CoChat
Sign Up
Cited by 3Artificial Intelligence in EducationIntelligent Tutoring Systems and Adaptive LearningOnline Learning and Analytics

AI Sensing and Intervention in Higher Education: Student Perceptions of Learning Impacts, Affective Responses, and Ethical Priorities

Open access
Han, Bingyi, Ma, Ying, Coghlan, Simon, McKay, Dana, Buchanan, George, Smith, Wally·2026·Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems
DOI 10.1145/3772318.3790360
Open this paper in CoChat — free account

Abstract

AI technologies that sense student attention and emotions to enable more personalised teaching interventions are increasingly promoted, but raise pressing questions about student learning, wellbeing, and ethics. In particular, students’ perspectives about AI sensing-intervention in learning are often overlooked. We conducted an online mixed-method experiment with Australian university students (N=132), presenting video scenarios varying by whether sensing was used (in-use vs. not-in-use), sensing modality (gaze-based attention detection vs. facial-based emotion detection), and intervention (by digital device vs. teacher). Participants also completed pairwise ranking tasks to prioritise six core ethical concerns. Findings revealed that students valued targeted intervention but responded negatively to AI monitoring, regardless of sensing methods. Students preferred system-generated hints over teacher-initiated assistance, citing learning agency and social embarrassment concerns. Students’ ethical considerations prioritised autonomy and privacy, followed by transparency, accuracy, fairness, and learning beneficence. We advocate designing customisable, social-sensitive, non-intrusive systems that preserve student control, agency, and well-being.

Verified 2026-10-25 · 38 references not checked yet

AI snapshot

Written from the paper’s full text.

TL;DR

In a two-stage online study with 132 Australian university students, AI sensing in learning scenarios produced negative affective and learning-impact ratings, while system-generated hints were preferred over teacher-mediated assistance. Students prioritised autonomy and privacy over other ethical concerns, including fairness, transparency, accuracy, and learning beneficence.

Key findings

  • Disclosing AI sensing sharply worsened perceptions: five of six learning-belief items and all six affective-response items were significantly less positive when sensing was in use; e.g., perceived efficiency fell from M = 3.60 to M = 2.56, anxiety rose from M = 3.03 to M = 4.22, and discomfort rose from M = 3.26 to M = 4.33. H1 was supported.
  • No significant differences were found between gaze-based attention detection and facial-based emotion detection on any item; H2 was rejected.
  • System-generated hints were rated significantly more positively than teacher-mediated assistance in both sensing not-in-use and sensing-in-use conditions; in the sensing-in-use condition, anxiety for human assistance was M = 4.50 versus M = 3.93 for system hints, and discomfort was M = 4.55 versus M = 4.11. H3 was supported.
  • Ethical priorities: Privacy (mean = 4.14) and Autonomy (mean = 4.10) drew the highest concern, while Fairness (mean = 2.73) was lowest. Pairwise wins were Autonomy N = 515, Privacy N = 462, Transparency N = 269, Learning Beneficence N = 261, Accuracy N = 248, and Fairness N = 240; Bradley-Terry probabilities were Autonomy 0.39, Privacy 0.27, Transparency 0.09, Learning Beneficence 0.09, Accuracy 0.08, and Fairness 0.08.

Methods

  • Two-stage online mixed-method study. Stage 1 used a within-subjects experiment in which 132 Australian university students viewed six video scenarios varying sensing use (in-use vs. not-in-use), sensing modality (gaze-based attention detection vs. facial-based emotion detection), and intervention form (system-generated hint vs. teacher-mediated assistance), then rated Likert items and provided open-ended responses. Stage 2 had participants rate 18 ethical-concern statements and complete 15 pairwise comparisons among six ethical concepts.
  • Participants were recruited via Prolific: 141 recruited, 9 excluded by attention check, final N = 132. All were Australian university students with English as a first language and ≥95% Prolific approval; compensation was GBP £13.64 per hour. Sample: 50% male, 47% female, 3% non-binary/third gender; ages 18–24 (51%), 25–35 (44%), 45–54 (5%); undergraduates 67%, postgraduates 17%, PhD 9%, other 7%.
  • Analysis used Mann-Whitney U tests with Bonferroni corrections for Likert comparisons, qualitative content analysis with descriptive coding for open-ended responses, descriptive statistics for ethical ratings, and the Bradley-Terry model for pairwise rankings. Cronbach’s alpha was 0.948 for Beliefs about Learning and 0.957 for Affective Responses; ethical scales were Privacy 0.90, Autonomy 0.82, Transparency 0.84, Fairness 0.72, Accuracy 0.68, and Learning 0.67.

