding the visual properties. If not considered, any actor may not be identifiable as such, in turn leading to a great potential for misuse. As a consequence, negative effects are to be foreseen regarding TRA, ACC, and HAO. Summarizing, the conceptual framework suggests the following essential low-level properties to classify a visual representation: The temporal coherence or synchronicity, reflecting the time difference between actors; the spatial coherence or location, which is the distance between actors; and a strong coherence of the embodiment, indicating if the actor even exists as such in the real world. Posing a substantial property for distance learning, the visual representation of an actor could help to better distinguish between ethical and unethical use.

The role property, previously discussed in Section 3.2.2, is essential to even differ between human actors and AIS, thus maintaining HAO and TRA and preventing misuse (ACC) (Figure 3). Relevant use cases, beyond the field of education, include for example medical applications. Fiske et al. (2019) and Shuaib et al. (2020) express their concerns about robots replacing therapists or doctors, whereas more recent works address potential unethical roles of AIS in concrete medical educational settings (Cornwall et al., 2024). With focus on education, various works investigate on the replacement of educators with AIS in academia (Karki and Karki, 2023) or higher education, also being “better equipped to deliver assessments” and grading (Chan and Tsi, 2024). Also, Okulich-Kazarin et al. (2023) provide insights into the possibility of replacing teachers with AIS. Besides, once an AIS is implemented for a specific use case, concerns about a robust traceability, explainability, and avoidance of unfair bias of decisions imply potential unethical use. Strategies to counteract are suggested by HLEG (2021), whereas “algorithms used, should be documented to the best possible standard and decisions made by an AI system should be understandable” (TRA), while it is recommended to use “oversight processes to determine AIS' purpose, constraints, requirements and decisions in a clear and transparent manner” (DNF). However, these requirements and mechanisms also require concrete and detailed models to effectively prevent potentials for unethical use. As a strong indicator of potentials for unethical replacing roles and use cases is a gradual social deterioration (also highlighted by the works of Friedman, 2023; Frambaugh-Kritzer and Petroelje Stolle, 2024 and Indurkhya and Sienkiewicz, 2024), this article stresses that for many use cases in education, AIS can enhance learning experiences, however, “cannot replace the vital role of human interaction in education” (Zaman, 2024).

The basic considerations of Section 3.2.2 on the multiplicity are entirely adopted to determine if at least N (1..*) one natural human educator and n (1..*) one natural human learner are participating in the interactive process of teaching and learning. In this context, potentials for unethical use may include the full or partly replacement of natural humans by AIS (Fitria, 2023; Etiubon and Etiubon, 2023). Thus, determining the multiplicity of AIS with m (0..*), helps to distinguish if such are even involved into the overall process, including all ethical implications, or not. Supplementary, the multiplicity is modeled in Figures 1, 2.

Once the roles and the multiplicity are well defined, an ethically justifiable, exemplary behavior of all involved actors to accomplish use cases and overarching goals in the intended interactive process of teaching and learning is essential. Closely connected to the goal of developing skills and competencies it is a fundamental concept in education (Gunderman, 2002; Lumpkin, 2008; Khusnidakhon, 2021; Dowd and Green, 2022; Jaedun et al., 2024). Educators do not simply deliver teaching content without any self-reflection. Rather, they are challenged to show exemplary behavior to the learners as a role-model, also motivating and engaging active participation (Nasir and Hossain, 2023). If the behavior of any AIS contradict this concept, this may result in a significant negative effect, in the same way affecting HAO and SEW. A complementary example of misuse, emphasizing the interplay between role and behavior, is shown in Figure 5.

