Pedagogical AI conversational agents in higher education: a conceptual framework and survey of the state of the art

May combine two claims.

Within the simulate and experience label, all the conversational agents are used for face-to-face educational programmes, with the aim of providing context to education for an improved student experience.

“Within the simulate and experience label, all the conversational agents are used for face-to-face educational programmes, with the aim of providing context to education for an improved student experience.”

Classified as probably not a substantial claim.

The pedagogic section of the framework presents three tiers of categorisation for the pedagogical functions of conversational agents.

“The pedagogic section of the framework presents three tiers of categorisation for the pedagogical functions of conversational agents.”

• AssumptionRests on an assumption: The 92 studies identified by the search strategy are representative of the state of the art of pedagogical AI conversational agents in higher education.

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

The first and highest level of categorisation is embodiment, in which every AI conversational agent is labelled as either embodied (some degree of visual or humanistic form) or disembodied (command line or text based, without visual form).

“The first and highest level of categorisation is embodiment, in which every AI conversational agent is labelled as either embodied, wherein the agent has some degree of visual or humanistic form, or disembodied, in which the agent is command line or text based, without visual form.”

• AssumptionRests on an assumption: The 92 studies identified by the search strategy are representative of the state of the art of pedagogical AI conversational agents in higher education.

May combine two claims.

The studies labelled under reflective skills and metacognitive tend to also be applied to programmes categorised as distance-education, however some studies in this domain are also applied in face-to-face contexts.

“The studies labelled under reflective skills and metacognitive (Bhartiya et al., 2019;Chang et al., 2022;Xiao et al., 2019;Yılmaz et al., 2018) tend to also be applied to programmes categorised as distance-education, however some studies in this domain are also applied in face-to-face contexts.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Studies within assessment domain tend to vary in agent type and mode of study, and is not as explored as other categorisations within the framework hence a smaller sample.

“Studies within this domain tend to vary in agent type and mode of study, and is not as explored as other categorisations within the framework hence a smaller sample.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

AI conversational agents categorised within the pastoral domain aid educational administration tasks to reduce staff workloads and increase efficiency, or build motivation and/or self-efficacy of students.

“The AI conversational agents categorised within the pastoral domain aid educational administration tasks to reduce staff workloads and increase efficiency, or build motivation and/or self-efficacy of students.”

• AssumptionRests on an assumption: The 92 studies identified by the search strategy are representative of the state of the art of pedagogical AI conversational agents in higher education.

May combine two claims.

Agents in mentoring and coaching can be used in both distance-education (n = 12/26, 46%) and face-to-face education (n = 14/26, 54%).

“agents can be used in both distance-education (n = 12/26, 46%) and face-to-face education (n = 14/26, 54%)”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Studies of AI conversational agents used for instruction aid teaching and learning through instruction or via building on knowledge of core competencies.

“Studies of AI conversational agents used for instruction aid teaching and learning through instruction or via building on knowledge of core competencies.”

• AssumptionRests on an assumption: The 92 studies identified by the search strategy are representative of the state of the art of pedagogical AI conversational agents in higher education.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The second level of categorisation, functional type, relates to the specific type of AI conversational agent design in terms of its aesthetic interface, which can be virtual human (anthropomorphic representation), avatar (embodied creature but not completely anthropomorphic or human-like), chatbot (text based communicational agent), or voice bot (audio based agent, allowing only verbal communi

“The second level of categorisation identified as functional type relates to the specific type of AI conversational agent design, in terms of its aesthetic interface, which can be virtual human which is an anthropomorphic representation, avatar which is an embodied creature but not completely anthropomorphic or human-like, chatbot which is a text based communicational agent, or voice bot which is”

• AssumptionRests on an assumption: The 92 studies identified by the search strategy are representative of the state of the art of pedagogical AI conversational agents in higher education.

May combine two claims.

Within the FAQ's and knowledge base category, all conversational agents aim to increase student participation by answering student queries, and are used in both distance education and face-to-face education.

“Within the FAQ's and knowledge base category, all conversational agents aim to increase student participation by answering student queries, and are used in both distanceeducation and face-to-face education”

Probably not a substantial claim: may not pass S3 (load-bearing).

The aim of the assessment agent is to provide feedback to learners.

“The aim of the agent is to provide feedback to learners”

May combine two claims.

From the studies reviewed, around half of the conversational agents were used in a distance-education programme delivery and the other half were used in a face-to-face programme delivery.

“From the studies reviewed, around half of the conversational agents were used in a distance-education programme delivery and the other half were used in a face-to-face programme delivery.”