Limitations

  • Stated: video scenarios did not fully simulate the experience of wrestling with a learning activity; future work could use actual systems. The sample focused on Australian higher education students, and expectations, cultural norms, and ethical concerns may vary across countries, educational systems, and age groups, including K–12. The study captured immediate reactions to hypothetical scenarios; long-term effects of using AI systems in education remain unexplored and require longitudinal study.
  • Evident from the text: qualitative coding was conducted by a single researcher (the first author), with peer-debrief sessions but not independent double coding; open-ended responses were short, typically 1–3 sentences. The study also relied on self-report responses to hypothetical scenarios rather than observed behaviour with deployed systems.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/3772318.3790360

Cites

125 of 170 references verified0 citation errors · 7 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

12 of 34 cited sources checked · 8 supported · 0 flagged · accuracy not yet measured

  • Premisecitation
    § 1 Introduction · p. 1

    Research on AIEd often overlooks deeper student-centred and learning factors.

    “Despite rapid technical advances, research on AI in Education (AIEd) often overlooks deeper student-centred and learning factors [17, 22].”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [17] [17] Melissa Bond, Hassan Khosravi, Maarten De Laat, Nina Bergdahl, Violeta Negrea, Emily Oxley, Phuong Pham, Sin Wang Chong and George Siemens. 2024. A Meta Systematic Review of Artificial Intelligence in Higher Education: A Call for Increased Ethics, Collaboration, and Rigour. International Journal of Educational Technology in Higher Education 21, 1, 4. - source
    • cites [22] [22] Christine Buchanan, M Lyndsay Howitt, Rita Wilson, Richard G Booth, Tracie Risling and Megan Bamford. 2021. Predicted Influences of Artificial Intelligence on Nursing Education: Scoping Review. JMIR nursing 4, 1, e23933. - source

      The Discussion’s Future Research section says more research is needed on the most effective ways to use AI as a tool in nursing education, supporting a narrower claim about gaps in educational research. It does not say that AIEd research generally overlooks deeper student-centred or learning factors, or contrast those factors with technical advances. Page numbers are unavailable in our copy. (LLM, conf=0.72)

      Experimental — accuracy not yet measured
  • Premisecitation
    § 3.1 Stage 1: Scenario-Based Experiment · p. 6

    Existing technology acceptance scales were not suitable to capture the specific affective and pedagogical concerns raised in prior observational and qualitative studies of AIEd systems.

    “Existing technology acceptance scales (e.g., [73]) were not suitable to capture the specific affective and pedagogical concerns raised in prior observational and qualitative studies of AIEd systems (e.g., [53, 56]).”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites [73] [73] Jihyun Kim, Kelly Merrill, Kun Xu and Deanna D Sellnow. 2020. My Teacher Is a Machine: Understanding Students' Perceptions of Ai Teaching Assistants in Online Education. International Journal of Human-Computer Interaction 36, 20, 1902-1911. - source
    • cites [53] [53] Bingyi Han, George Buchanan and Dana Mckay. 2023. Learning in the Panopticon: Examining the Potential Impacts of Ai Monitoring on Students. In Proceedings of Proceedings of the 34th Australian Conference on Human-Computer Interaction . Association for Computing Machinery, Canberra, ACT, Australia, 9-21. http://dx.doi.org/10.1145/3572921.3572937 - source
    • cites [56] [56] Bingyi Han, Sadia Nawaz, George Buchanan and Dana McKay. 2025. Students' Perceptions: Exploring the Interplay of Ethical and Pedagogical Impacts for Adopting Ai in Higher Education. International Journal of Artificial Intelligence in Education . http://dx.doi.org/10.1007/s40593-024-00456-4 - source
  • Premisecitation
    § 1 Introduction · p. 2

    Crucial AI ethics considerations are equally underexamined in AIEd.

    “Crucial AI ethics considerations are equally underexamined in AIEd [60, 103].”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [60] [60] Wayne Holmes, Kaska Porayska-Pomsta, Ken Holstein, Emma Sutherland, Toby Baker, Simon Buckingham Shum, Olga C Santos, Mercedes T Rodrigo, Mutlu Cukurova and Ig Ibert Bittencourt. 2021. Ethics of Ai in Education: Towards a Community-Wide Framework. International Journal of Artificial Intelligence in Education , 1-23. - source
    • cites [103] [103] Amy Ogan. 2019. Reframing Classroom Sensing: Promise and Peril. Interactions 26, 6, 26-32. - source

      In ETHICS AND FAIRNESS, Ogan says serious ethics and fairness questions about classroom sensing still need close examination, including who can access and benefit from the data. CAUTIONARY TALES also describes a lack of frameworks to guide its use. These passages support the concern for classroom sensing, but do not establish that AI ethics considerations are underexamined across AIEd generally. (LLM, conf=0.83)

      Experimental — accuracy not yet measured
  • Interpretationown data
    § 5.1 Possibilities and challenges regarding AI sensing-intervention systems · p. 13

    The mechanism of sensing that promises personalised, responsive teaching also risks undermining students' learning experiences.

