Pith. sign in

REVIEW 5 major objections 6 minor 134 references

This paper proposes the first framework for deciding when, what, and how AI digital teachers should be used in online learning.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-03 08:16 UTC pith:CIBYS35H

load-bearing objection Useful When/What/How organizing framework for digital-teacher design, but the 'actionable' claim outruns the evidence: learner state isn't operationalized, validation is self-referential, and survey numbers don't add up. the 5 major comments →

arxiv 2601.17434 v2 pith:CIBYS35H submitted 2026-01-24 cs.HC

Co-Designing Digital Humans for Online Learning: A Framework for Human-AI Pedagogical Integration

Xiaokang Lei , Ching Christie Pang , Yuyang Jiang , Xin Tong , Pan Hui This is my paper
classification cs.HC
keywords digital teachersonline learningMOOCshuman-AI integrationco-designlearner preferencesinstructional designAI in education
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Digital teachers—AI avatars that deliver instruction in online courses—are being deployed faster than design knowledge about them. This paper aims to fix that by building a framework that tells course designers when a digital teacher should step in, what content it should teach, and how it should look and behave. The authors combine a review of 87 prior works, a survey of 132 MOOC learners, and co-design workshops with 18 pedagogy, design, and AI experts. The framework centers on a 'when–what–how' design space and seven key factors—such as learner state and course design—that are meant to drive every integration decision. The paper claims this is the first framework tailored specifically to digital teachers in online education, while acknowledging it has not yet been validated in real deployment scenarios.

Core claim

On its own terms, the paper establishes a design framework for integrating digital teachers into online learning, structured along three dimensions: when to use them (availability and delivery triggers), what to teach (six content types grounded in established educational taxonomies), and how to design them (modality and paradigm). The framework maps seven key factors—digital teacher technology, learner state, course design, contextual information, system goal, learner goal, and learner profile—onto these dimensions to produce actionable guidance. It is derived from a design-space analysis of 87 published works, a 132-learner survey, and iterative co-design workshops with 18 experts. The pap

What carries the argument

The central mechanism is the when–what–how design space combined with key-factor mapping. The 'when' dimension distinguishes availability from manual/automatic delivery triggers; the 'what' dimension defines six content categories (orientation, conceptual, procedural, affective, inquiry-based, evaluative); the 'how' dimension specifies visual and audio modalities plus design paradigms. A closed feedback loop—learner feedback feeding back into content and behavior—makes the framework self-iterative. Together these elements turn a taxonomy into a decision-support tool for designers.

Load-bearing premise

The framework's actionability rests on the assumption that the 18 experts and 132 learners who shaped it represent the wider MOOC population, and that expert endorsement during co-design is a valid test of effectiveness; the paper itself concedes it lacks empirical validation in specific scenarios and operational guidance for abstract factors like learner state.

What would settle it

A randomized controlled trial comparing a real MOOC designed with this framework against a conventionally designed version of the same course, measuring learning gains, completion rates, and satisfaction; if the framework-guided course shows no advantage, the claim of actionability is undercut.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the framework is right, course designers can systematically decide whether a digital teacher should teach, assist, or evaluate at a given moment, rather than treating the avatar as a uniform content-delivery tool.
  • The survey results imply that learners want digital teachers to be available on demand (manual trigger) but also appreciate proactive, just-in-time assistance when their state indicates confusion or disengagement.
  • The framework suggests that conceptual, procedural, and evaluative content are well suited to digital teachers, while inquiry-based tasks may remain better with human instructors—a division of labor that could shape curriculum planning.
  • The closed-loop feedback mechanism implies digital teachers should be designed to improve continuously from learner input, not deployed as static artifacts.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • One testable extension: the when–what–how structure could generalize beyond MOOCs to K-12 or corporate training, where the same three questions arise and the key factors (learner state, course design, context) still apply.
  • The 'privacy advantage' that experts identified suggests digital teachers may be particularly valuable for sensitive topics like mental health counseling—an application the framework's affective-content category leaves implicit.
  • If learner state is to be a driving factor, the framework will need operational definitions, as the authors concede; a natural next step is mapping abstract states like confusion or boredom to measurable signals such as quiz scores, response times, or facial expressions.
  • The framework's actionability could be tested by building two versions of the same MOOC—one designed with the framework, one without—and comparing completion and satisfaction, which the paper has not done.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

5 major / 6 minor

Summary. The paper proposes a framework for integrating digital teacher avatars into online learning, organized around three design questions—when to use digital teachers, what content to teach, and how to design the interaction—and a set of key factors (learner state, learner profile, course design, contextual information, system goal, digital teacher technology). The framework is derived from a design-space analysis of 87 papers, a survey of 132 learners, and three co-design workshops with 18 experts. The authors claim this is the first framework tailored specifically to digital teachers in online education and that it provides actionable guidance for educators, designers, and researchers.