• AssumptionRests on an assumption: The 92 studies identified by the search strategy are representative of the state of the art of pedagogical AI conversational agents in higher education.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Cognitive AI conversational agents build the metacognitive functions of students by promoting creative thinking, facilitating experiences or reducing their psychological natural responses.

“Finally, the cognitive AI conversational agents build the metacognitive functions of students by promoting creative thinking, facilitating experiences or reducing their psychological natural responses.”

• AssumptionRests on an assumption: The 92 studies identified by the search strategy are representative of the state of the art of pedagogical AI conversational agents in higher education.

Classified as probably not a substantial claim.

The characteristics of AI conversational agents were organised into two overarching themes: pedagogical applications and technological functions.

“The characteristics of AI conversational agents were organised into two overarching themes, namely pedagogical applications and technological functions of AI conversational agents.”

• AssumptionRests on an assumption: The 92 studies identified by the search strategy are representative of the state of the art of pedagogical AI conversational agents in higher education.

May combine two claims.

Most studies within the simulate and experience label (instructional) are used in healthcare-based education programmes, and are medical technology (MedTech) based research studies.

“Most studies within this label are used in healthcare-based education programmes, and are medical technology (MedTech) based research studies.”

Classified as probably not a substantial claim. The quote only partly states this claim.

The third and lowest level of categorisation, conversational agent intents, includes eight sections: FAQ's & Knowledge Base, Motivate & Inspire, Admin & Management, Communication, Reflective Skills & Metacognitive, Simulate & Experience, Assessments, and Mentoring & Coaching.

“(5) Reflective Skills & Metacognitive where the conversational agent promotes analytical, divergent, critical and/or creative thinking in learners' to aid learning, (6) Simulate & Experience where the conversational agent provides a platform to stimulate a particular context or role and/or the agent to interact with human emotion to some degree, for example as a patient in healthcare education, (7”

• AssumptionRests on an assumption: The 92 studies identified by the search strategy are representative of the state of the art of pedagogical AI conversational agents in higher education.

Classified as probably not a substantial claim.

The first and highest level of categorisation, pedagogical purposes, is an overarching primary function of pedagogical application, stated as instructional, pastoral and cognitive.

“The first and highest level of categorisation identified as pedagogical purposes is an overarching primary function of pedagogical application, stated as instructional, pastoral and cognitive.”

• AssumptionRests on an assumption: The 92 studies identified by the search strategy are representative of the state of the art of pedagogical AI conversational agents in higher education.

The quote only partly states this claim.

The majority of cognitive agents tend to have intents focused on the simulate and experience category.

“Regarding conversational agent intents, the majority agents tend to have intents focused on the simulate and experience category”

May combine two claims.

The conversational agents within the high-level instructional category labelled as simulate and experience are also used for face-to-face educational programmes, however they are used for an instructional teaching and learning purpose to build student knowledge, as opposed to the cognitive high-level category wherein they are assisting the cognitive functions of students, such as psychological.

“Similar to the cognitive high-level category, the conversational agents within the highlevel instructional category labelled as simulate and experience are also used for faceto-face educational programmes, however they are used for an instructional teaching and learning purpose to build student knowledge, as opposed to the cognitive high-level category wherein they are assisting the cognitive”

Probably not a substantial claim: may not pass S3 (load-bearing).

The third and lowest level of categorisation, features, indicates the implementation of a specific type of technology, including five sections: Machine Learning, Social/IM Link, External Data Source, Natural Language Processing (API or Native/Natural), and Custom/Unspecified Development.

“This layer of categorisation includes five sections; (1) Machine Learning, wherein the AI conversational agent has additional learning capabilities such as eye tracking, emotion sensors, gaze detection, biometrics etc. to make the conversation more intelligent, emotionally, or otherwise, (2) Social/IM Link which uses a social media or instant messaging (IM) platform as the primary user interface,”

• Also stated in: § Technological functions · plain

  “(4) Natural Language Processing (NLP) for understanding human language, which can have two sub-categories, firstly Application Programming Interface (API) wherein the agent connects to an external NLP system such as Facebook (Wit.ai),Python libraries, IBM Watson etc. and secondly Native/Natural NLP wherein the agent uses NLP using built-in libraries or source code, and lastly (5) Custom/Unspec”

• AssumptionRests on an assumption: The 92 studies identified by the search strategy are representative of the state of the art of pedagogical AI conversational agents in higher education.

May combine two claims.

In terms of conversational agent intents, the largest category to be explored by researchers is Motivate & Inspire label, wherein most conversational agents (n = 13/17, 76%) are used within face-to-face education programmes while some (n = 4/17, 23%) are used in distance-learning programmes.

“In terms of the conversational agent intents, the largest category to be explored by researchers is Motivate & Inspire label, wherein most conversational agents (n = 13/17, 76%) are used within face-to-face education programmes while some (n = 4/17, 23%) are used in distance-learning programmes”

May combine two claims.