    “Thus, the very mechanism of sensing that promises personalised, responsive teaching also risks undermining students' learning experiences [53, 56].”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [53] [53] Bingyi Han, George Buchanan and Dana Mckay. 2023. Learning in the Panopticon: Examining the Potential Impacts of Ai Monitoring on Students. In Proceedings of Proceedings of the 34th Australian Conference on Human-Computer Interaction . Association for Computing Machinery, Canberra, ACT, Australia, 9-21. http://dx.doi.org/10.1145/3572921.3572937 - source
    • cites [56] [56] Bingyi Han, Sadia Nawaz, George Buchanan and Dana McKay. 2025. Students' Perceptions: Exploring the Interplay of Ethical and Pedagogical Impacts for Adopting Ai in Higher Education. International Journal of Artificial Intelligence in Education . http://dx.doi.org/10.1007/s40593-024-00456-4 - source
  • Findingown data
    § Abstract · p. 1

    Students prioritised ethical concerns in the order: autonomy and privacy highest, followed by transparency, accuracy, fairness, and learning beneficence.

    “Students' ethical considerations prioritised autonomy and privacy, followed by transparency, accuracy, fairness, and learning beneficence.”
    • Also stated in: § 1 Introduction · p. 2 · hedged
      “Regarding ethical views, students were most concerned about autonomy and privacy, suggesting they highly value control over learning processes and data in interactions with educational technology.”
    Quote in the paperSubstantial claim
  • Classified as probably not a substantial claim. 20
  • May combine two claims. 31

  • Classified as probably not a substantial claim.

    Premisecitation
    § 2.1 Student experiences with AI sensing-intervention · p. 3

    Some students prefer automated assistance over human support to avoid feeling judged.

    “Some research suggests some students prefer automated assistance over human support to avoid feeling judged [54, 113]”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [54] [54] Bingyi Han, Simon Coghlan, George Buchanan and Dana McKay. 2025. Who Is Helping Whom? Student Concerns About Ai-Teacher Collaboration in Higher Education Classrooms. Proceedings of the ACM on Human-Computer Interaction 9, 2, 1-32. - source

      Section 4.3 reports that some students may rely on the system instead of seeking help directly from teachers. It describes the system as a way to obtain teacher assistance, however, and does not say students take this route to avoid feeling judged. Page numbers are unavailable in our copy. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites [113] [113] Paul Rodway and Astrid Schepman. 2023. The Impact of Adopting Ai Educational Technologies on Projected Course Satisfaction in University Students. Computers and Education: Artificial Intelligence 5, 100150. - source

  • May combine two claims.

    Premisecitation
    § 1 Introduction · p. 1

    Discomfort from being observed by AIEd may trigger behavioural change and undermine the technology's effectiveness.

    “For instance, discomfort from being observed by AIEd may trigger behavioural change [53] and undermine the technology's effectiveness [54].”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [53] [53] Bingyi Han, George Buchanan and Dana Mckay. 2023. Learning in the Panopticon: Examining the Potential Impacts of Ai Monitoring on Students. In Proceedings of Proceedings of the 34th Australian Conference on Human-Computer Interaction . Association for Computing Machinery, Canberra, ACT, Australia, 9-21. http://dx.doi.org/10.1145/3572921.3572937 - source
    • cites [54] [54] Bingyi Han, Simon Coghlan, George Buchanan and Dana McKay. 2025. Who Is Helping Whom? Student Concerns About Ai-Teacher Collaboration in Higher Education Classrooms. Proceedings of the ACM on Human-Computer Interaction 9, 2, 1-32. - source

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

    Premisecitation
    § 2.1 Student experiences with AI sensing-intervention · p. 3

    Intelligent Tutoring Systems offering personalised automated hints and explanations have been shown to improve learning outcomes and receive positive evaluation from students.

    “Meanwhile, ITS's offering personalised automated hints and explanations have been shown to improve learning outcomes and receive positive evaluation from students [76, 130].”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites [76] [76] Ekaterina Kochmar, Dung Do Vu, Robert Belfer, Varun Gupta, Iulian Vlad Serban and Joelle Pineau. 2020. Automated Personalized Feedback Improves Learning Gains in an Intelligent Tutoring System. In Proceedings of International conference on artificial intelligence in education . Springer, 140-146. - source
    • cites [130] [130] William Villegas-Ch, Diego Buenano-Fernandez, Alexandra Maldonado Navarro and Aracely Mera-Navarrete. 2025. Adaptive Intelligent Tutoring Systems for Stem Education: Analysis of the Learning Impact and Effectiveness of Personalized Feedback. Smart Learning Environments 12, 1, 1-31. - source

  • May combine two claims.

