pith. sign in

arxiv: 2604.06178 · v2 · submitted 2026-02-03 · 💻 cs.HC

"Help Me, But Don't Track Me": Intervention Timing and Privacy Boundaries for Process-Aware AI Tutors

Jane Hanqi Li , Yuhong Zhang , Jiaqi Liu , Tzyy-Ping Jung , Amy Eguchi This is my paper

Pith reviewed 2026-05-16 07:19 UTC · model grok-4.3

classification 💻 cs.HC
keywords AI tutorsproactive interventionprivacy boundariesstudent preferencesintervention timinglearning process dataK-12 mathematics
0
0 comments X

The pith

Secondary students prefer AI math tutors that offer hints instead of answers, intervene gradually rather than constantly, and avoid tracking attention or behavior.

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

The paper reports results from a survey of 330 Chinese secondary students (Grades 7-11) on desired behaviors for generative AI tutors in mathematics. Students consistently chose autonomy-preserving help such as hints over direct answers. They also selected graduated proactive support that begins with small hints and escalates only when needed, rather than frequent interruptions. Students accepted sharing problem-solving steps and mistake patterns but rejected collection of attention or behavior signals. These patterns supply early empirical guidance for balancing timely support with learner agency and privacy in process-aware K-12 systems.

Core claim

Students preferred autonomy-preserving support such as hints over direct answers, favored graduated proactive support over constant interruption, and drew clear privacy boundaries around learning-process data by accepting problem-solving steps and mistake patterns while rejecting attention- or behavior-related signals.

What carries the argument

Survey of student preferences on intervention timing and acceptable uses of learning-process data for process-aware AI tutors.

Load-bearing premise

Self-reported survey preferences from one sample of students will match actual acceptance and behavior once process-aware AI tutors are deployed.

What would settle it

Deploy two versions of a live AI tutor—one using graduated hints with data limited to steps and errors, the other using direct answers and attention tracking—then compare student engagement, dropout rates, and self-reported comfort.

Figures

Figures reproduced from arXiv: 2604.06178 by Amy Eguchi, Jane Hanqi Li, Jiaqi Liu, Tzyy-Ping Jung, Yuhong Zhang.

Figure 1
Figure 1. Figure 1: Privacy boundaries: What learning-process data can AI use? (Multiple selection; N=330) Overall acceptance of process-aware AI support. Most students found process-aware help ac￾ceptable: 61.3% (202/330) rated it as "mostly" or "completely acceptable," 30.5% (101/330) were neutral, and 8.2% (27/330) found it not acceptable. Data boundaries. Students drew clear boundaries around what types of learning-proces… view at source ↗
Figure 2
Figure 2. Figure 2: Ordinal logistic regression: Proactive perceptions predicting acceptance of process-aware AI support Together, these results suggest that acceptance is shaped by a benefit-cost tradeoff: students are more receptive when process-aware support feels adaptive and useful, but less receptive when it introduces interruption costs. We elaborate on the implications of this pattern in Section 5.2. 5. Discussion We … view at source ↗
read the original abstract

As generative AI (GenAI) tools are increasingly used as informal tutors for mathematics learning, future systems may become more proactive and process-aware in deciding when and how to offer support. Yet such support raises an important design tension: help that is timely may also feel interruptive or overly monitoring. To inform the design of process-aware AI tutors, we surveyed 330 secondary school students in China (Grades 7--11) about their preferred tutoring behaviors, attitudes toward proactive intervention, and acceptable use of learning-process data. We found three design-relevant patterns. First, students preferred autonomy-preserving support, such as hints over direct answers. Second, they favored graduated proactive support over constant interruption, preferring small hints first and stronger assistance only as needed. Third, they drew clear privacy boundaries around learning-process data: students were comfortable with problem-solving steps and mistake patterns, but substantially less comfortable with attention- or behavior-related signals. Together, these findings offer early empirical guidance for designing AI tutors that balance timely support with learner agency, and personalization with perceived privacy boundaries in K-12 contexts.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper reports results from a survey of 330 Chinese secondary students (Grades 7-11) on preferences for process-aware AI tutors in mathematics. Key findings include a preference for autonomy-preserving support such as hints over direct answers, graduated proactive interventions over constant interruption, and differentiated privacy boundaries (comfort with problem-solving steps and mistake patterns but lower comfort with attention- or behavior-related signals).