Face-to-face agents are generally guided by an instructor or facilitator, tend to be more diverse in embodiment and functional type, and include chatbots, virtual humans or avatars, and voice-bots.

“The agents used for face-to-face education are generally guided by an instructor or facilitator and therefore tend to be more diverse in their embodiment and functional type, and accordingly this category can include chatbots”

May combine two claims.

The conversational agents within the high-level Pastoral category labelled as Communication are also used for distance-learning programmes, however they are used for pastoral support purposes to aid the learning process by engaging students, as opposed to the Cognitive high-level category wherein they are used as a tool for interaction between students.

“Similar to the Cognitive high-level category, the conversational agents within the high-level Pastoral category labelled as Communication are also used for distancelearning programmes, however they are used for pastoral support purposes to aid the learning process by engaging students, as opposed to the Cognitive high-level category wherein they are used as a tool for interaction between students”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Each study acknowledges that agents can provide emotional support and foster a positive learning environment.

“Each study acknowledges that agents can provide emotional support and foster a positive learning environment.”

May combine two claims.

The most common functional type of pedagogical conversational agent is the chatbot, which is mostly disembodied with text-based inputs and outputs.

“The most common functional type of agent is the chatbot, which is sometimes embodied (Scholten et al., 2019;Xie et al., 2021) but mostly presents as a disembodied agent, having a text-based inputs and outputs.”

May combine two claims.

The conversational agents within the high-level Pastoral category labelled as FAQ & Knowledge Base are mostly used for distance-learning programmes (n = 9/10, 90%) to aid student learning and/or act as a student support mechanism, as opposed to the Instructional high-level category wherein they are used as a tool for increasing student participation by answering student queries.

“Furthermore, similar to the Instructional high-level category, the conversational agents within the high-level Pastoral category labelled as FAQ & Knowledge Base are mostly used for distance-learning programmes (n = 9/10, 90%) to aid student learning and/or act as a student support mechanism, as opposed to the Instructional high-level category wherein they are used as a tool for increasing student”

May combine two claims.

Most agents within the pastoral group are labelled as distance-education, which shows a contrasting overview to the instructional group, mainly since pastoral AI conversational agents supplement and support teaching and learning as opposed to direct in-class instruction.

“Concerning the mode of study categorisation, most agents within the pastoral group are labelled as distance-education, which shows a contrasting overview to the instructional group, mainly since pastoral AI conversational agents supplement and support teaching and learning as opposed to direct in-class instruction.”

May combine two claims. The quote only partly states this claim.

Distance-learning agents are generally used by students in their own time and include mostly disembodied chatbots, with some embodied virtual humans.

“The AI conversational agents used for distance-learning are generally used by students in their own time (non-scheduled) and include mostly disembodied chatbots”

Probably not a substantial claim: may not pass S3 (load-bearing).

Across these studies, a common theme emerges regarding the cognitive load management facilitated by agents.

“Across these studies, a common theme emerges regarding the cognitive load management facilitated by agents.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Social/IM link agents tend to be disembodied chatbots that use platforms such as Telegram, WhatsApp and Facebook messenger.

“Social/IM link tend to be disembodied chatbots that use platforms such as Telegram, WhatsApp and Facebook messenger”

Probably not a substantial claim: may not pass S4 (based).

Studies in educational technology tend to focus on either the pedagogical impact of their agents or the technological advancement of their agent.

“However, studies within this remit tend to focus on either the pedagogical impact of their agents, or the technological advancement of their agent”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Virtual human conversational agents are always embodied agents and span over all three pedagogical purposes (cognitive, instructional and pastoral).

“Virtual human conversational agents are always embodied agents and span over all three pedagogical purposes (cognitive, instructional and pastoral).”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Kim and Baylor (2006) categorise roles statically, while Tao et al. (2008) advocate for a more flexible, dynamic role adaptation.

“However, while Kim and Baylor (2006) categorise roles statically, Tao et al. (2008) advocate for a more flexible, dynamic role adaptation.”

May combine two claims.

The voice bot is the least popular functional type of pedagogical conversational agent and is only used within the cognitive pedagogical purpose.

“The last and least-popular functional type of pedagogical conversational agent is the voice bot, which is only used within the cognitive pedagogical purpose (Morris & Chen, 2020, 2021).”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The agents within the Motivate & Inspire category tend to have anthropomorphic aesthetic characteristics.

“The agents within this category tend to have anthropomorphic aesthetic characteristics.”

May combine two claims.

There is a lack of focus on cognitive agents in general and instructional agents specifically focusing on assessment.