Significance. If the framework is taken as a structured synthesis of existing knowledge, it is a useful contribution: it organizes scattered literature on pedagogical agents, AI tutors, and multimodal learning into a coherent vocabulary, and it incorporates stakeholder input from learners and experts. The paper's explicit acknowledgment of limitations and future directions is also a strength. The central claim, however, goes beyond synthesis to actionability: the manuscript states that the framework 'provides a structured approach for designers and a valuable reference for researchers.' That claim is currently under-supported because the key construct of learner state is not operationalized, and the only usefulness evidence comes from co-design participants evaluating their own output. The paper is therefore best treated as a promising framework proposal that needs additional validation and specification before it can support prescriptive use.

major comments (5)
  1. [Section 3.2, Participant breakdown] The survey participant numbers are internally inconsistent. The text reports group 1 ('worked on AI products') as n=92, 8.3%, while 92/132 = 69.7%; group 2 is also n=92, 69.7%. The five group counts sum to 213, not 132, and the reported 0.03% for group 4 should be 3.0%. This appears to be a typographical error (likely group 1 should be n=11), but as written it undermines confidence in all survey descriptive statistics. The authors must correct the table/text and re-verify every percentage before publication.
  2. [Sections 4.1, 4.2, and 5.3, learner-state operationalization] The framework routes many design decisions through 'learner state'—e.g., automatic assistance triggered by 'learner confusion through context-aware multimedia,' 'attention monitoring via test results, physiological data,' and pacing adjustments 'via quizzes.' Yet Section 5.3 concedes that the framework 'lacks empirical validation in more specific scenarios and operational guidance for abstract factors like learner state.' Without a specification of what data sources, thresholds, or trigger-to-action mappings define learner state, a designer cannot apply the framework; it remains an analytic taxonomy rather than an actionable guide. This gap is load-bearing because the paper's central contribution is actionability. The authors should either provide concrete operational definitions for at least the most common cases, or explicitly revise the contribution to a conceptual framework rather th
  3. [Section 3.3 and Section 5.1, validation of usefulness] The only evidence that the framework is 'very useful' (quotes from E1, S1–S2, D1, D3, T3) comes from the expert workshop participants who had just helped refine the framework. This is a self-referential evaluation: the co-designers are endorsing a product they co-created. There is no independent application by designers who did not participate in framework development, and no outcome measure such as design quality, implementation feasibility, learner engagement, or learning gain. A small external usability test—e.g., asking fresh designers to apply the framework to a concrete MOOC scenario and comparing the resulting designs to a baseline—would substantially strengthen the claim. As it stands, the paper does not demonstrate that the framework is actionable beyond the group that built it.
  4. [Finding 4, Figure 3, comparative modality claims] The text states learners 'expressed a strong preference' for AR-based MOOCs (51.5%) over VR-based MOOCs (47.7%) and traditional MOOC videos (40.9%). These differences are presented without inferential statistics, error bars, or per-condition sample sizes. Because the percentages are top-2-box scores from the same overall sample, the 10.6 percentage-point gap between AR and video may or may not be meaningful; the reported values alone do not establish a 'strong' preference. This matters because the modality comparison feeds directly into the 'How to Design' dimension. The authors should either report appropriate statistical tests and confidence intervals or soften the claim to descriptive tendency.
  5. [Section 3.2, sampling and generalizability] The survey sample is a convenience sample, with the majority of respondents located in Guangzhou and Chongqing, and the expert workshop drew 18 participants from a single technology university. The paper uses these data to support general design recommendations for MOOCs, which serve a global and heterogeneous learner population. While this is a known limitation of many HCI studies, the manuscript should explicitly discuss how the sample composition might bias the preference findings (e.g., digital literacy, cultural attitudes toward AI avatars) and temper the generalizability claims in the Conclusion. This is not a fatal flaw, but it is part of the evidence supporting the framework's broad applicability.
minor comments (6)
  1. [Introduction, Section 1] The phrase 'three-phrase method' should be 'three-phase method.'
  2. [Section 5.3] Typo: 'Undoubtely' should be 'Undoubtedly.'
  3. [Figure 1] Figure 1 is extremely dense and is not systematically referenced in the text. Consider decomposing it into separate panels for 'When/What/How' and the key factors, and pointing readers to each panel at the relevant section.
  4. [Finding 1] The sentence 'While most participants have MOOC experience (n=94, 71.2%; satisfaction: 76.6%)' is ambiguous: it is unclear whether 76.6% is the percentage of the 94 MOOC users who were satisfied, or of all 132 participants. Please clarify the denominator.
  5. [Section 3.2, group percentages] Group 4's percentage is reported as 0.03%; with n=4/132 it should be 3.0%. Also, group 5 (n=0) could be omitted to avoid confusion.
  6. [Section 3.3] The expert participant age distribution is reported as M=25.41, SD=3.5, but the age range is not given. Including the range would help the reader assess the sample's representativeness of 'experts.'