Although, co-located in presence participation of human actors in teaching and learning in-class can have significant positive impact on the educational experience and performance, a variety of modern educational processes involve distance learning. However, many facets regarding natural human presence are still hard to maintain over wider distances, from a social and interpersonal perspective (Leo et al., 2021; Anastasakis et al., 2023; Kornfield et al., 2021), but also from a technical-scientific perspective (Tsankova and Manolova, 2022; Dustova, 2023). Once improved, such facets could pose a vital basis for facilitating the overall quality of developing skills and competencies during distance learning. Nevertheless, in such settings, natural human presence may be more elusive and more prone to misuse (Figure 4), particularly with focus on HAO, TRA, and ACC. In these regards related work underlines “concerns with accountability, agency, and surveillance in online learning” (Seo et al., 2021), or stresses “a real danger of AI becoming pervasive in every sense where those involved may be exposed to risks without being aware of them” (Kamalov et al., 2023). Other works highlight issues with HAO and ACC, since (online) virtual AIS “will enable conversations that are hardly distinguishable from real conversations with humans, but they will also raise concerns about bias, transparency, and accountability (Piñeiro-Mart́ın et al., 2023).” Concluding, on one hand, the location of the human actors clearly effects the overall teaching and learning experience. On the other hand it can have great impact on the potentials for misuse of AIS in distant educational settings.

There is ongoing debate as to whether synchronous settings clearly provide benefits to the overall teaching and learning experience of students. If social aspects are concerned, related research investigates on the development of social networks and emotional interaction as part of an online monitoring framework for teaching and learning (Spadavecchia and Giovannella, 2010). Throughout analyzing text-message based interactions they found that dense social interaction can be also achieved in asynchronous online-settings, however socialization takes more time. Besides, Giovannella et al. (2011) found that teacher/learner interactions, like grading, can be seamlessly integrated, either in blended or pure online settings. Other related work suggests that both synchronous and asynchronous settings can enhance the overall quality of teaching and learning, whereas the preference which setting to choose strongly depends on the type of learner (Higley, 2013). Nevertheless, with focus on social aspects, related work concedes that during blended teaching/learning in general students “deem very important the collaboration/interaction among peers and with the teachers” (Giovannella, 2021) or, with synchronous settings, “feel a stronger sense of connection to their peers and instructor and stay engaged with course activities (Yamagata-Lynch, 2014).” Becoming specifically relevant during the COVID-19 pandemic, various works investigated on the efficacy of blended and hybrid learning, studying and also supporting synchronous settings (Priess-Buchheit, 2020; Raes et al., 2020). More related research indicates that during the pandemic a majority of students preferred a synchronous setting. However, asynchronous settings were typically chosen in case of technical difficulties, whereas benefits of synchronous learning are that the students “can interact in real time (Almpanis and Joseph-Richard, 2022).” Emphasizing the property of synchronicity, more related work indicates synchronicity as a “significant role player” (Hepburn and Borthwick, 2021) with a positive overall effect on the (blended) teaching/learning experience, if technically feasible (Katai and Iclanzan, 2023; Belt and Lowenthal, 2023).

Especially in distant settings, the characterization of natural human actors with focus on their visual representation is of great importance. In the worst-case, the representations of all actors significantly differ from the real-world, or are even non-existing (Table 1, “Purely Virtual”). On one hand, such representations are a basic concept for other systems, like multimedia platforms (Herman et al., 2020) or avatars in gaming (Szolin et al., 2023). On the other hand, if the goal is to preserve natural human presence in educational processes with focus on TRA and ACC, this property may be handled with care. If distant and/or asynchronous settings are combined with purely virtual representations of AIS, the room for misinterpretation and misuse may be significantly widened (Figure 3) and must be addressed accordingly (Bardzell et al., 2007). In the worst-case, preventing AIS from mimicking representations of natural human actors may become practically impossible (Figure 4).

To better justify the ethical use of AIS in education, the presented low-level framework is substantiated with concrete guidelines, summarized as activity diagram (Figure 6). Whilst concrete questions help to justify potentials for unethical use, the diagram provides practical, and easy to implement yes/no decisions for using AIS as part of an intended process. Thus, clarity regarding an ethical use in education may be drastically increased. The diagram is based on the novel set of properties, derived in Section 3.2 and reflected on in Section 3.3, and relates to typical educational settings modeled in Figure 2.