Significance. If the self-reported patterns hold, the work supplies early empirical input for HCI design of proactive AI tutors that respect learner agency and privacy in K-12 settings. The sample size supports pattern identification, though the absence of behavioral validation limits direct applicability to deployed systems.

major comments (2)
  1. [Methods] Methods section: The manuscript provides no information on survey question design, pilot validation, sampling procedure, response rate, or statistical analysis used to derive the three headline patterns, preventing assessment of how well the data support the stated claims.
  2. [Discussion] Discussion section: The central design guidance rests on the untested assumption that cross-sectional self-reported preferences will predict actual acceptance and behavior when students interact with deployed process-aware tutors; no prototype testing, behavioral logs, or longitudinal data are reported to support this mapping.
minor comments (1)
  1. [Abstract] Abstract: The summary of findings could explicitly note that all patterns derive from stated attitudes rather than observed interaction data.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our survey study. We address the two major comments point by point below, indicating the revisions we will make.

read point-by-point responses

  1. Referee: [Methods] Methods section: The manuscript provides no information on survey question design, pilot validation, sampling procedure, response rate, or statistical analysis used to derive the three headline patterns, preventing assessment of how well the data support the stated claims.

    Authors: We agree that the current Methods section omits these critical details. In the revised manuscript we will expand the section to describe the survey instrument development (including item sources from prior HCI and privacy literature), pilot validation steps, sampling approach (targeted recruitment of Chinese secondary students via educational platforms), response rate, and the descriptive and inferential statistical procedures used to identify the three patterns. These additions will allow readers to assess the support for the reported findings. revision: yes

  2. Referee: [Discussion] Discussion section: The central design guidance rests on the untested assumption that cross-sectional self-reported preferences will predict actual acceptance and behavior when students interact with deployed process-aware tutors; no prototype testing, behavioral logs, or longitudinal data are reported to support this mapping.

    Authors: We concur that self-reported preferences from a single cross-sectional survey cannot be assumed to predict real-world acceptance or usage of deployed tutors. The study was designed as an initial exploration of student attitudes rather than a behavioral validation. In the revised Discussion we will explicitly acknowledge this limitation, reframe the results as preliminary design guidance, and outline the need for future prototype-based and longitudinal studies to test translation to actual behavior. revision: partial

Circularity Check

0 steps flagged

No circularity: direct empirical survey with no derivations or self-referential reductions

full rationale

This is a cross-sectional survey study reporting stated preferences from 330 students on tutoring behaviors, proactive intervention, and data privacy. No equations, fitted models, predictions, ansatzes, or uniqueness theorems appear in the derivation chain. All three headline patterns are direct aggregates of survey responses rather than quantities defined by the paper's own choices or reduced via self-citation. The study is therefore self-contained against external benchmarks with no load-bearing circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that survey responses validly capture preferences that generalize beyond the sample and translate to real AI use; no free parameters or invented entities are introduced.

axioms (1)
  • domain assumption Self-reported survey responses from secondary students accurately reflect their preferences for AI tutor behaviors and data use
    Standard assumption in user studies; the abstract provides no validation against actual usage data.

pith-pipeline@v0.9.0 · 5508 in / 1149 out tokens · 31942 ms · 2026-05-16T07:19:49.133278+00:00 · methodology

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

29 extracted references · 29 canonical work pages

  1. [1]

    J. W. Lai, Adapting self-regulated learning in an age of generative artificial intelligence chatbots, Future Internet 16 (2024) 218

    work page 2024

  2. [2]