Circularity Check

1 steps flagged

Framework is largely a literature synthesis; main circularity is that the 'very useful' validation comes from the same co-design experts who iteratively shaped the framework, with no independent application test.

specific steps
  1. other [Section 3.3 (Study 2: Co-design Workshops), Procedure & Results; also Contribution (4) in Section 1]
    "Based on literature and survey data, an initial framework was developed and subsequently refined through three stakeholder workshops (learners and experts). ... After each session, two researchers applied affinity diagramming [132] to analyze data and iteratively refine the framework for subsequent stages. ... Experts participants highly recommended the framework, describing it as 'very useful' (E1, S1-2, D1, D3, T3)."

    The 'very useful' judgment is collected from the same participants whose suggestions were incorporated into the framework during iterative refinement. Contribution (4) then states that the workshops 'demonstrated that the proposed framework provides valuable insights and support for educators,' so the cited evidence for the framework's value is the co-design process that produced the framework itself. The endorsement is self-referential: after affinity-diagramming changes driven by expert feedback (Recommendations 1–4), the experts were asked to rate a framework they had just helped shape. No independent designer or learner applied the framework, and no outcome measure was collected, so the validation reduces to the co-designers approving their own output.

full rationale

The framework itself is mostly a synthesis of external inputs rather than a fitted prediction. The when/what/how design space is grounded in design-space analysis and external pedagogical taxonomies (Bloom, Gagne), the key factors are drawn from 87 reviewed works, and the learner survey is independent empirical input. The admission in Section 5.3 that the framework 'lacks empirical validation in more specific scenarios and operational guidance for abstract factors like learner state' is a correctness/actionability limitation, not a circularity. The author self-citations ([9], [26]) support background and feasibility claims but are not the sole justification for any load-bearing conclusion, since external references accompany them. No quantitative prediction is made from fitted parameters and no uniqueness theorem is imported. The main circularity is therefore the self-referential expert validation used to support the claim that the framework is actionable and valuable; this raises the score to 4 but not higher, because the central taxonomy still has independent content from the literature and survey.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 1 invented entities

The framework is a hand-built taxonomy: content types, factor lists, and dimension splits are chosen from literature and workshop consensus, not derived from data. No numeric parameters are fitted, but the categorical choices are free parameters because different choices would alter the framework's recommendations. The axioms are standard pedagogical/diagnostic assumptions in this subfield, but they are not justified with evidence specific to digital teachers.

free parameters (5)
  • Six content types = Orientation, Conceptual, Procedural, Affective, Inquiry-based, Evaluative
    Chosen from Bloom's Taxonomy and Gagne's Nine Events (Section 3.1). These categories determine the framework's content mapping; different taxonomies would change the framework.
  • When-What-How design space = Three dimensions
    Imported from design space analysis literature (refs 58–60) and hand-mapped to key factors (Section 3.1). The mapping is a design choice, not derived from data.
  • Platform/Digital-Teacher dimension split = Two branches after workshops
    Experts added Guidance, Credentialing, and Networking (Section 3.3, Recommendation 1); this split is a hand-added categorical choice.
  • Modality priorities = Text and video prioritized over graphics; audio secondary
    Based on expert consensus (Section 4.3). This ranking is not independently validated with learners.
  • Top-2-box analysis thresholds = Likert 1–7; top-2-box = 6–7
    Used in Section 3.2 for findings; no significance tests or confidence intervals accompany these numeric thresholds.
axioms (5)
  • domain assumption Bloom's Taxonomy and Gagne's Nine Events are valid bases for classifying instructional content
    Invoked in Section 3.1 to define the six content types; the framework's 'what' dimension depends on this pedagogical classification.
  • domain assumption Learner state can be reliably sensed from multimodal data and acted upon in real time
    Assumed throughout Section 3.1 and Section 4.1 (e.g., automatic assistance based on confusion/attention detection). The paper provides no empirical proof of real-time sensing reliability.
  • domain assumption The convenience samples (132 learners, 18 experts) are representative of MOOC populations
    Section 3.2 participants are mostly students in Chinese cities; Section 3.3 experts come from a single technology university. Generalizability is assumed for the framework's claims.
  • domain assumption Expert co-design feedback is a valid measure of framework usefulness
    The workshops both refine the framework and provide the endorsement quoted in Section 5.1; this self-validation is treated as evidence of effectiveness.
  • domain assumption Digital teachers can improve learning outcomes over traditional methods
    Stated in the Introduction with cited references [3–5, 7]; the framework assumes this underlying benefit without a controlled test here.
invented entities (1)
  • Closed-loop learner feedback mechanism no independent evidence
    purpose: To continuously optimize digital teacher outputs using learner feedback (questionnaires, multimodal data), proposed in Section 3.3 Recommendation 4 and Section 4.3.
    This is a proposed design mechanism, not tested empirically; no falsifiable prediction outside the paper is provided.