    Premisecitation
    § 5.2 Shape of students' ethical concerns · p. 14

    Higher education students have broader expectations of independence and a preference for agency in AI-assisted classrooms.

    “This aligns with higher education students' broader expectations of independence [134], and their preference for agency in AI-assisted classrooms [139].”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites [134] [134] Alexander Whitelock-Wainwright, Dragan Gašević, Ricardo Tejeiro, Yi-Shan Tsai and Kate Bennett. 2019. The Student Expectations of Learning Analytics Questionnaire. Journal of Computer Assisted Learning 35, 5, 633-666. http://dx. doi.org/10.1111/jcal.12366 - source
    • cites [139] [139] Kexin Bella Yang, Vanessa Echeverria, Zijing Lu, Hongyu Mao, Kenneth Holstein, Nikol Rummel and Vincent Aleven. 2023. Pair-Up: Prototyping Human-Ai Co-Orchestration of Dynamic Transitions between Individual and Collaborative Learning in the Classroom. In Proceedings of Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems . Association for Computing Machinery, Hamburg, Germany, Article 453. http://dx.doi.org/10.1145/3544548. 3581398 - source

      Section 6.3.1 reports that students wanted some control over their partners and the timing of transitions, supporting the preference-for-agency point attributed to this citation. But section 4.1 identifies the participants as sixth- to eighth-grade students, not higher education students, so the cited study does not establish that preference for the population named. The independence point is attached to a different citation [134]. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Findingown data
    § 4.2 Stage 1: Qualitative reports about learning impacts and affective responses to AI sensing-intervention · p. 12

    Students found sensing intrusive and disruptive to learning.

    “These qualitative fi ndings help explain students' negative reactions to sensing (H1) regardless of the two approaches (H2), which they found intrusive and disruptive to learning.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing).

    Findingown data
    § 4.2 Stage 1: Qualitative reports about learning impacts and affective responses to AI sensing-intervention · p. 11

    Participants suggested that a system gathering classroom data on average completion time and offering a hint without anyone knowing would be the best method.

    “'If it gathered classroom data on average completion time and offered a hint without anyone knowing, that would be the best method.' (P31)”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 4.1 Stage 1: Believed learning impacts and affective response towards AI sensing-intervention · p. 8

    In the sensing-in-use condition, ratings shifted sharply negative on both believed learning impacts and affective responses.

    “In the sensing-in-use condition, ratings shifted sharply negative on both believed learning impacts and affective responses.”
    Quote in the paperSubstantial claim

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

    Findingnone given
    § 5.3 Design Implications for AI sensing-intervention in education · p. 14

    Students sought agency not just at the start (opt-in/opt-out), but dynamically throughout use.

    “Students sought agency not just at the start (opt-in/opt-out), but dynamically throughout use.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 5.1 Possibilities and challenges regarding AI sensing-intervention systems · p. 13

    Affective responses and learning beliefs tended to align: positive emotions corresponded with more positive learning perceptions, and vice versa.

    “Our results suggest that affective responses and learning beliefs tended to align. Positive emotions corresponded with more positive learning perceptions, and vice versa.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 4.3 Stage 3: Students' ethical concerns · p. 12

    Statements relating to Privacy (mean = 4.14) and Autonomy (mean = 4.10) garnered the highest level of concern from participants, while Fairness (mean = 2.73) was rated the least concerning.

    “Overall, statements relating to Privacy (mean = 4.14) and Autonomy (mean = 4.10) garnered the highest level of concern from participants, while Fairness (mean = 2.73) was rated the least concerning.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The six ethical principles (Autonomy, Transparency, Privacy, Fairness, Accuracy, Learning Beneficence) are the key ethical principles for AI sensing-intervention in education.

  • May combine two claims.

    Findingown data
    § 4.1 Stage 1: Believed learning impacts and affective response towards AI sensing-intervention · p. 8

    In the sensing not-in-use condition, system-generated hints were perceived significantly more positively than human teacher assistance for efficiency, control of pace, and distraction, with no difference for personalisation, grade performance, and over-helping.

    “In the sensing not-in-use condition (Table.7), system-generated hints were perceived significantly more positively than human teacher assistance for 3 items of believed learning impacts ( efficiency, control of peace, distraction ), with no difference for 3 items ( personalisation, grade performance, over-helping ).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Findingown data
    § 4.2 Stage 1: Qualitative reports about learning impacts and affective responses to AI sensing-intervention · p. 11

    Participants questioned whether the system could accommodate diverse learning styles, disabilities, cultural nuances, and neurodivergence.

