Keywords

1 Introduction

Dialogue systems and conversational agents, including chatbots, are becoming ubiquitous in modern society. Chatbots – also called machine conversation systems, virtual agents, dialogue systems, and chatterbots – comprise computer programs that are used to simulate auditory and/or textual conversations with users, or other chatbots using natural languages [46, 56, 59]. Even though chatbots have been around for decades, with their origins dating to the 1950s with the so-called Turing test, a renewed interest in chatbots appeared in 2016, due to massive advances in artificial intelligence (AI) and a major usage shift from online social networks to mobile-messaging applications [10]. Nowadays, chatbots are becoming a trend in many fields such as personal assistants on mobile devices, in customer service, in medicine and healthcare, and lately in education [45, 56].

1.1 Chatbots in Education

Given that nowadays a significant part of the learning and teaching process utilizes Information and Communication Technologies (ICT), chatbot technology can be considered an important innovation for e-learning [11] for addressing the needs of the learning community (i.e., learners and instructors) [8]. A chatbot solution might offer support for different teaching and learning tasks, depending on its architecture and technology used (i.e., retrieval-based models, [36] and generative models, [47]). On the one hand, chatbots have the potential to facilitate students’ learning and offer interactive learning experiences for the students, and even smoothen the transition of secondary-education students into university environment or help to increase university enrolment (e.g., the Pounce chatbot introduced at the Georgia State University for safeguarding students’ successfully transition to college). On the other hand, chatbots can ease the instructors’ workload, by acting as teaching assistants and enacting the role of tutors, e.g., by answering student’s questions and to Frequently Asked Questions (FAQs), by sending reminders to the students for upcoming deadlines, or even by conducting online assessments.

Chatbot solutions have been already applied in multiple educational contexts, such as, in health and well-being interventions [1, 7], in medical education [28], in mathematics education [30], and in language learning [4, 17, 25, 26, 32]. Also, their use in learning scenarios can be identified in recent studies, in which chatbots appear to function as teaching assistants [18], as tutors supporting students in learning general knowledge subjects [13], for motivating students [40, 57], for feedback provision and metacognitive thinking triggering [29, 35] and as formative assessment solutions (e.g., peer assessment chatbots in Lee and Fu study [34]). Similar chatbot uses, serving teaching and learning purposes, have been identified in the literature review of educational chatbots for the Facebook Messenger [48], such as recommending learning content to users, providing feedback, questions, and answers (Q&A), setting goals and monitoring learning progress.

In recent years there has been an attempt to further exploit the capabilities of educational chatbots, especially for large-scale learning scenarios at universities or in massive open online courses (MOOCs), since chatbots have the potential to provide individual support and feedback to students with no further financial or organizational costs for the providers [56]. Nevertheless, the effectiveness of chatbots in education is complex and depends on a variety of factors, such as students’ individual differences and personality traits [22, 49], their educational background, their social and technological skills [38, 39] their self-efficacy and self-regulated skills [49], and instructors’ willingness and their attitudes towards using chatbots in routine teaching [8]. Also, when chatbots do not meet certain requirements such as easy use and access, they add little value to the learning process, including cognitive, meta-cognitive, and affective dimensions, and there is a high possibility of not being used by the students [20]. Further, one of the major challenges relates to the development and sustainable use of a smart chatbot, capable of delving into content-related topics and offering personalized guidance and feedback to the learners [37].

Therefore, the design of new chatbot solutions and their meaningful integration in higher education presupposes a good understanding of users’ personal traits, their needs, and expectations, as well as, an examination of their perceptions towards educational technology [27, 43], the adoption of an appropriate pedagogy [19], and last but not least, a confrontation of technological challenges and potential limitations, that come relate to the Natural Language Processing (NLP) research field [58]. In addition, current chatbot applications suggest that conversational user interfaces still face substantial challenges, also related to the human-computer interaction (HCI) field [16].

1.2 Users’ Expectations for the Use of Chatbots in Higher Education

Several chatbot solutions have already been designed, developed, and tested, as resulting from the review of literature on chatbots in education. Yet, chatbot initiatives often ignore user needs and user experiences [10]. As the field of chatbots is gradually expanding, developers and designers have an urgent need to understand the user needs that motivate the future use of chatbots, which is required to make successful automatic conversational interfaces [10].