pith-pipeline@v1.3.0-alltime-deepseek · 17469 in / 9591 out tokens · 97616 ms · 2026-08-03T08:16:35.109438+00:00 · methodology

0 comments
read the original abstract

Artificial intelligence (AI) and large language models (LLMs) are reshaping education, with virtual avatars emerging as digital teachers capable of enhancing engagement, sustaining attention, and addressing instructor shortages. Aligned with the Sustainable Development Goals (SDGs) for equitable quality education, these technologies hold promise yet lack clear guidelines for effective design and implementation in online learning. To fill this gap, we introduce a framework specifying when, what, and how digital teachers should be integrated. Our study combines (1) a design space analysis of 87 works across AI, educational technology, design, and HCI, (2) a survey of 132 learners' practices and preferences, and (3) three co-design workshops with 18 experts from pedagogy, design, and AI. It provides actionable guidance for educators, designers, and HCI researchers, advancing opportunities to build more engaging, equitable, and effective online learning environments powered by digital teachers.

Figures

Figures reproduced from arXiv: 2601.17434 by Ching Christie Pang, Pan Hui, Xiaokang Lei, Xin Tong, Yuyang Jiang.

Figure 1
Figure 1. Figure 1: An overview of our proposed framework. From left to right: an example [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Motivation and Perceived Effectiveness of Digital Teachers. 0 10 20 30 40 50 Top-2-Box % (ratings 6 7) MOOC video VR MOOC AR MOOC 40.9% 47.7% 51.5% Presentation Preferences Top-2-Box (6 7) [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 5
Figure 5. Figure 5: On-site photos of the workshops. (UX/UI designer) (n = 3), and learner (n = 6). All of them were familiar with AI and digital teacher concepts. For the study, they were randomly assigned to one of three balanced, interdisciplinary groups, each comprising one or two expert from every domain. Procedure Pre-workshop materials provided context on digital teachers for expert participants. Sessions utilized a la… view at source ↗
Figure 6
Figure 6. Figure 6: Affinity diagram analysis process for iterative framework development. [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

134 extracted references · 11 canonical work pages

  1. [1]

    International Marketing Review (2024)

    Oliveira, F.G., Belitski, M., Kakabadse, N., Theodorakopoulos, N.: Unveiling the potential of digital human avatars in modern marketing strategies. International Marketing Review (2024)

    2024

  2. [2]

    Weeraskara, U.: Digital Avatars: What is the Future? SSRN 5073872 (2024)

    2024

  3. [3]

    International Transactions on Education Technology (ITEE)3(1), 1–8 (2024)

    Yu, C., Yao, G., et al.: Enhancing student engagement with ai-driven personalized learning systems. International Transactions on Education Technology (ITEE)3(1), 1–8 (2024)

    2024

  4. [4]

    https://doi.org/10.14689/ejer.2024.110.003

    Basri, W.S.: Effectiveness of AI-powered tutoring systems in enhancing learn- ing outcomes.Eurasian Journal of Educational Research110(1), 33–48 (2024). https://doi.org/10.14689/ejer.2024.110.003

    work page doi:10.14689/ejer.2024.110.003 2024

  5. [5]

    In: Driving Quality Education Through AI and Data Science, pp

    Jothikumar, K., Anisha, S., Varshini, S.A., Meenakshi, A.J., Gomathi, V.: AI tu- tors and virtual classrooms: Revolutionizing distance learning. In: Driving Quality Education Through AI and Data Science, pp. 503–530. IGI Global, Hershey (2025)

    2025

  6. [6]

    https://www.un.org/sustainabledevelopment/inequality (2015), last accessed 09 Jan 2026

    United Nations: Sustainable development goals. https://www.un.org/sustainabledevelopment/inequality (2015), last accessed 09 Jan 2026

    2015

  7. [7]

    In: CHI ’25, Article No

    Tan, C.T., Atmosukarto, I., Tandianus, B., et al.: Exploring the impact of avatar representations in AI chatbot tutors on learning experiences. In: CHI ’25, Article No. 1072. ACM, New York (2025). https://doi.org/10.1145/3706598.3713456

    arXiv 2025

  8. [8]

    In: CHI ’25, Article No

    Fu, Y., et al.: Understanding children’s avatar making in social on- line games. In: CHI ’25, Article No. 136. ACM, New York (2025). https://doi.org/10.1145/3706598.3713262 18 X. Lei et al

    arXiv 2025

  9. [9]

    Pang, C.C., Zhao, Y., Yin, Z., et al.: Artificial human lecturers: Initial findings from Asia’s first AI lecturers in class to promote innovation in education. In: LCT

  10. [10]

    Interactive Learning Environments 30(1), 34–48 (2022)