“There appears to be a lack of focus on cognitive agents in general and instructional agents specifically focusing on assessment.”

May combine two claims.

From the surveyed pedagogical conversational agents, some used native NLP libraries, but many linked to common and sometimes open-source NLP systems through API connections.

“From the pedagogical conversational agents surveyed, some used native NLP libraries such as the study by Hsu et al. (2021) that uses a custom Python library, but many agents linked to common and sometimes opensource NLP systems through API connections to services such as Google's DialogFlow (Gonda & Chu, 2019), Juji.ai (Gupta & Chen, 2022), NLP.js (Winkler et al., 2020), IBM's Watson (Gonda et al.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The highest-level categorisation of the technological functions is the cyber-physical and visual embodiment of the conversational agents which, based on the thematic analysis conducted, has been classified as embodied and disembodied.

“The highest-level categorisation of the technological functions is the cyber-physical and visual embodiment of the conversational agents which, based on the thematic analysis conducted, has been classified as embodied and disembodied.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Every pedagogical agent analysed within the remit of this survey can be considered within the construct of embodiment.

“Every pedagogical agent analysed within the remit of this survey can be considered within the construct of embodiment.”

May combine two claims.

The majority of AI conversational agents exploit NLP capabilities and have provisions for connecting to various data sources.

“In reference to the technological functions, the majority of AI conversational agents exploit NLP capabilities to ensure users can effectively and easily communicate with the agents and have provisions for connecting to various data sources to draw meaningful conversations.”

Classified as probably not a substantial claim.

There is a lack of research that focuses on the use of LLMs for pedagogical purposes.

“Due to its recent inception, there is a lack of research that focuses on the use of LLMs for pedagogical purposes.”

Classified as probably not a substantial claim. May combine two claims.

They highlight the need for agents to adapt to individual learners' emotional and motivational needs, suggesting that personalised interactions can lead to better educational outcomes.

“They highlight the need for agents to adapt to individual learners' emotional and motivational needs, suggesting that personalised interactions can lead to better educational outcomes.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

All sources recognise that human-like gestures and facial expressions contribute to a more relatable and effective learning experience.

“All sources recognise that human-like gestures and facial expressions contribute to a more relatable and effective learning experience.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

All these studies recognise the importance of scalability in deploying pedagogical agents across various educational settings.

“All these studies recognise the importance of scalability in deploying pedagogical agents across various educational settings.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The review highlighted the diverse applications of these agents in various educational contexts.

“The review has highlighted the diverse applications of these agents in various educational contexts.”

Probably not a substantial claim: may not pass S3 (load-bearing).

When comparing these studies, several shared themes emerge regarding strategic implementation in educational contexts, such as scalability and task management.

“When comparing these studies, several shared themes emerge regarding strategic implementation in educational contexts, such as scalability and task management.”

Classified as probably not a substantial claim.

The conceptual framework presented in the study has two overarching themes: pedagogical applications and technological functions of pedagogical AI conversational agents.

“The conceptual framework presented in this study has two overarching themes which include the pedagogical applications, and technological functions of pedagogical AI conversational agents.”

May combine two claims.

The frameworks share a common goal of enhancing educational outcomes through the strategic use of conversational agents, yet they differ in their focus areas, ranging from psychological and social dynamics to cognitive processes and technological functions.

“The frameworks share a common goal of enhancing educational outcomes through the strategic use of conversational agents, yet they differ in their focus areas, ranging from psychological and social dynamics to cognitive processes and technological functions.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The study found many disembodied chatbots having NLP capabilities, which creates an intelligent agent able to identify users' intentions from verbal cues in natural language.

“this study has found many disembodied chatbots having NLP capabilities, which creates an intelligent agent, as the agent is able to identify users' intentions from verbal cues in natural language.”

May combine two claims. The quote only partly states this claim.

Of the 92 studies, 50 were conducted in face-to-face mode and 42 in distance education mode.

“Face-to-face | 50”

• Also stated in: § Conclusions · plain

  “Distance | 42”

May combine two claims. The quote only partly states this claim.

Of the 92 studies, the pedagogical purposes were cognitive (14), pastoral (37), and instructional (41).

“Cognitive | 14”

• Also stated in: § Conclusions · plain

  “Pastoral | 37”

• Also stated in: § Conclusions · plain

  “Instructional | 41”

May combine two claims.