Brandtzaeg and Følstad [10] outline some of the new needs and challenges posed by the emergent trend of chatbots. Those include, among others, the need for: (i) having an effective and efficient accomplishment of productivity tasks, (ii) obtaining assistance or information, (iii) using chatbots for fun or entertainment, (iv) addressing social and relational factors, (v) reducing loneliness, or enabling socialization. Previous research along these lines and with the focus on educational chatbots has also already been done to some extent, projecting possible uses of chatbots in education [21, 33, 48, 50, 55]. Findings of those studies outline the potential use of chatbots as career advisors [33, 54], as intelligent tutors answering students’ questions [21, 55], as means for improving users’ soft skills [54], and as means for formative assessment implementation with the provision of qualitative feedback [55]. Moreover, users’ expectations for the design characteristics of chatbots have been already explored (i.e., recognition; visibility of system status; anthropomorphism in communication; knowledge expertise, linguistic consistency; realistic interaction) which may enhance the feeling of trust and support students in a personalized and interactive way [50].

2 Research Questions of the Study

A few educational chatbot solutions have already been designed and developed in Higher Education, being yet an emerging educational technology within many countries. The examination of users’ needs and expectations has been explored to some extent, with a focus on chatbots’ characteristics and HCI implications. However, an investigation of how higher-education students and educators envision the meaningful integration of chatbots in higher education in a broader scope is still missing. In response to this, the present investigation aims to document the envisioned pedagogical uses of chatbots in higher education and the benefits and challenges related to their use. The following two research questions are being addressed in this study: (1) What are the envisioned pedagogical uses of chatbots in higher education as proposed by learners, instructors with relevant experiences, and experts in the domain of chatbots? (2) What are the benefits and challenges of using chatbots in education as perceived by experts in their proposed pedagogical uses of chatbots?

3 Methodology

This is a qualitative injury. Data collection is based on in-depth, semi-structured interviews with various stakeholders, including students, instructors as well as experts in the field of AI and educational chatbots.

3.1 Participants

Instructors and students from different local universities in Cyprus were invited to participate in this study. Ten instructors and eight undergraduate students, with different backgrounds (i.e., Multimedia and Graphics Art, Computers Science, Sport & Exercise Science), volunteered to participate in semi-structured interviews. Also, all the participants had experienced at least once an interaction with a chatbot, regardless of the field of application, as confirmed by the interview data. Additionally, experts from the field of AI and educational chatbots participated in the study. Two came from academia (one from each organization: Bellwether College Consortium, University College London - Knowledge Lab) and two came from the industry (one from each organization: Sintef Digital, CYENS Centre of Excellence). A total of 22 interviews were conducted via teleconferencing. All the participants consent to anonymously use the data for research purposes.

3.2 Data Collection

Data from in-depth, semi-structured interviews were considered to be the most important source, serving the needs of our research aims, as they provided us with rich and deep insights into the concepts under investigation [14]. The semi-structured interviews were driven by two protocols, developed by the authors. The first protocol was used with higher-education instructors and students and included questions on the previous experience of the users with chatbots (e.g., Have you used a chatbot before? With reference to this previous experience, what would you change in the chatbot(s), so that it would better address your needs?) and their needs and expectations on a potential chatbot use (e.g., In your opinion, could a chatbot facilitate your learning/teaching progress? How? Please explain your reasoning.). A slightly adapted version of the first protocol was used with the experts, including questions on the state-of-the-art on pedagogical chatbots in higher education in terms of areas of application (e.g., What is the state-of-the-art on pedagogical chatbots in higher education? In terms of instructional designs integrating chatbots? In terms of areas of application and learning scenarios?) and the potential of educational chatbots to facilitate teaching and learning (e.g., How could chatbots facilitate students’ learning in higher education?).

The second protocol, was used merely with experts and included questions on the added value of chatbots in education (i.e., What is the added value for chatbots in higher education?), and the benefits and challenges raised in the proposed areas of application, the potential and/or possible constraints of chatbots to offer personalized learning experiences (e.g., What is the potential of chatbots in offering personalized learning experiences, including personalized feedback to learners? What types of features should chatbots have, serving this purpose? What are the possible constraints?). All the interviews were recorded upon the interviewees’ consent and transcribed verbatim.

3.3 Data Analysis

For addressing the first research question of the study, the first dataset was utilized (use of first interview protocol). The study adopted a grounded theory approach in order to address the first research question. Specifically, the analysis of participants’ interview transcripts followed an iterative analysis was conducted as described in Strauss and Corbin [51, 52] and Thai, Chong, and Agrawal [53]. The coding operations employed for the data analysis included an open-coding process for concepts and categories, axial-coding for relationship between categories at their property and dimension levels, and selective-coding for our results’ formation. All these required constant comparison as an inherent task.