    “Participants questioned whether the system could accommodate diverse learning styles, disabilities, cultural nuances, and neurodivergence.”
    Quote in the paperSubstantial claim

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

    Findingown data
    § 4.2 Stage 1: Qualitative reports about learning impacts and affective responses to AI sensing-intervention · p. 11

    Participants worried the system lacked flexibility and could misinterpret behaviours, disadvantaging those not conforming to expected patterns.

    “They worried it lacked fl exibility and could misinterpret behaviours, disadvantaging those not conforming to expected patterns.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 4.3 Stage 3: Students' ethical concerns · p. 13

    Even though Autonomy and Privacy are clearly more important to participants, the highest probability of wins was 39%, indicating that no single concern completely overshadows the others within the participant sample.

    “It should be noted that even though Autonomy and Privacy are clearly more important to participants, the highest probability of wins in this context was 39%, indicating that no single concern completely overshadows the others within our participant sample.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The six ethical principles (Autonomy, Transparency, Privacy, Fairness, Accuracy, Learning Beneficence) are the key ethical principles for AI sensing-intervention in education.

  • May combine two claims.

    Findingown data
    § 4.1 Stage 1: Believed learning impacts and affective response towards AI sensing-intervention · p. 8

    Five of six learning beliefs items (personalisation, efficiency, control of pace, grade performance, distraction) and all six affective responses items were significantly less positive when sensing was disclosed, with only over-helping unaffected.

    “Five of six learning beliefs items (personalisation, efficiency, control of pace, grade performance, distraction) and all six items from affective responses (enjoyable, confident, anxious, uncomfortable, worried, fearing mistakes) were significantly less positive when sensing was disclosed, with only over-helping unaffected.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 4.1 Stage 1: Believed learning impacts and affective response towards AI sensing-intervention · p. 10

    When sensing was disclosed, system-generated hints were associated with sharp increases in anxiety (mean 3.93 vs 2.22), discomfort (mean 4.11 vs 2.56), and fear of mistakes (mean 3.54 vs 2.33).

    “System hints were now seen as more distracting (mean = 3.74, changed from 2.77) and emotionally taxing, with sharp increases in anxiety (mean = 3.93, changed from 2.22). discomfort (mean = 4.11, changed from 2.56), and fear of mistakes (mean = 3.54, changed from 2.33).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingcitation
    § 5.1 Possibilities and challenges regarding AI sensing-intervention systems · p. 13

    Perceived benefits proved fragile once students realised that interventions derived from sensing; positive perceptions shifted to neutral or negative, with both emotion and attention detection provoking discomfort, anxiety, and beliefs that the system would be distracting.

    “Yet these perceived benefits proved fragile once students realised that interventions derived from sensing. Positive perceptions shifted to neutral or negative, with both emotion and attention detection provoking discomfort, anxiety, and beliefs that the system would be distracting.”
    Quote in the paperSubstantial claim

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

    Findingown data
    § 4 Findings · p. 8

    The final dataset included 132 participants.

    “Our fi nal dataset included 132 participants aged 18 ∼ 24 (51%), 25 ∼ 35 (44%), 45 ∼ 54 (5%), with none below 18 or above 55.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 4.1 Stage 1: Believed learning impacts and affective response towards AI sensing-intervention · p. 8

    In the sensing-in-use condition, system-generated hints were rated significantly more positively than human teacher assistance across nearly all items in believed learning impacts except over-helping, and all items in affective responses.

    “In the sensing-in-use condition (Table.8), system-generated hints again were rated significantly more positively than human teacher assistance, across nearly all items in believed learning impacts except over-helping , and all items in affective responses.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Findingown data
    § 4 Findings · p. 8

    All participants were current university students.

    “All participants were current university students, including undergraduates (67%), postgraduates (17%), PhD (9%), and others such as honours and diplomas (7%).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 4.2 Stage 1: Qualitative reports about learning impacts and affective responses to AI sensing-intervention · p. 11

    Participants frequently reported feelings of anxious (P50), uncomfortable (P39), and distracted (P90).

    “Participants frequently reported feelings of 'anxious' (P50), 'uncomfortable' (P39), and 'distracted' (P90).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 4.2 Stage 1: Qualitative reports about learning impacts and affective responses to AI sensing-intervention · p. 12

    Many participants raised broader ethical concerns around privacy and data misuse (44 participants) and accuracy in interpreting emotional or cognitive states (58 participants).

    “Many also raised broader ethical concerns around privacy and data misuse (44 participants) and accuracy in interpreting emotional or cognitive states (58 participants).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 4.3 Stage 3: Students' ethical concerns · p. 12

    In 1980 paired comparisons, Autonomy (N = 515) and Privacy (N = 462) received the highest total wins, indicating participants placed the most importance on them; Transparency (N = 269), Learning Beneficence (N = 261), and Accuracy (N = 248) scored lower, with Fairness (N = 240) last.