    Du, X., Zhang, M., Shelton, B.E., Hung, J.L.: Learning anytime, anywhere: a spatio-temporal analysis for online learning. Interactive Learning Environments 30(1), 34–48 (2022). https://doi.org/10.1080/10494820.2019.1663161

    arXiv 2022

  11. [11]

    Educational Researcher 36(2), 96–102 (2007)

    Fletcher, J.D., Tobias, S., Wisher, R.A.: Learning anytime, anywhere: Advanced distributed learning and the changing face of education. Educational Researcher 36(2), 96–102 (2007)

    2007

  12. [12]

    The New York Times2(12), 2012 (2012)

    Pappano, L.: The Year of the MOOC. The New York Times2(12), 2012 (2012)

    2012

  13. [13]

    Sustainability15(6), 4782 (2023)

    Alenezi, M., Wardat, S., Akour, M.: The need of integrating digital education in higher education: Challenges and opportunities. Sustainability15(6), 4782 (2023). https://doi.org/10.3390/su15064782

    work page doi:10.3390/su15064782 2023

  14. [14]

    Working Papers on Ed- ucation Policy, vol

    Pedro, F., Subosa, M., Rivas, A., Valverde, P.: Artificial intelligence in education: Challenges and opportunities for sustainable development. Working Papers on Ed- ucation Policy, vol. 7. UNESCO, Paris (2019)

    2019

  15. [15]

    In: Ad- vances in Technological Innovations in Higher Education, pp

    Kaswan, K.S., Dhatterwal, J.S., Ojha, R.P.: AI in personalized learning. In: Ad- vances in Technological Innovations in Higher Education, pp. 103–117. CRC Press (2024)

    2024

  16. [16]

    IEEE Access10, 81323–81342 (2022)

    Murtaza, M., Ahmed, Y., Shamsi, J.A., Sherwani, F., Usman, M.: AI-based per- sonalized e-learning systems: Issues, challenges, and solutions. IEEE Access10, 81323–81342 (2022). https://doi.org/10.1109/ACCESS.2022.3195821

    arXiv 2022

  17. [17]

    GSC Advanced Research and Reviews18(2), 261–271 (2024)

    Ayeni, O.O., et al.: AI in education: A review of personalized learning and educa- tional technology. GSC Advanced Research and Reviews18(2), 261–271 (2024)

    2024

  18. [18]

    Lee, L.H., et al.: All one needs to know about Metaverse: A complete survey on technological singularity, virtual ecosystem, and research agenda. Found. Trends HCI18(2–3), 100–337 (2024). https://doi.org/10.1561/1100000095

    doi:10.1561/1100000095 2024

  19. [19]

    AI Magazine27(2), 96 (2006)

    Swartout, W.R., et al.: Toward virtual humans. AI Magazine27(2), 96 (2006). https://doi.org/10.1609/aimag.v27i2.1883

    work page doi:10.1609/aimag.v27i2.1883 2006

  20. [20]

    Schroeder, N.L., Craig, S.D.: Learning with virtual humans: In- troduction to the special issue. J. Res. Technol. Educ.53(1), 1–7 (2021).https://doi.org/10.1080/15391523.2020.1863114

    arXiv 2021

  21. [21]

    In: Handbook on So- cially Interactive Agents, vol

    Lane, H.C., Schroeder, N.L.: Pedagogical agents. In: Handbook on So- cially Interactive Agents, vol. 2, pp. 307–330. ACM, New York (2022). https://doi.org/10.1145/3563659.3563669

    arXiv 2022

  22. [22]

    https://doi.org/10.1016/j.edurev.2010.07.004

    Heidig, S., Clarebout, G.: Do pedagogical agents make a difference to stu- dent motivation and learning? Educational Research Review6(1), 27–54 (2011). https://doi.org/10.1016/j.edurev.2010.07.004

    doi:10.1016/j.edurev.2010.07.004 2011

  23. [23]

    In: Companion Proc

    Zhang, K., et al.: Intelligent interaction platform for personalized dig- ital tutor: Advancing empathetic and adaptive learning experiences. In: Companion Proc. WWW 2025, pp. 2947–2950. ACM, New York (2025). https://doi.org/10.1145/3701716.3715182

    arXiv 2025

  24. [24]

    Rudolph, J., Tan, S., Tan, S.: War of the chatbots: Bard, Bing Chat, ChatGPT, Ernie and beyond. J. Appl. Learn. Teach.6(1), 1–6 (2023). https://doi.org/10.37074/jalt.2023.6.1.23

    work page doi:10.37074/jalt.2023.6.1.23 2023

  25. [25]

    IEEE Trans

    Gao,B.,Yan,J.,Zhong,R.:Howdigitalteacherappearanceanthropomorphismim- pacts digital learning satisfaction. IEEE Trans. Learn. Technol.18, 438–457 (2025). https://doi.org/10.1109/TLT.2025.3560032 Co-Designing Digital Humans for Online Learning 19

    arXiv 2025

  26. [26]