Our first step was to discover concepts through detailed examination of the data. We started with open coding; this was a line-by-line coding in which key statements used by the participants were noted, attempting to unitize the data. Open coding resulted in a total of 32 codes. Throughout the process, we maintained memos on the process to allow for selective coding, an iterative process of grouping and regrouping the codes into core categories. However, as the analysis progressed, the code system was continually trimmed down and refined for better understanding of the data and concepts underlying in the data. In the end, 23 codes were kept (see Table 1).

Next, we carried out axial-coding (i.e., comparison of data observations) to find the relationships that existed between the different codes corresponding to groups of conceptual properties and dimensions [52]. With the help of the literature, we categorized the codes under more abstract higher-order classifications in two different levels: (i) categories: different dimensions of the learning process (i.e., cognitive, metacognitive, and affective domain, administrative work); (ii) sub-categories: specific type of activities (e.g., counselling) and/or aspects of learning (e.g., emotional aspect). Following that, we went through a third coding processes, using the codes, the sub-categories and the categories which resulted in the previous steps. During this process, we related the categories found in axial-coding to the core category which represents the main theme of this study, that is the learning process. With the use of techniques, such as, the use of diagrams, sorting, and reviewing through memos, we concluded to three higher-order categories, reflecting the chronological intervention of chatbots in the learning process.

For addressing the second research question of the study, the second dataset was utilized (use of second interview protocol with domain experts). In this case, we adopted a thematic analysis approach [41] involving an open coding process followed by clustering of the emerged codes into broader themes portraying the (i) benefits and (ii) challenges of the proposed pedagogical uses of chatbots.

4 Results

4.1 Envisioned Pedagogical Uses of Chatbots in Higher Education (RQ1)

The data analysis revealed various envisioned pedagogical uses of chatbots in higher education which can potentially facilitate learning (Table 1). Those uses can be clustered based on the domain of learning that they serve (e.g., cognitive and affective) and instruction-related tasks (i.e., administration) and also the phase in which they are situated with respect to the learning process (i.e., prospective, on-going, and retrospective intervention).

Table 1. Envisioned pedagogical uses of chatbots in higher education
Full size table

Proactive Intervention.

This higher-level category refers to uses of chatbots in education which focuses on the cognitive learning domain (i.e., assessing students’ prior knowledge and setting-up learning goals), on social aspects of the affective domain, such as, the promotion of inter-social interactions among learners (i.e., the chatbot connecting students’ with common educational paths and initiating an interaction among those) and on the easing of administrative work (i.e., the chatbot providing administrative information for the university to freshmen) for smoothening the transition of secondary-education students into the university environment. Those uses were proposed by instructors and experts, whereas the students did not make any reference to the potential proactive use of chatbots. Indicative quotes from the interviews tangling the above-mentioned uses are given below (Table 2).

Table 2. Indicative quotes for the proactive intervention of chatbots in higher education
Full size table

On-Going Intervention.

This higher-level category refers to uses in which the chatbot can facilitate the on-going learning process and can be clustered in two different learning domains: cognitive and affective. In addition, a category corresponding to administrative work that facilitates or accompanies the learning and teaching process also emerged through the data analysis.

The first category involves uses with the chatbots targeting cognitive aspects of learning (see Table 3). Namely, one use relates to the role of the chatbot as a remote tutor in the learning process. Experts referred to the role of the chatbot as a remote tutor helping the instructor in lecturing and delivering content to the students and responding to students’ content-related questions. Students and instructors also envisioned chatbots as tutors for responding to students’ content-related questions. Specifically, the instructors-interviewees pointed out the need of having a chatbot responding to students’ content-related questions, rather than administrative matters, since the latter can be also addressed now through the mass amount of information that students already receive in social networks and e-learning systems. Specific functionalities that the chatbots could have for serving this use were also proposed. A functionality that the chatbot could have, is posing a question at the end of the conversation for assuring that the conversation has successfully taken place and the content-related issues have been resolved. Another functionality could allow the chatbot to share relevant resources with the students. Apart from the pedagogical value of this use of chatbots in the learning process, it also benefits the instructors in the sense that is saves time, but also the chatbot could be available at any time of the day, thus helping the students to receive immediate support, as acknowledged by the interviewees.