    “The results reveal that Autonomy (N = 515) and Privacy (N = 462) received the highest total wins, indicating that participants placed the most importance on them. Transparency (N = 269), Learning Beneficence (N = 261), and Accuracy (N = 248) scored lower, with Fairness (N = 240) last.”
    • Also stated in: § 5.2 Shape of students' ethical concerns · p. 14 · plain
      “Pairwise comparisons reinforced this hierarchy, with students prioritising autonomy and privacy over potential learning benefits.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The six ethical principles (Autonomy, Transparency, Privacy, Fairness, Accuracy, Learning Beneficence) are the key ethical principles for AI sensing-intervention in education.

  • May combine two claims.

    Findingown data
    § 5.2 Shape of students' ethical concerns · p. 13

    Students ranked lack of consent before deployment and opt-out options (autonomy) highest, followed by not being informed of data use and associated risks (transparency), and having learning struggles or emotions revealed to others (privacy).

    “Ranked highest were the lack of consent before deployment and opt-out options (autonomy), followed by not being informed of data use and associated risks (one kind of transparency), and having learning struggles or emotions revealed to others (privacy).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The six ethical principles (Autonomy, Transparency, Privacy, Fairness, Accuracy, Learning Beneficence) are the key ethical principles for AI sensing-intervention in education.

  • May combine two claims.

    Findingown data
    § 4.2 Stage 1: Qualitative reports about learning impacts and affective responses to AI sensing-intervention · p. 11

    Participants worried about the system's inability to account for everyday emotional fluctuations, leading to misinterpretation of their emotional state and inappropriate intervention.

    “Participants also worried about the system's inability to account for everyday emotional fl uctuations, leading to misinterpretation of their emotional state and inappropriate intervention.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 5.2 Shape of students' ethical concerns · p. 13

    Concerns about erroneous interpretation of behaviour and emotions (accuracy) were moderate, while understanding how algorithms work (another kind of transparency) and fairness issues, such as algorithmic bias, evoked the least concern.

    “Concerns about erroneous interpretation of behaviour and emotions (accuracy) were moderate, while understanding how algorithms work (another kind of transparency) and fairness issues, such as algorithmic bias, evoked the least concern.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The six ethical principles (Autonomy, Transparency, Privacy, Fairness, Accuracy, Learning Beneficence) are the key ethical principles for AI sensing-intervention in education.

  • May combine two claims.

    Findingreasoning
    § 5.3 Design Implications for AI sensing-intervention in education · p. 14

    Students generally favoured system-generated hints, but only when these preserved autonomy and comfort.

    “Students generally favoured system-generated hints, but only when these preserved autonomy and comfort.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning
    § 7 Conclusion · p. 15

    Ethical priorities, especially autonomy and privacy, outweighed perceived learning benefits, with fairness, transparency, and accuracy considered secondary.

    “Ethical priorities, especially autonomy and privacy, outweighed perceived learning benefits, with fairness, transparency, and accuracy considered secondary.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The six ethical principles (Autonomy, Transparency, Privacy, Fairness, Accuracy, Learning Beneficence) are the key ethical principles for AI sensing-intervention in education.

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

    Study design and conductown data
    § 3.1 Stage 1: Scenario-Based Experiment · p. 6

    The items were piloted with 10 students to ensure clarity and ease of interpretation, with minor wording revisions made based on their feedback.

    “The items were piloted with 10 students to ensure clarity and ease of interpretation, with minor wording revisions made based on their feedback.”
    Quote in the paperNot a substantial claim

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

    Study design and conductown data
    § 3.1 Stage 1: Scenario-Based Experiment · p. 6

    Weekly peer-debrief sessions were conducted within the research team to ensure analysis transparency.

    “To ensure analysis transparency, weekly peer-debrief sessions were conducted within the research team to review code definitions, category boundaries, and disconfirming cases.”
    Quote in the paperNot a substantial claim

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

    Study design and conductreasoning
    § 3.2 Stage 2: Ethics Pairwise Ranking Tasks · p. 7

    The items and statements used in Stage 2 reflect the review of salient ethical principles from established frameworks.

    “The items and statements used in Stage 2 reflect our review of salient ethical principles from established frameworks (Section 2.2.1).”
    Quote in the paperNot a substantial claim

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

    Study design and conductnone given
    § 3.1 Stage 1: Scenario-Based Experiment · p. 5

    All instructions were carefully worded with neutral terms to avoid biasing participants' perceptions.

    “All instructions were carefully worded with neutral terms such as ' … the system observes you', avoiding expressions that might bias participants' perceptions.”
    Quote in the paperNot a substantial claim

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

    Study design and conductnone given
    § 3 Study Method · p. 5

    Ethical approval for the study was granted by the University of Melbourne.