    In: SIGGRAPH Asia 2024 Educator’s Forum, Article No

    Zhao, Z., Yin, Z., Sun, J., Hui, P.: Embodied AI-guided interactive digital teachers for education. In: SIGGRAPH Asia 2024 Educator’s Forum, Article No. 6. ACM, New York (2024). https://doi.org/10.1145/3680533.3697070

    arXiv 2024

  27. [27]

    In: CHI ’25, Article No

    Bodonhelyi, A., et al.: From passive watching to active learning: Empowering proactive participation in digital classrooms with AI video assistant. In: CHI ’25, Article No. 16. ACM, New York (2025). https://doi.org/10.1145/3706598.3713513

    arXiv 2025

  28. [28]

    Wang, A., et al.: Generative AI for medical education: Insights from a case study with medical students. In: Ext. Abstracts CHI 2025, Art. 303. ACM (2025). https://doi.org/10.1145/3706599.3721208

    arXiv 2025

  29. [29]

    In: In: CHI ’07, pp

    Johnsen, K., et al.: The validity of a virtual human experience for in- terpersonal skills education. In: In: CHI ’07, pp. 1049–1058. ACM (2007). https://doi.org/10.1145/1240624.1240784

    arXiv 2007

  30. [30]

    In: Proc

    Yang, S., et al.: Debugging with an AI tutor: Investigating novice help- seeking behaviors. In: Proc. ICER 2024, pp. 84–94. ACM, New York (2024). https://doi.org/10.1145/3632620.3671092

    arXiv 2024

  31. [31]

    In: Proc

    Baradari, D., et al.: NeuroChat: A neuroadaptive AI chatbot for customizing learning experiences. In: Proc. CUI 2025, Article No. 57. ACM, New York (2025). https://doi.org/10.1145/3719160.3736623

    arXiv 2025

  32. [32]

    Chen, S., et al.: An integrated model for predicting pupils’ acceptance of artifi- cially intelligent robots as teachers. Educ. Inf. Technol.28(9), 11631–11654 (2023). https://doi.org/10.1007/s10639-023-11601-2

    work page doi:10.1007/s10639-023-11601-2 2023

  33. [33]

    In: Proc

    Tu, S., et al.: LittleMu: Deploying an online virtual teaching assistant via hetero- geneous sources integration. In: Proc. CIKM 2023, pp. 4843–4849. ACM, New York (2023). https://doi.org/10.1145/3583780.3615484

    arXiv 2023

  34. [34]

    Hayashi, Y.: Explanation activities with a pedagogical agent in an online task. In: Ext. Abstracts CHI 2015, pp. 1457–1460. ACM, New York (2015). https://doi.org/10.1145/2702613.2732751

    arXiv 2015

  35. [35]

    In: Proc

    Xu, T., et al.: AI-driven virtual teacher for enhanced educational efficiency: Lever- aging large pretrain models. In: Proc. AAAI 2024. AAAI Press, Washington DC (2024)

    2024

  36. [36]

    In: CHI ’24, pp

    Han, A., et al.: Teachers, parents, and students’ perspectives on integrating gen- erative AI into elementary literacy. In: CHI ’24, pp. 1–17. ACM, New York (2024). https://doi.org/10.1145/3613904.3642438

    arXiv 2024

  37. [37]

    In: CHI ’25, Article No

    Zhu, Z., et al.: Exploring LLM-powered role and action-switching pedagogical agents for history education in VR. In: CHI ’25, Article No. 1066. ACM, New York (2025). https://doi.org/10.1145/3706598.3713109

    arXiv 2025

  38. [38]

    In: CHI ’21, Article No

    Petersen, G.B., Mottelson, A., Makransky, G.: Pedagogical agents in educational VR: An in the wild study. In: CHI ’21, Article No. 482. ACM, New York (2021). https://doi.org/10.1145/3411764.3445760

    arXiv 2021

  39. [39]

    Peng, Z., et al.: DesignQuizzer: A community-powered conversational agent for learning visual design. Proc. ACM HCI (CSCW)8(44), 1–40 (2024). https://doi.org/10.1145/3637321

    work page doi:10.1145/3637321 2024

  40. [40]

    In: Proc

    Fan, H., et al.: LessonPlanner: Assisting novice teachers to prepare pedagogy- driven lesson plans with LLMs. In: Proc. UIST 2024, Article No. 146. ACM, New York (2024). https://doi.org/10.1145/3654777.3676390

    arXiv 2024

  41. [41]

    In: Proc

    Chen, Y., et al.: Empowering private tutoring by chaining large lan- guage models. In: Proc. CIKM 2024, pp. 354–364. ACM, New York (2024). https://doi.org/10.1145/3627673.3679665

    arXiv 2024

  42. [42]