The technological functions of AI conversational agents were found to be either embodied or disembodied, and to belong to one of four functional types (virtual human, avatar, chatbot, or voice bot), and to have one or more of five technological functions (machine learning, natural language processing, external data source, social and instant messaging link, or custom and/or unspecified development

“The technological functions of AI conversational agents were found to be either embodied or disembodied, and belong to one of four functional types; (1) virtual human, (2) avatar, (3) chatbot or (4) voice bot, and to have one (or more) of five technological functions; (1) machine learning, (2) natural language processing, 3) external data source, (4) social and instant messaging link or (5) custom”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Each framework addresses task coordination uniquely, contributing to an understanding of how agents support instructional goals.

“Each framework addresses task coordination uniquely, contributing to an understanding of how agents support instructional goals.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The conversational agents that exploit natural language processing systems also fall into one of two sub-types: native and natural or application programming interface systems.

“The conversational agents that exploit natural language processing systems also fall into one of two sub-types; native and natural or application programming interface systems.”

May combine two claims.

Pedagogical applications of AI conversational agents were found to belong to one of three pedagogical purposes (pastoral, instructional, or cognitive) and to specify into one of two modes of study (face-to-face or distance-education).

“Pedagogical applications of AI conversational agents were found to belong to one of three pedagogical purposes; (1) pastoral, (2) instructional or (3) cognitive, and specify into one of two modes of study, face-to-face or distance-education.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Studies of conversational agents used for Instruction aid teaching and learning through instruction or via building on knowledge of core competencies.

“Studies of conversational agents used for Instruction aid teaching and learning through instruction or via building on knowledge of core competencies.”

Classified as probably not a substantial claim. May combine two claims.

Anomalies and irregularities in the thematic analysis were rectified through validation, ensuring all studies in the literature dataset found a home in each thematic layer, arriving at a consensus pool of overarching themes and sub-themes.

“Any anomalies found (such as where a study would not be identified within a set a theme) or other irregularities and inconsistencies were rectified through validation, which ensured that all the studies in the literature dataset found a home in each of the thematic layers, finally arriving at a consensus pool of overarching themes and sub-themes that set the foundation on which the findings of”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Pedagogical applications and technological functions complement each other, as increased technological functions can enable a wider range of pedagogical applications, and more specified pedagogical applications often require more specialised technological functions.

“Both of these overarching themes complement each other, as increased technological functions can enable a wider range of pedagogical applications, and more specified pedagogical applications often require more specialised technological functions.”

• AssumptionRests on an assumption: The 92 studies identified by the search strategy are representative of the state of the art of pedagogical AI conversational agents in higher education.

Probably not a substantial claim: may not pass S3 (load-bearing).

The review process was partially conducted based on the conventions and guidelines applied by Petersen et al. (2008).

“The review process of this study is partially conducted based on the conventions and guidelines applied by Petersen et al. (2008).”

• cites Petersen et al. (2008)Not checked yet

  Not checked in the free run.

Classified as probably not a substantial claim.

The technological functions of pedagogical AI conversational agents can be considered in terms of embodiment (embodied/disembodied) and functional type and features.

“The technological functions of the agents are also considered in terms of embodiment (embodied/disembodied) and functional type and features.”

• AssumptionRests on an assumption: The 92 studies identified by the search strategy are representative of the state of the art of pedagogical AI conversational agents in higher education.

May combine two claims.

Conversational agents have been demonstrated to supplement and advance education in various forms, such as providing assessment of learners' work, increasing peer dialogue, providing a context for education (e.g., role-play in healthcare education), and answering student queries.

“Conversational agents have been demonstrated to supplement and advance education in various forms, such as providing assessment of learners work (Maryadi et al., 2017), increasing peer dialogue (Wang et al., 2020), providing a context for education, for example in coordinating role-play in healthcare education (Shorey et al., 2019), and answering student queries (Gonda & Chu, 2019).”

• cites Maryadi et al., 2017Not checked yet

  Not checked in the free run.

• cites Wang et al., 2020Not checked yet

  Not checked in the free run.

• cites Shorey et al., 2019 sourceNot checked yet

  Not checked in the free run.

Classified as probably not a substantial claim. May combine two claims.

Pedagogical AI conversational agents can primarily be considered in terms of their pedagogical applications and pedagogical purposes, which include pastoral, instructional and cognitive, and are further considered in terms of mode of study and intent.

“The conceptual framework proposes that pedagogical AI conversational agents can primarily be considered in terms of their pedagogical applications and their pedagogical purposes, which include pastoral, instructional and cognitive, and are further considered in terms of mode of study and intent.”

• AssumptionRests on an assumption: The 92 studies identified by the search strategy are representative of the state of the art of pedagogical AI conversational agents in higher education.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Due to the recent global Covid-19 pandemic in 2020, distance education has significantly increased in many higher educational institutes and has become the normal standard.

“According to Yarmak et al. (2021), due to the recent global Covid-19 pandemic in 2020, distance education has significantly increased in many higher educational institutes (HEI) and has become the normal standard”

• cites Yarmak et al. (2021)Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

The databases selected for the survey are the most relevant sources for computer-related studies.