    N. F. Davar, M. A. A. Dewan, X. Zhang, Ai chatbots in education: challenges and opportunities, Information 16 (2025) 235

    work page 2025

  3. [3]

    Stańdo, Ż

    J. Stańdo, Ż. Fechner, A. Dąbrowicz-Tlałka, K. Kujawska, M. M. Musielak, Exploring AI chatbots for learning mathematics: Students’ perspectives on accuracy and educational value, in: International Conference on Artificial Intelligence in Education, Springer Nature Switzerland, Cham, 2025, pp. 282–288

    work page 2025

  4. [4]

    B. S. Bloom, The 2 sigma problem: the search for methods of group instruction as effective as one-to-one tutoring, Educational Researcher 13 (1984) 4–16

    work page 1984

  5. [5]

    C. C. Lin, A. Y. Huang, O. H. Lu, Artificial intelligence in intelligent tutoring systems toward sustainable education: a systematic review, Smart Learning Environments 10 (2023) 41

    work page 2023

  6. [6]

    Maity, A

    S. Maity, A. Deroy, Generative AI and its impact on personalized intelligent tutoring systems, arXiv preprint, 2024.arXiv:2410.10650

    work page arXiv 2024

  7. [7]

    K. R. Koedinger, V. Aleven, Exploring the assistance dilemma in experiments with cognitive tutors, in: R. Luckin, et al. (Eds.), Artificial Intelligence in Education, Springer, 2007, pp. 239–246

    work page 2007

  8. [8]

    how do you know that i don’t understand?

    A. Sarrafzadeh, S. Alexander, F. Dadgostar, C. Fan, A. Bigdeli, “how do you know that i don’t understand?” a look at the future of intelligent tutoring systems, Computers in Human Behavior 24 (2008) 1342–1363

    work page 2008

  9. [9]

    International Journal of Human-Computer Studies106, 63–76 (2017) https://doi.org/10.1016/j.ijhcs

    S. D’Mello, A. Olney, C. Williams, P. Hays, Gaze tutor: A gaze-reactive intelligent tutoring system, International Journal of Human-Computer Studies 70 (2012) 377–398. doi: 10.1016/j.ijhcs. 2012.01.004

    work page doi:10.1016/j.ijhcs 2012

  10. [10]

    Khuman, The impact of non-verbal communication in teaching: Enhancing educational effec- tiveness, A Global Journal of Humanities 7 (2024) 89

    P. Khuman, The impact of non-verbal communication in teaching: Enhancing educational effec- tiveness, A Global Journal of Humanities 7 (2024) 89

    work page 2024

  11. [11]

    S. Hutt, J. Hardey, N. Chen, N. Mohammad, S. D’Mello, Mind wandering in classrooms: Insights from classroom videos and neural networks, Proceedings of the ACM on Human-Computer Interaction 3 (2019) Article 206. doi:10.1145/3359298

    work page doi:10.1145/3359298 2019

  12. [12]

    Kraus, D

    M. Kraus, D. Betancourt, W. Minker, Does it affect you? social and learning implications of using cognitive-affective state recognition for proactive human-robot tutoring, in: 2023 32nd IEEE International Conference on Robot and Human Interactive Communication (RO-MAN), IEEE, 2023, pp. 928–935

    work page 2023

  13. [13]

    Maniktala, C

    M. Maniktala, C. Cody, T. Barnes, M. Chi, Avoiding help avoidance: Using interface design changes to promote unsolicited hint usage in an intelligent tutor, International Journal of Artificial Intelligence in Education 30 (2020) 637–667

    work page 2020

  14. [14]

    Marwiang, M

    M. Marwiang, M. Prasertsang, P. Junpeng, Enhancing students’ learning outcomes in mathematics through intelligent tutoring systems based on real-time feedback, Journal of Education and Learning 14 (2025) 186–196

    work page 2025

  15. [15]

