Pith. sign in

REVIEW 3 major objections 3 minor 47 references

Human-AI collaboration or obedient and often clueless AI in instruct, serve, repeat dynamics?

T0 review · 3 major / 3 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Students mostly issue commands to AI rather than collaborating, and the interaction shows no connection between problem complexity or prompt length and grades.

desk verdict The paper asks the right question about student-LLM interaction, but its headline 'Instructive pattern' rests on unreported coding reliability and a null correlation read as proof of shallow cognition. read the letter →

arxiv 2508.10919 v1 pith:57DBFU3A submitted 2025-08-03 cs.HC cs.AI

Mohammed Saqr , Kamila Misiejuk , Sonsoles López-Pernas This is my paper
classification cs.HCcs.AI
keywords human-AIcollaborationLLMinteractionpatternsinstructivedynamicstransitionnetworkanalysissequencecognitivedeptheducationtechnologypromptcomplexity
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to show that current large language models, when used by students on a complex problem, do not behave as collaborative partners: the typical interaction is a one-way stream of instructions from student to model, with the model faithfully executing each order. The authors coded student-AI chat logs and tracked how interaction types evolved, finding a dominant "Instructive" pattern marked by iterative ordering and frequent misalignment between prompts and outputs. They also report that neither assignment complexity nor prompt length correlated with final grades, which they interpret as evidence that the interactions lacked cognitive depth. The stakes are practical: if this picture holds, then expecting LLMs to stimulate or align with student thinking by default is misplaced, and AI design should prioritize cognitive alignment and genuine negotiation.

What carries the argument

The argument is carried by a qualitative coding scheme that classifies each student-AI exchange into interaction types such as "instructive" versus "collaborative negotiation," together with transition network analysis and sequence analysis that map how one type leads to the next across the thread. Partial correlation networks and chi-square tests with mosaic plots then connect interaction patterns to assignment complexity and grades. The "Instructive pattern" is the central object: it names the dominant, repeated trajectory in which students give orders and the model serves them without negotiation.

What would settle it

A re-coding study of comparable student-AI threads using an independent codebook that counts concrete negotiation behaviors—challenging the model's answer, asking for justification, modifying a suggested solution—would falsify the dominance claim if such behaviors appear in a substantial share of interactions. The null-correlation claim would be falsified by a study with a validated depth measure (e.g., reasoning traces or revision quality) showing that prompt complexity or task difficulty positively predicts depth even when grades do not.

Watch

Extended reading notes

Core claim

The paper's central claim is that in a cognitively demanding task, human-AI interaction is mostly an "instruct, serve, repeat" loop rather than collaboration. Students issued instructions; the AI served them; the next step repeated the sequence, so the overall trajectory was iterative ordering, not collaborative negotiation such as questioning assumptions, weighing alternatives, or jointly framing the problem. The authors found long threads in which student prompts and AI outputs were misaligned, which they describe as a lack of synergy, and a null correlation between assignment complexity, prompt length, and grades. Their conclusion is that LLMs, optimized to follow instructions rather than to be cognitive partners, currently make it harder, not easier, for students to engage in cognitively stimulating or aligned collaboration.

Load-bearing premise

The load-bearing premise is that the qualitative coding categories—especially the line between "instructive" and "collaborative negotiation"—are valid and reliable enough to measure collaboration, and that the absence of correlations between complexity, prompt length, and grades can be interpreted as evidence of shallow cognitive depth rather than as a measurement artifact.