“The databases selected for the survey are presented in Table 1, and were selected as they are the most relevant sources for computer-related studies (Dyba et al., 2007;Kitchenham & Charters, 2007).”

• cites Dyba et al., 2007Not checked yet

  Not checked in the free run.

• cites Kitchenham & Charters, 2007Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

There was only one coder involved in the thematic analysis, and therefore interrater reliability consideration was not required.

“There was only one coder involved in the process and therefore interrater reliability consideration was not required.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Pastoral agents aid educational administration tasks to reduce workloads and increase efficiency or build motivation and/or self-efficacy of students.

“The conversational agents categorised within the Pastoral domain aid educational administration tasks to reduce workloads and increase efficiency or build motivation and/or self-efficacy of students.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Although studies relate to aiding programmes in either of the two modes of delivery, the findings of the studies can be intermingled as the benefit of the experiments are not limited to the mode in which they are placed.

“Furthermore, although studies relate to aiding programmes in either of the two modes of delivery, the findings of the studies can be intermingled as the benefit of the experiments are not limited to the mode in which they are placed.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing). May combine two claims.

The selection, extraction, and analysis processes are not an exact science, and the results may have elements of subjective research.

“It must be noted that the selection, extraction, and analysis processes are not an exact science as there are no absolute rules, nor container in the proposed research domain to follow or adhere to, hence the results may have elements of subjective research.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Cognitive agents build the metacognitive functions of the students by promoting creative thinking, giving experiences or reducing the psychological natural responses.

“Cognitive agents build the metacognitive functions of the students by promoting creative thinking, giving experiences or reducing the psychological natural responses.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Marín (2014) introduces a social dimension, suggesting that emotional and social aspects are as crucial as cognitive support in the learning process.

“Marín (2014) introduces a social dimension, suggesting that emotional and social aspects are as crucial as cognitive support in the learning process.”

• cites Marín (2014)Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Tao et al. (2008) propose a framework that integrates psychological incentive mechanisms, suggesting that Expert, Mentor, and Motivator agents stimulate learners' internal motivation.

“Tao et al. (2008) propose a framework that integrates psychological incentive mechanisms, suggesting that Expert, Mentor, and Motivator agents stimulate learners' internal motivation.”

• cites Tao et al. (2008)Not checked yet

  Not checked in the free run.

May combine two claims.

Within developed countries, education-focused applications appear to be slower in adopting the latest advancements in digital transformation technologies and gaining investment in cutting-edge technologies compared to sectors such as finance and pharma.

“Within developed countries, and especially in a post-Covid United Kingdom, education focused applications appear to be slower in adopting the latest advancements in digital transformation technologies and gaining investment in cutting-edge technologies, while sectors such as finance and pharma appear to benefit more from public sector investment”

• cites Bell, 2023Not checked yet

  Not checked in the free run.

• cites Hacker & Dreifus, 2017Not checked yet

  Not checked in the free run.

• cites Johnson et al., 2023Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Zhao et al. (2014) emphasise agents' ability to adapt to learners' emotional states, noting that timely feedback can enhance learner motivation.

“Zhao et al. (2014) emphasise agents' ability to adapt to learners' emotional states, noting that timely feedback can enhance learner motivation.”

• cites Zhao et al. (2014)Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

Their contribution lies in emphasising scalability, allowing pedagogical agents to operate across various educational contexts and disciplines.

“Their contribution lies in emphasising scalability, allowing pedagogical agents to operate across various educational contexts and disciplines.”

• cites Nakano et al. (2008)Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Disembodied conversational agents generally lack bodily form and thus, are centred on a text-based graphical user interface.

“Disembodied conversational agents generally lack bodily form and thus, are centred on a text-based graphical user interface.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Marín (2014) emphasises agents' roles in fostering social interaction and emotional support, suggesting that virtual peer interactions contribute to empathy and social equality among learners, thereby improving learning outcomes.

“Marín (2014) emphasises agents' roles in fostering social interaction and emotional support, suggesting that virtual peer interactions contribute to empathy and social equality among learners, thereby improving learning outcomes.”

• cites Marín (2014)Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Fakinlede et al. (2013) highlight agents as virtual companions, not only providing academic support but also fostering positive learning environments that mitigate isolation.

“Fakinlede et al. (2013) highlight agents as virtual companions, not only providing academic support but also fostering positive learning environments that mitigate isolation.”

• cites Fakinlede et al. (2013)Not checked yet

  Not checked in the free run.

May combine two claims.

Moreno (2004) focuses on cognitive processing, proposing that agents aid in organising learning materials and managing cognitive load.