    In: CHI ’25, Article No

    Jin, H., et al.: TeachTune: Reviewing pedagogical agents against diverse student profiles. In: CHI ’25, Article No. 1073. ACM, New York (2025). https://doi.org/10.1145/3706598.3714054 20 X. Lei et al

    arXiv 2025

  43. [43]

    In: CHI ’25, Article No

    Rogers, K., et al.: Playing dumb to get smart: Evaluating an LLM-based teachable agent. In: CHI ’25, Article No. 126. ACM, New York (2025). https://doi.org/10.1145/3706598.3713644

    arXiv 2025

  44. [44]

    In: CHI ’24, Article No

    Jin, H., et al.: Teach AI how to code: Using large language models as teachable agents. In: CHI ’24, Article No. 652. ACM, New York (2024). https://doi.org/10.1145/3613904.3642349

    arXiv 2024

  45. [45]

    In: Proc

    Hellas, A., et al.: Experiences from integrating large language model chatbots into the classroom. In: Proc. SIGCSE Virtual 2024, pp. 46–52. ACM, New York (2024). https://doi.org/10.1145/3649165.3690101

    arXiv 2024

  46. [46]

    In: Proc

    Jetzinger, F., Baumer, S., Michaeli, T.: Artificial intelligence in compulsory K-12 Computer Science classrooms. In: Proc. SIGCSE 2024, pp. 590–596. ACM, New York (2024). https://doi.org/10.1145/3626252.3630782

    arXiv 2024

  47. [47]

    Cai, Z., et al.: Integrating LLM-based conversational AI in small group collabora- tive learning. In: Ext. Abstracts CHI 2024, Article No. 37. ACM, New York (2024). https://doi.org/10.1145/3613905.3650868

    arXiv 2024

  48. [48]

    Shochcho, M.I., et al.: Improving user engagement and learning outcomes in LLM- based Python tutor. In: Ext. Abstracts CHI 2025, Article No. 337. ACM, New York (2025). https://doi.org/10.1145/3706599.3720240

    arXiv 2025

  49. [49]

    Education Sciences12(5), 308 (2022)

    Amado, C., et al.: MOOCs design: A conceptual framework for contin- uous teacher training in Portugal. Education Sciences12(5), 308 (2022). https://doi.org/10.3390/educsci12050308

    work page doi:10.3390/educsci12050308 2022

  50. [50]

    JUCS23(9), 845–867 (2017)

    Kasch, J., et al.: A framework towards educational scalability of open online courses. JUCS23(9), 845–867 (2017). https://doi.org/10.3217/jucs-023-09-0845

    work page doi:10.3217/jucs-023-09-0845 2017

  51. [51]

    Educational Media International52(4), 239–252 (2015)

    Conole, G.: Designing effective MOOCs. Educational Media International52(4), 239–252 (2015). https://doi.org/10.1080/09523987.2015.1125989

    arXiv 2015

  52. [52]

    Educational Psychologist57(3), 178–191 (2022)

    Archambault, L., Leary, H., Rice, K.: Pillars of online pedagogy: A framework for teaching in online learning environments. Educational Psychologist57(3), 178–191 (2022). https://doi.org/10.1080/00461520.2022.2051513

    arXiv 2022

  53. [53]

    Medical Teacher46(10), 1362–1368 (2024)

    Saaiq, M., et al.: Digital teaching: Developing a structured digital teach- ing competency framework. Medical Teacher46(10), 1362–1368 (2024). https://doi.org/10.1080/0142159X.2024.2308782

    arXiv 2024

  54. [54]

    ReCALL18(1), 105–121 (2006)

    Hampel, R.: Rethinking task design for the digital age: A frame- work for language teaching. ReCALL18(1), 105–121 (2006). https://doi.org/10.1017/S0958344006000711

    work page doi:10.1017/s0958344006000711 2006

  55. [55]

    Pedagogies19(3), 327–336 (2024)

    Tan, S.C., et al.: Introduction to digital pedagogy: a proposed frame- work for design and enactment. Pedagogies19(3), 327–336 (2024). https://doi.org/10.1080/1554480X.2024.2396944

    arXiv 2024

  56. [56]

    In: CHI ’20, pp

    Winkler, R., Hobert, S., Salovaara, A., et al.: Sara, the lecturer: Improving learning in online education with a scaffolding-based conversational agent. In: CHI ’20, pp. 1–14. ACM, New York (2020). https://doi.org/10.1145/3313831.3376781

    arXiv 2020

  57. [57]

    ACM Trans

    Deng, Y., et al.: Towards goal-oriented intelligent tutoring systems in online edu- cation. ACM Trans. Inf. Syst. (2025). https://doi.org/10.1145/3760401

    work page doi:10.1145/3760401 2025

  58. [58]