Interviewees further proposed a tutoring role of chatbots for offering tutorials for basic concepts of a course, supporting the learners while studying a new language, and even delivering a specific lesson that a student missed in the class. In addition, the instructors proposed specific instances in which remote tutoring, namely meaningful conversations between chatbots and students for learning purposes, could be beneficial for the students. One of these instances includes the scenario in which the chatbot can act as a guide in problem-based learning; in other words, the chatbot can offer step-by-step support in problem solving tasks to the students. In another instance, the chatbot could intervene in critical benchmarks and check points of the learning process, interacting with the students and making sure that the core learning objectives have been met. In addition, interviewees from the three categories (expert, instructor, students) proposed a mediating role of the chatbot in the learning and teaching process. In this occasion, the chatbot, while interacting with the students in content-related topics, has the capability to understand when the instructor should intervene in the chatbot-student conversation; therefore, the chatbot is acting as the mediator with the instructor to intervene in the conversation.

Table 3. Indicative quotes for the on-going intervention of chatbots in higher education in the cognitive domain
Full size table

Another major sub-category under the cognitive domain of chatbot educational uses includes the implementation of formative assessment. In this case, the chatbots can be designed for assessing students’ conceptual understanding on a topic, offer personalized and direct feedback to the learners and even identify learning difficulties of the learners. However, one of the experts stressed the importance of having well-designed chatbots for this educational use, as there might some risk that the chatbot will cause frustration and irritation to the students if it fails to provide meaningful responses.

The second category of the on-going chatbot uses serves affective aspects of the learning process, namely, emotional, motivational, and social aspects of learning (Table 4). The interviewees proposed the potential of chatbots to offer psychological support to the students and to fulfil the feelings of loneliness and anxiety (i.e., emotional), to motivate the students during their learning journey and query them in short conversations about their satisfaction with their performance (i.e., motivational), as well as, to promote inter-social bonding of learners and learners with mentors (i.e., social). For the social bonding idea, the experts referred to personalization and adaption of chatbots to respond to language, cultural and social norms of the students. Nevertheless, the social bonding idea was faced with some concerns from the experts’ side. Cultural differences were cited as a potential barrier in designing technological solutions that can really serve social bonding and explained that this is an area that needs further research and development. Besides this, the application of chatbots for social bonding was seen with caution by some of the interviewees (i.e., instructors and students), as they could not see any value of chatbots mediating in the social bonding with other humans.

Table 4. Indicative quotes for the on-going intervention of chatbots in higher education in the affective domain
Full size table

Last, the fourth category of on-going chatbots uses in the learning and teaching process encompasses complemented administration tasks (Table 5). In particular, the chatbot could be used for course counselling (i.e., course counselling, students’ affairs counselling, and services), for reservations (i.e., arrange and book meetings between a student and the instructor or for making a reservation in the laboratory and other university premises), and for course information and administrative processes (i.e., collecting students’ assignments throughout the semester, reminding students for course deadlines, facilitating and offering support during the submission process of assignments and responding to FAQs of the students that deal with administration). Last, the chatbots can be used for module feedback provision that typical takes place at the end of the academic semester and as part of the instructors’ evaluation. Students’ responses could be shared directly with the instructor, thus leaving room for improvements in the course delivery.

Table 5. Indicative quotes for the on-going intervention of chatbots in higher education in complemented administration
Full size table

Retrospective Intervention.

Uses in which the chatbot can facilitate the learning process retrospectively were proposed by experts and students; those uses have been clustered under the metacognitive domain. Chatbots may appear as personal reflective tools, getting in short conversations with the students, but also the instructors, for helping them to reflect on the learning and teaching process. Also, the chatbots can be used for summative assessment, with the implementation of quizzes and tests that measure students’ conceptual understanding of a topic (see Table 6).

Table 6. Indicative quotes for the retrospective intervention of chatbots in higher education in the metacognitive domain
Full size table

4.2 Benefits and Challenges of the Proposed Pedagogical Uses (RQ2)

As it has been revealed from the interview data, the envisioned uses of educational chatbots bears several advantages but also challenges, often linked to technological affordances and constraints. Among the advantages (see Table 7), the cognitive offload of the users was mentioned. That is, from the users’ perspective cognitive offload was explained in the sense that they do not need to memorize pieces of information, deadlines, etc., when a chatbot can offer this type of information easy and quickly via a short interaction with the user. Then, from the instructors’ perspective, cognitive offload was explained in the sense that the chatbot can reply to FAQs. More benefits include the low development cost, the capability of chatbots to address the needs of large-scale classes, the chatbots’ availability to interact with the users at any time of the day and their potential to interact proactively with the users and display precisely the information that the users are searching for. In addition, the type of interaction that takes place between chatbots and users can be characterized as a human-like and enjoyable and can be considered appropriate for online teaching and learning. In addition, the personalization aspect has been acknowledged as an advantage, since a sufficiently advanced chatbot should be able to accommodate the particularities of the user or the student interacting with a chatbot.