    “Ethical approval for the study was granted by the University of Melbourne.”
    Quote in the paperNot a substantial claim

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

    Study design and conductreasoning
    § 6 Limitations and future work · p. 15

    The long-term effects of using AI systems in education, such as changes in behaviour or attitudes over time, remain unexplored.

    “The long-term effects of using AI systems in education, such as changes in behaviour or attitudes over time, remain unexplored.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data
    § 4.1 Stage 1: Believed learning impacts and affective response towards AI sensing-intervention · p. 8

    The results support H1: Students perceive AI systems that rely on sensing more negatively than systems that do not use sensing, both in terms of potential learning impacts and affective responses.

    “These results support H1 - Students perceive AI systems that rely on sensing more negatively than systems that do not use sensing, both in terms of (a) potential learning impacts and (b) affective responses.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The sample of Australian university students is representative of higher education students more broadly.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationcitation
    § 5.2 Shape of students' ethical concerns · p. 14

    A value sensitive design approach to AIEd would accommodate ethical considerations, such as needs for autonomy and privacy, that could affect student well-being, trust, and learning.

    “A value sensitive design approach to AIEd would accommodate ethical considerations, such as needs for autonomy and privacy, that could affect student well-being, trust, and learning.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [18] [18] Alan Borning and Michael Muller. 2012. Next Steps for Value Sensitive Design. In Proceedings of Proceedings of the SIGCHI conference on human factors in computing systems . 1125-1134. - source

  • May combine two claims.

    Interpretationown data
    § 4.4 Summary of Findings · p. 13

    Students were most concerned about autonomy and privacy, suggesting students highly value control over their data and learning process in AIEd interactions.

    “Ethically speaking, students were most concerned about autonomy and privacy, suggesting that students highly value control over their data and learning process in AIEd interactions.”
    Quote in the paperSubstantial claim

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

    Premisecitation
    § 6 Limitations and future work · p. 15

    Expectations, cultural norms, and ethical concerns can vary across countries, educational systems, and age groups.

    “Expectations, cultural norms, and ethical concerns can vary across countries, educational systems, and age groups.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [57] [57] Geert Hofstede. 1980. Culture's Consequences: International Differences in Work-Related Values London. UK: Sage.[Google Scholar] . -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationown data
    § 4.1 Stage 1: Believed learning impacts and affective response towards AI sensing-intervention · p. 8

    The results support H3: Students perceive system-generated assistance more positively than human-delivered assistance, both in terms of potential learning impacts and affective responses.

    “These results support H3 - Students perceive system-generated assistance more positively than human-delivered assistance, both in terms of (a) potential learning impacts and (b) affective responses.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The sample of Australian university students is representative of higher education students more broadly.

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

    Premisecitation
    § 5.2 Shape of students' ethical concerns · p. 14

    Autonomy in education also encompasses broader control over what and how one studies, and the ability to make learning choices without undue direction from technologies.

    “As both prior research [56] and our qualitative fi ndings emphasise, autonomy in education also encompasses broader control over what and how one studies, and the ability to make learning choices without undue direction from technologies [56].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [56] [56] Bingyi Han, Sadia Nawaz, George Buchanan and Dana McKay. 2025. Students' Perceptions: Exploring the Interplay of Ethical and Pedagogical Impacts for Adopting Ai in Higher Education. International Journal of Artificial Intelligence in Education . http://dx.doi.org/10.1007/s40593-024-00456-4 - source

  • May combine two claims.

    Interpretationown data
    § 5.2 Shape of students' ethical concerns · p. 14

    Students' ethical priorities around autonomy and privacy directly shape their negative responses to sensing and teacher-initiated interventions, and explain their preference for private, system-generated hints.

    “Taken together, students' ethical priorities around autonomy and privacy directly shape their negative responses to sensing and teacher-initiated interventions, and explain their preference for private, system-generated hints.”
    Quote in the paperSubstantial claim

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

    Study design and conductnone given
    § 3.3 Participants · p. 8

    We recruited 141 participants via Prolific, with 9 excluded based on an attention check to ensure data quality.

    “We recruited 141 participants via the crowdsourcing platform Prolific, with 9 excluded based on an attention check to ensure data quality.”
    Quote in the paperSubstantial claim

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

    Study design and conductcitation
    § 3.3 Participants · p. 8

    A power analysis suggested a minimum of 106 participants were required, based on a medium effect size f = 0.35 with a significance level of 0.05 at 80%.

    “A power analysis suggested a minimum of 106 participants were required, based on a medium effect size [29] f = 0.35 with a significance level of 0.05 at 80%.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [29] [29] Jacob Cohen. 2013. Statistical Power Analysis for the Behavioral Sciences . routledge, - source

  • May combine two claims.