    J. Ahn, F. Campos, H. Nguyen, M. Hays, J. Morrison, Co-designing for privacy, transparency, and trust in K–12 learning analytics, in: LAK21: 11th International Learning Analytics and Knowledge Conference, 2021, pp. 55–65. doi:10.1145/3448139.3448145

    work page doi:10.1145/3448139.3448145 2021

  16. [16]

    Mintz, W

    J. Mintz, W. Holmes, L. Liu, M. Perez-Ortiz, Artificial intelligence and K–12 education: Possibilities, pedagogies and risks, Computers in the Schools 40 (2023) 325–333

    work page 2023

  17. [17]

    Paolucci, S

    C. Paolucci, S. Vancini, R. T. Bex II, C. Cavanaugh, C. Salama, Z. de Araujo, A review of learning analytics opportunities and challenges for K–12 education, Heliyon 10 (2024)

    work page 2024

  18. [18]

    P. M. Regan, J. Jesse, Ethical challenges of edtech, big data and personalized learning: Twenty-first century student sorting and tracking, Ethics and Information Technology 21 (2019) 167–179

    work page 2019

  19. [19]

    Rehman, A

    A. Rehman, A. Dæhlen, I. Heldal, J. C. W. Lin, Privacy preservation and identity tracing prevention in AI-driven eye tracking for interactive learning environments, IEEE Access (2025)

    work page 2025

  20. [20]

    Snoek, A

    A. Snoek, A. M. Brouwer, I. V. Stuldreher, P. Haselager, D. Horstkötter, Wearables for tracking mental state in the classroom: ethical considerations from the literature and high school students, Frontiers in Neuroergonomics 6 (2025) 1536781

    work page 2025

  21. [21]

    T. Yang, R. Hasan, Discovering privacy harms from education technology by analyzing user reviews, in: Proceedings of the 23rd Workshop on Privacy in the Electronic Society (WPES ’24), 2024, pp. 186–192

    work page 2024

  22. [22]

    A. K. Shrestha, A. Barthwal, M. Campbell, A. Shouli, S. Syed, S. Joshi, J. Vassileva, Navigating AI to unpack youth privacy concerns: An in-depth exploration and systematic review, arXiv preprint, 2024.arXiv:2412.16369

    work page arXiv 2024

  23. [23]

    L. M. Rizki, Y. S. Kusumah, Unveiling the needs: Developing AI-based chatbots in blended learning to enhance computational thinking and self-regulated learning in mathematics education, The Eurasia Proceedings of Educational and Social Sciences 41 (2025) 11–24

    work page 2025

  24. [24]

    Peeters, V

    A. Peeters, V. Robinson, C. Rubie-Davies, Theories in use that explain adolescent help seeking and avoidance in mathematics, Journal of Educational Psychology 112 (2020) 533

    work page 2020

  25. [25]

    H. Wood, D. Wood, Help seeking, learning and contingent tutoring, Computers & Education 33 (1999) 153–169

    work page 1999

  26. [26]

    D. J. DeFeo, D. Bonin, M. Ossiander-Gobeille, Waiting and help-seeking in math tutoring exchanges, Journal of Developmental Education (2017) 14–22

    work page 2017

  27. [27]

    Kennedy, A study of students’ fears of seeking academic help from teachers, The Journal of Classroom Interaction (1997) 11–17

    E. Kennedy, A study of students’ fears of seeking academic help from teachers, The Journal of Classroom Interaction (1997) 11–17

    work page 1997

  28. [28]

    P. Long, K. D. Neff, Self-compassion is associated with reduced self-presentation concerns and increased student communication behavior, Learning and Individual Differences 67 (2018) 223–231

    work page 2018

  29. [29]

    A. L. Beerwinkle, The use of learning analytics and the potential risk of harm for K–12 students participating in digital learning environments, Educational Technology Research and Development 69 (2021) 327–330

    work page 2021