Table 7. Indicative quotes from the experts’ interview data on the benefits of the proposed pedagogical uses of chatbots
Full size table

However, the application of chatbots may be accompanied with challenges (Table 8). Personalization, even though being viewed as a potential benefit, it has been also listed among the challenges, as the technology is not ready yet to fully support this functionality, according to the experts in the domain. Related to the previous constraint, another challenge proposed by the experts, includes the limitation of a chatbot to handle complex conversations and the high error rate at the beginning of their use, which can evolve via machine learning techniques. As the domain expert-interviewees explained, this challenge affects in a great extent the successful application of pedagogical scenarios for students’ learning support, assessment, and personalized feedback provision. Another major challenge that was revealed in our analysis, deals with ethical considerations around the application of chatbots in education. This challenge bears several dimensions. First, a chatbot-student interaction can be viewed as an educational intervention which raises ethical questions in case of undesirable effects. Second, an orthogonal ethical consideration involves data privacy issues; that is the data that is gathered from a chatbot interacting with a student and its further use for research or policy making purposes. Finally, the experts argued that sustainability of use, high maintenance cost and potential user experience (UX) problems which might cause feelings of frustration to the users, were also outlined. Related to the latter point, it was explicated by one of the experts that it is necessary for the users (e.g., learners) to be aware of the capabilities of the chatbots and adjust their expectations accordingly, to avoid feelings of frustration.

Table 8. Indicative quotes from the experts’ interview data on the challenges of the proposed pedagogical uses of chatbots
Full size table

5 Discussion

5.1 Envisioned Pedagogical Uses of Chatbots in Higher Education (RQ1)

In this study, we sought to identify how higher-education users and experts in the domain of educational chatbots envision pedagogical uses of chatbots in higher education, as well as the benefits and challenges related to their use. The qualitative analysis relied on 22 in-depth, semi-structured interviews with learners, instructors, and experts in the domain of AI and chatbots. Our analysis disclosed envisioned pedagogical uses applicable for three distinct phases of the learning process: prospectively, on-going, and retrospectively. It further revealed a wide spectrum of potential uses of chatbots that the particular interviewees consider appropriate and meaningful to be integrated into higher education, covering core dimensions of learning in the cognitive [3, 9] and affective (emotional, motivational, social) [31] learning domains. In addition to the above-mentioned categories, we have documented uses of chatbots that support the conduction of administrative tasks that relate to the learning and teaching processes. Using technology effectively in administrative activities could provide more access to information resources and ease the teaching and learning processes [42]. Likewise, the proposed chatbot uses have the potential to increase administrative efficiency and even lead to innovative administrative approaches.

Even though several envisioned pedagogical uses have been already identified as chatbot solution in previous studies, such as chatbots acting as teaching assistants [18], as tutors [13], for feedback provision and metacognitive thinking triggering [29, 35], and as formative assessment solutions (e.g., peer assessment chatbots in Lee and Fu study [34]), still our findings portray potential uses of chatbots in a wider spectrum of educational applications. In addition, our analysis provides a useful categorization of chatbots’ uses into three distinct phases of the learning process and into different learning domains. In this sense, our findings depict in a non-fragmented manner the uses of chatbots in education, as envisioned by the participants of this study.

Of particular interest is the fact that chatbots have been suggested by our participants as means to enacting the role of the facilitator in problem-based tasks, while also intervening in the learning process, and interacting with the students in check points. This use aligns with two learning approaches. First, it aligns with the inquiry-based learning approach [2], in which the instructor appears as a facilitator intervening in the so-called check points for discussing with the learners; likewise, when appropriately trained, chatbots could enact this specific role. Second, this chatbot use relates to the problem-based learning model [44], during which the learners are engaged in self-directed learning, to examine and solve a given problem. Overall, chatbots were envisioned by the interviewees as facilitators in problem-based and inquiry-based learning and teaching. Such pedagogies adhering to principles of active learning and collaboration fall under the category of socio-constructivist learning theories (i.e., constructivism by Piaget and socio-culturalism by Vygotsky). Even though current chatbot solutions do function as teaching assistants and tutors in the classrooms [13, 18], yet the framing of those solutions within an appropriate pedagogy seems to be still scant and little research has been conducted with this focus [57]. The need to design chatbots to support such pedagogical uses, with implications in pedagogical contexts, is imperative. This requires an understanding of learning as an active process on the behalf of learners, whose prior knowledge and the interaction with the social environment has a catalyst role to play in their knowledge construction.