    Interpretationreasoning
    § 5.2 Shape of students' ethical concerns · p. 14

    The prominence of consent and opt-out options reflects students' strong desire for greater agency and their foundational role: without being informed and able to choose, students cannot meaningfully evaluate or respond to other risks.

    “The prominence of consent and opt-out options reflect not only students' strong desire for greater agency in how technology is integrated into their learning, but also their foundational role: without being informed and able to choose, students cannot meaningfully evaluate or respond to other risks.”
    Quote in the paperSubstantial claim

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

    Study design and conductnone given
    § 5.2 Shape of students' ethical concerns · p. 14

    Our measures captured only one aspect of autonomy-the ability to accept or decline system use.

    “Meanwhile, our measures captured only one aspect of autonomy -the ability to accept or decline system use.”
    Quote in the paperNot a substantial claim
45 of 45

  • The paper relies on this without support.

    No existing study has examined how students themselves evaluate ethical trade-offs.

    “No existing study has examined how students themselves evaluate these trade-offs.”

  • The paper relies on this without support.

    The connection between AI-detected emotions and learning is more ambiguous and context-dependent than the connection between attention and learning.

    “whereas the connection between AI-detected emotions and learning is more ambiguous and context-dependent.”

  • The paper relies on this without support.

    The sample of Australian university students is representative of higher education students more broadly.

    “Participants were all university students in Australia, with English as their fi rst language and a minimum approval rate of 95% on the Prolific platform.”

  • The paper relies on this without support.

    The video-scenario method validly elicits students' real perceptions and reactions to AI sensing-intervention systems.

    “This study used video scenarios to elicit perceptions of AI sensingintervention during puzzle tasks. While effective for capturing reactions, we did not fully simulate the experience of wrestling with a learning activity.”

  • The paper relies on this without support.

    The six ethical principles (Autonomy, Transparency, Privacy, Fairness, Accuracy, Learning Beneficence) are the key ethical principles for AI sensing-intervention in education.

    “we analysed the results from 1980 (15*132) paired comparisons participants made between two of the six key ethical principles of Autonomy, Transparency, Privacy, Fairness, Accuracy, and Learning Beneficence (Table.4).”
5 of 5

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Adjusted p-value is smaller than the raw p-value

    Major

    Table 8 reports a raw p-value of 0.01 but an adjusted p-value of 0.00 for grade performance. The stated Bonferroni correction cannot reduce a p-value; the distraction row has the same discrepancy.

  • Scenario 3 reverses the assigned intervention

    Minor

    Table 1 assigns a system hint to Scenario 3, whereas Appendix A describes Scenario 3 as teacher assistance.

  • Abstract puts learning beneficence last despite higher win count

    Minor

    The abstract's ordered list places learning beneficence after accuracy and fairness. The reported pairwise wins instead put learning beneficence above both, with fairness last.

  • Ethics findings are labeled Stage 3 despite a two-stage design

    Minor

    The methods assign the ethics tasks to Stage 2, but the corresponding findings section is titled Stage 3.

  • Scenario 5 reverses the assigned intervention

    Minor

    Table 1 assigns a system hint to Scenario 5, whereas Appendix A describes Scenario 5 as teacher assistance.

  • Figure 2 is described as both mean and median ratings

    Minor

    The figure caption calls its plotted ratings means, while the findings text says it plots medians and interquartile ranges.

6 of 6

Cites

Flagged first, then by citations

43 of the 142 references couldn’t be matched to a known paper.

Cited by

Sorted by citations

Authors
Han, B., Ma, Y., Coghlan, S., McKay, D., Buchanan, G., & Smith, W.
Affiliations
  • The University of Melbourne (AU)
  • RMIT University (AU)
ORCID
Venue
Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems
Publisher
ACM
Published
2026-04-13
Pages
1-21
DOI
10.1145/3772318.3790360
OpenAlex
W7128721667
Semantic Scholar
2f52bdc4407180900fe6e950b91fb63eb8ef9635
Cited by
3
References
142
Citations by year
2026: 2
FWCI
23.73
Type
conference paper
Open access
Licence
cc-by
Locations
  • CHI Conference on Human Factors in Computing Systems (publishedVersion, cc-by)
  • arXiv (Cornell University) (submittedVersion, cc-by)
  • arXiv (Cornell University) (cc-by)
Topics
  • Physical Sciences › Computer Science › Information Systems › Artificial Intelligence in Education
  • Physical Sciences › Computer Science › Artificial Intelligence › Intelligent Tutoring Systems and Adaptive Learning
  • Physical Sciences › Computer Science › Computer Science Applications › Online Learning and Analytics
Keywords
student attention detection, emotion recognition, student perceptions, ethical considerations, student independence, privacy, learning agency, facial emotion recognition, higher education

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-25.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-25.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review