    HCI6(3-4), 201–250 (1991)

    MacLean, A., et al.: Questions, options, and criteria: Elements of design space analysis. HCI6(3-4), 201–250 (1991)

    1991

  59. [59]

    Introduction-the why, what, when, and how

    Elliott, J.H., et al.: Living systematic review: 1. Introduction-the why, what, when, and how. J. Clin. Epidemiol.91, 23–30 (2017)

    2017

  60. [60]

    In: CHI ’23, Article No

    Xu, X., et al.: XAIR: A framework of explainable AI in aug- mented reality. In: CHI ’23, Article No. 202. ACM, New York (2023). https://doi.org/10.1145/3544548.3581500

    arXiv 2023

  61. [61]

    Petter, S., DeLone, W., McLean, E.: Measuring information systems success: Mod- els, dimensions, measures. Eur. J. Inf. Syst.17(3), 236–263 (2008) Co-Designing Digital Humans for Online Learning 21

    2008

  62. [62]

    Wang, W.T., Wang, C.C.: An empirical study of instructor adoption of web-based learning systems. Comput. Educ.53(3), 761–774 (2009)

    2009

  63. [63]

    Al-Adwan, A.S.: Investigating the drivers and barriers to MOOCs adoption: The perspective of TAM. Educ. Inf. Technol.25(6), 5771–5795 (2020)

    2020

  64. [64]

    In: CHI ’22, pp

    Yeh, S.F., et al.: How to guide task-oriented chatbot users, and when. In: CHI ’22, pp. 1–13. ACM, New York (2022)

    2022

  65. [65]

    In: CHI ’19, pp

    Roy, Q., Zhang, F., Vogel, D.: Automation accuracy is good, but high controlla- bility may be better. In: CHI ’19, pp. 1–8. ACM, New York (2019)

    2019

  66. [66]

    Nahum-Shani, I., et al.: Just-in-time adaptive interventions (JITAIs) in mobile health. Ann. Behav. Med.52(6), 446–462 (2018)

    2018

  67. [67]

    Interact

    Sun, G.Y., et al.: Teachers’ self-efficacy and continuance commitment of using LMS. Interact. Learn. Environ., 1–18 (2021)

    2021

  68. [68]

    Online Learning23(4), 214–233 (2019)

    Sadaf, A., Martin, F., Ahlgrim-Delzell, L.: Impact of Quality Matters-certified online courses. Online Learning23(4), 214–233 (2019)

    2019

  69. [69]

    Brain Sciences15(2), 203 (2025)

    Gkintoni, E., et al.: Redefining learning efficacy through educational neuroscience and AI. Brain Sciences15(2), 203 (2025)

    2025

  70. [70]

    David McKay, New York (1956)

    Bloom, B.S., et al.: Taxonomy of educational objectives. David McKay, New York (1956)

    1956

  71. [71]

    Anderson,L.W.,etal.:Ataxonomyforlearning,teaching,andassessing.Longman, New York (2001)

    2001

  72. [72]

    Holt, Rinehart and Winston, New York (1985)

    Gagne, R.M.: The conditions of learning and theory of instruction, 4th edn. Holt, Rinehart and Winston, New York (1985)

    1985

  73. [73]

    Elaasri, R., Bouziane, A.: Applying the Quality Matters (QM) rubric to ENT platforms. Eur. J. Open Educ.4(2) (2019)

    2019

  74. [74]

    In: CHI ’18, pp

    Coppers, S., et al.: Intellingo: An intelligible translation environment. In: CHI ’18, pp. 1–13. ACM, New York (2018)

    2018

  75. [75]

    arXiv:1901.00770 (2019)

    Schneider, J., Handali, J.: Personalized explanation in machine learning. arXiv:1901.00770 (2019)

    Pith/arXiv arXiv 1901

  76. [76]

    Guo, F., et al.: Dual-view multi-modal contrastive learning for recommender sys- tems. Comput. Electr. Eng.116, 109213 (2024)

    2024

  77. [77]

    In: Hughes, G

    Cuevas, J.: Visual and auditory learning: Sensory modality effects on memory. In: Hughes, G. (ed.) Student Achievement, pp. 29–54. Nova Science Publishers, Hauppauge (2017)

    2017

  78. [78]

    de Oliveira Neto, J.D., et al.: Online learning: audio or text? Educ. Technol. Res. Dev.63(4), 555–573 (2015)

    2015

  79. [79]

    Xu, Wei., Zhou, Yuhan.: Course video recommendation with multimodal informa- tion. Brit. J. Educ. Technol.51(5), 1734–1747 (2020)

    2020

  80. [80]

    CoRR abs/1312.6034 (2013)

    Simonyan, K., et al.: Deep inside convolutional networks: Visualising image classi- fication. CoRR abs/1312.6034 (2013)

    Pith/arXiv arXiv 2013

Showing first 80 references.