5.2 Benefits and Challenges of the Proposed Pedagogical Uses (RQ2)

The realization and potential success of these pedagogical uses is associated with their technological affordances and constraints. First, a human-like interaction with chatbots, as described by some of our participants, can be considered appropriate for online teaching and learning, especially in the post-covid era that we are experiencing nowadays. Further, among the advantages that chatbots bring into education, is their low development cost, as acknowledged by our participants, consistent with what is already proposed by previous scholars [56]. However, interviewees in this study also argued about the high costs of maintaining the use of chatbots, which might have implications for the widespread use of chatbots by educational organizations. More advantages include the cognitive offload of the users (e.g., bots responding to FAQs), chatbots’ availability to interact with the users at any time of the day, and their potential to interact proactively with the users. Consistent with previous authors, this proactive role can be achieved because chatbots can initiate a ‘dialogue’ with the user and adjust the content of the communication appropriately, considering the user’s location or clickstreams, making the user feel that s/he is personally addressed [24].

The appropriateness of chatbots for large-scale classes has been acknowledged by our participants. One would think that chatbots, as applied in education, can be a solution to the inadequate individual support that students receive in large-scale courses and/or MOOCs [23], with no further financial and organizational costs for the providers. In sum, chatbots can potentially provide essential individual student support especially in large scale classrooms, in which the provision of individualized feedback and support is demanding for educators. Moreover, the interactions among chatbots and the users can be automatically analyzed (e.g., by employing sentiment analysis as a proxy to measure users’ satisfaction, as proposed by Feine, Morana, and Gnewuch, [15]), thus providing another advantage of using chatbots in education. Analysis of this nature can be used to understand the users’ requirements and therefore improve the service or product that the chatbot is serving.

Despite all these advantages, there are several constraints that hinder the maximum performance of educational chatbots. For instance, even though chatbots can evolve via machine learning techniques and through evaluating conversations with users, the error rate at which a chatbot works is initially high [36]. Also, according to the same authors, even though chatbots can simplify the administrative work of educators by disclosing supplementary information to students about their courses, they often fail to solve content issues. This becomes a bit problematic when chatbots are meant to be used for meaningful formative assessment purposes and content-related guidance provision. Let alone, as indicated by the experts in our study, when chatbots do not comprehend the users’ requests and questions, even in administrative matters, they could cause frustration originating from ineffective communication.

Shortcomings in the use of chatbots include the usual disconnect between the vision of what AI powered chatbots or intelligent tutoring systems could be, and what they really are [5]. This could be attributed to the approaches used in practice, which are mainly simple. Also, research innovations in the field, often do not get integrated into the systems deployed at scale; that is, systems being used at scale in education are generally not representative of the full richness that research systems demonstrate. Therefore, even though there is an initial intend from researchers to develop systems that can use reinforcement learning to improve themselves [6], few systems incorporate this capacity [5].

Finally, another major challenge deals with ethical considerations on the application of chatbots in education. As with all novel technologies, chatbots also entail ethical and privacy implications. Besides, chatbots comprise AI entities and, therefore, they must be subject to the ethical standards applicable to AI [12]. As derived from our data, especially in the field of education, chatbots may be hindered by ethical considerations in relation to the actual educational intervention, but also the type of interactions that take place, and the way the conversation data are being further used and exploited. Those concerns are expected to be encountered in chatbots’ integration in an educational context, and thus, appropriate attention should be given to each potential chatbot use, according to the purpose that the chatbot serves each time.

6 Implications and Limitations

The findings of this study can have implications for researchers and educators in higher education, but also, software designers and developers in the field of chatbots. The proposed pedagogical uses of chatbots that can be exploited for the creation of specific pedagogical scenarios, accounting also for the added value of a particular chatbot use, the learning context, users’ characteristics, and particular needs, chatbot technology, and pedagogies that are deemed appropriate in each scenario.

Finally, in this work, the participants (learners, instructors, and one of the four experts) came from a single country, as they were reached by convenience. In future work, the authors should aim to have participants from a wider region and other cultures. Our results, although not generalizable, offer an in-depth analysis of current and future opportunities and challenges related to the use of chatbots in higher education. This study contributes new knowledge in the area, and we hope it will spark interest, research, and development in the field of educational chatbots in higher education.