“Moreno (2004) focuses on cognitive processing, proposing that agents aid in organising learning materials and managing cognitive load.”

• cites Moreno (2004)Not checked yet

  Not checked in the free run.

Classified as probably not a substantial claim.

This research has adopted the view stipulated by Ziemke (2023) which posits that embodied conversational agents have 'organismic' and physical attributes which make them anthropomorphic based on their visual aesthetic and/or ability of interaction with users.

“This research has adopted the view stipulated by Ziemke (2023) which posits that embodied conversational agents have”

• cites Ziemke (2023)Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Mayer and DaPra (2012) integrate social agency, arguing that agents can enhance comprehension through rapport-building, while Zhao et al. (2014) highlight the adaptability of agents in providing personalised support.

“Mayer and DaPra (2012) integrate social agency, arguing that agents can enhance comprehension through rapport-building, while Zhao et al. (2014) highlight the adaptability of agents in providing personalised support.”

• cites Mayer and DaPra (2012)Not checked yet

  Not checked in the free run.

• cites Zhao et al. (2014)Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Disembodied conversational agents are more common in pedagogical contexts due to being easier to develop, requiring less visual programming expertise and computational power for rendering media outputs.

“Such conversational agents are more common in pedagogical contexts due to being easier to develop, requiring less visual programming expertise and computational power for rendering media outputs.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Kim and Baylor (2006) classify pedagogical agents into three roles: Expert, Mentor, and Motivator.

“Kim and Baylor (2006) classify pedagogical agents into three roles: Expert, Mentor, and Motivator.”

• cites Kim and Baylor (2006)Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

The use of LLMs is likely to become much more common in light of new vendors developing AI conversational agents using this approach.

“The use of LLMs is likely to become much more common in light of new vendors developing AI conversational agents using this approach (Dilmegani, 2023).”

• cites Dilmegani, 2023Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

Lemon et al. (2002) highlights natural language processing capabilities as crucial for creating an immersive learning environment.

“Lemon et al. (2002) highlights natural language processing capabilities as crucial for creating an immersive learning environment.”

• cites Lemon et al. (2002)Not checked yet

  Not checked in the free run.

May combine two claims.

Komatani et al. (2006) emphasise multi-threaded dialogue management, showing how agents' ability to manage multiple conversation threads enhances user engagement.

“Komatani et al. (2006) emphasise multi-threaded dialogue management, showing how agents' ability to manage multiple conversation threads enhances user engagement.”

• cites Komatani et al. (2006)Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

Large language models (LLMs) are a rapidly emerging technology for the development of AI conversational agents.

“One rapidly emerging technology is the use of large language models (LLMs) for the development of AI conversational agents.”

May combine two claims.

Min et al. (2019) and Mayer and DaPra (2012) discuss embodiment, with findings indicating that human-like gestures improve interaction quality and foster social presence.

“Min et al. (2019) and Mayer and DaPra (2012) discuss embodiment, with findings indicating that human-like gestures improve interaction quality and foster social presence.”

• cites Min et al. (2019)Not checked yet

  Not checked in the free run.

• cites Mayer and DaPra (2012)Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

Tao et al. (2008) propose that agents should switch flexibly among roles such as knowledge provider, emotional stimulator, and organiser to better meet diverse student needs.

“Tao et al. (2008) propose that agents should switch flexibly among roles such as knowledge provider, emotional stimulator, and organiser to better meet diverse student needs, enhancing adaptability in learning contexts.”

• cites Tao et al. (2008)Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Voice bots have limitations such as lack of visual engagement and dependence on a singular human sense, which makes them less inclusive in terms of actual use.

“It is not explored due to the various and obvious limitations of the voice bot, such as lack of visual engagement and dependence on a singular human sense, which also makes it less inclusive in terms of actual use.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Students consistently need to reflect on their knowledge, methods of learning and revision techniques to build better memory retention, gain a better understanding of their subject matter, and avoid unconscious biases from directing their critical judgement through implicit assumptions.

“Students consistently need to reflect on their knowledge, methods of learning and revision techniques to build better memory retention, gain a better understanding of their subject matter, and avoid unconscious biases from directing their critical judgement through implicit assumptions (Brookfield, 2021).”

• cites Brookfield, 2021Not checked yet

  Not checked in the free run.

May combine two claims.

Lemon et al. (2002) focus on adaptable dialogue systems, and Marín (2014) provides structured role definitions across educational contexts.

“As previously discussed, Lemon et al. (2002) focus on adaptable dialogue systems, and Marín (2014) provides structured role definitions across educational contexts.”

• cites Lemon et al. (2002)Not checked yet

  Not checked in the free run.

• cites Marín (2014)Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Academic and faculty staff satisfaction in higher education is an area of interest for universities as it is linked to staff effectiveness and retainment.

“Academic and faculty staff satisfaction in higher education is an area of interest for universities as it is linked to staff effectiveness and retainment (Gillespie et al., 2001).”

• cites Gillespie et al., 2001 sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Reflections are an important tool for the teacher or instructor, as reflective practice allows progress in pedagogical ability by reflecting on sessions and identifying areas of weakness and implementing perpetual improvements.

“reflections are also an important tool for the teacher or instructor, as reflective practice allows progress in pedagogical ability, by reflecting on sessions and identifying areas of weakness and implementing perpetual improvements, consistently (Tay et al., 2023).”

• cites Tay et al., 2023Not checked yet

  Not checked in the free run.

May combine two claims.

The literature review and conceptual framework demonstrated the growing trend and potential benefits of using AI conversational agents for pedagogical purposes in higher education.

“this literature review and conceptual framework has demonstrated the growing trend and potential benefits of using AI conversational agents for pedagogical purposes in higher education.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Student satisfaction is a key performance indicator for universities, and in many countries such as the UK, institutions have a legal responsibility to declare student satisfaction results to regulatory bodies, and student satisfaction also affects legalities related to student-university relationship in contract law.

“Student satisfaction is a key performance indicator for universities, and in many countries such as the UK, institutions have a legal responsibility to declare student satisfaction results to regulatory bodies, and student satisfaction also affects legalities related to student-university relationship in contract law (Gaffney-Rhys & Jones, 2010).”

• cites Gaffney-Rhys & Jones, 2010Not checked yet

  Not checked in the free run.

May combine two claims.

In many higher education institutes, teaching and faculty staff employ considerable time on assessment and feedback but have very little professional development in how to design assessments.

“In many higher education institutes, teaching and faculty staff employ considerable time on assessment and feedback,”

• cites McConlogue, 2020Not checked yet

  Not checked in the free run.

Classified as probably not a substantial claim.

Reflection is a crucial aspect of learning, especially in higher education.

“Another crucial aspect of learning is reflection, especially in higher education (Harvey et al., 2016;Moon, 2004).”

• cites Harvey et al., 2016 sourceNot checked yet

  Not checked in the free run.

• cites Moon, 2004Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

Assessment processes are one of the most cited factors linked to causing student stress and anxiety in higher education.

“assessment processes are one of the most cited factors linked to causing student stress and anxiety in higher education (McConlogue, 2020).”

• cites McConlogue, 2020Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

The conceptual framework developed in this study provides a comprehensive understanding of the key components and design principles of pedagogical AI conversational agents and offers a useful framework for future research and development in this field.

“The conceptual framework developed in this study provides a comprehensive understanding of the key components and design principles of pedagogical AI conversational agents and offers a useful framework for future research and development in this field.”

Classified as probably not a substantial claim.

The conceptual framework developed in the current study distinguishes itself from frameworks in the wider literature by offering a comprehensive approach that integrates both pedagogical applications and technological considerations, making it particularly suited to the complexities of higher education.

“The conceptual framework developed in the current study distinguishes itself from frameworks presented in the wider literature to date by offering a comprehensive approach that integrates both pedagogical applications and technological considerations, making it particularly suited to the complexities of higher education.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

A cause of stress and burnout is not necessarily having a high number of students, but rather the lack of management tools available to handle a high number of students.

“When analysed, it is apparent that a cause of stress and burnout is not necessarily having a high number of students, but rather the lack of management tools available to handle a high number of students.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

A major factor in faculty staff dissatisfaction is occupational stress, a primary cause of which is exposure to high numbers of students, especially tuition of postgraduates, which strongly predicts the experience of burnout.

“A major factor in faculty staff dissatisfaction is occupational stress, a primary cause of which is”

• cites Watts & Robertson, 2011Not checked yet

  Not checked in the free run.

Classified as probably not a substantial claim.

The conceptual framework developed in the current study builds upon and extends existing frameworks by providing a comprehensive categorisation of pedagogical conversational agents that integrates both pedagogical applications and technological functions.

“the conceptual framework developed in the current study builds upon, and extends existing frameworks, by providing a comprehensive categorisation of pedagogical conversational agents that integrates both pedagogical applications and technological functions.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Many pedagogical instructional theories are based on the process of reflection, for instance Experiential Learning Theory which posits that learning happens through experimentation, reflection, thought and action.

“Many pedagogical instructional theories are based on the process of reflection, for instance Experiential Learning Theory which posits that learning happens through experimentation, reflection, thought and action (Kolb, 1984).”

• cites Kolb, 1984Not checked yet

  Not checked in the free run.