Editorial extensions

If this is right

  • If the finding holds, course designs that hand students a chatbot and expect collaboration will need scaffolding that explicitly teaches negotiation moves.
  • LLM interfaces could be changed to ask clarifying questions, request justifications, or flag contradictions, shifting the default from instruction-following to joint problem-solving.
  • The reported null correlations imply that prompt length and assignment difficulty are poor proxies for the depth of AI-assisted work; educators should look at process logs, not final prompts or grades.
  • Sequence and transition analyses of interaction logs could become a routine diagnostic for whether AI use is collaborative or merely directive.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • My reading: the null correlation between complexity and grades may say less about cognitive depth and more about grades measuring the final artifact rather than the thinking process; a process-level outcome such as revision quality might correlate with complexity even when grades do not.
  • My reading: the dominant instructive pattern could partly reflect students' habits from search engines, where one query is followed by an answer; comparing the same chatbot with an interface that prompts reflection would test whether the model's design or the student's habit causes the pattern.
  • My reading: "lack of synergy" is not always a failure; for routine subtasks, quick instruction-following may be appropriate. A task-specific threshold for when negotiation is needed would make the critique more actionable.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 3 minor

Summary. This manuscript reports an observational study of student interactions with an LLM while solving a complex problem. The abstract describes qualitative coding of interaction turns, followed by transition network analysis, sequence analysis, partial correlation networks, chi-square tests, and Pearson-residual shaded mosaic plots. The headline findings are a dominant "Instructive" pattern characterized by iterative ordering rather than collaborative negotiation, long threads showing misalignment between prompts and outputs, and null correlations between assignment complexity, prompt length, and student grades, which the authors interpret as evidence of a lack of cognitive depth. The paper concludes that current LLMs are optimized for instruction-following rather than cognitive partnership. The review copy of the full text is badly corrupted by character-encoding errors, so the assessment below relies on the abstract and on the methodological and evidentiary claims it reports.

Significance. If the results hold, the paper would make a useful contribution by shifting from simple frequency counts of human-AI interaction to an analysis of dynamics and evolution, using sequence analysis and transition networks. The abstract offers a concrete, potentially falsifiable claim about a dominant Instructive pattern and draws attention to a real design issue: LLMs may be optimized for instruction-following rather than for collaborative cognitive engagement. However, the current evidence base is not sufficient to establish that claim, because the qualitative coding scheme is not described, no inter-rater reliability is reported, and the null correlations are interpreted as substantive evidence without effect sizes or power considerations. The manuscript is therefore not yet convincing, but the core questions and methodological direction are worth pursuing.

major comments (3)
  1. [Abstract (coding and reliability)] The central finding of a dominant "Instructive pattern" rests entirely on qualitative coding of student-AI interactions into categories such as "instructive" versus "collaborative negotiation." The abstract provides no codebook definitions, no example utterances, no decision criteria, no information about the number of coders, and no inter-rater reliability statistic (e.g., Cohen's kappa). All subsequent analyses—transition networks, sequence analysis, chi-square tests, and mosaic plots—consume these categories as measured facts. As reported, the headline distinction between "iterative ordering" and "collaborative negotiation" is not separable from coder expectation, and the study could merely reflect the coding lens rather than the interaction pattern. This is a load-bearing omission that must be addressed, either by reporting reliability and transparency of the coding procedure or by reframing the claims as exploratory.
  2. [Abstract (null correlation inference)] The claim that "no significant correlations between assignment complexity, prompt length, and student grades" suggests "a lack of cognitive depth, or effect of problem difficulty" is an over-reading of null results. The abstract reports no effect sizes, confidence intervals, sample size, or power analysis. An absence of significant correlation in an observational sample can equally arise from low statistical power, restricted range in assignment complexity, noisy grade measures, or a poorly chosen proxy such as prompt length. As written, the conclusion treats a null result as confirming evidence for a substantive cognitive interpretation, which is not warranted by the reported statistics. The authors should either report formal equivalence testing or effect-size bounds, or substantially soften the cognitive-depth conclusion.
  3. [Abstract (causal framing)] The conclusion that "current LLMs, optimized for instruction-following rather than cognitive partnership, compound their capability to act as cognitively stimulating or aligned collaborators" makes a mechanistic and quasi-causal claim about LLM design goals and their effects. The study as described is observational: it does not manipulate model optimization objectives, compare multiple models with controlled capacities, or randomly assign conditions. The abstract's evidence can support, at most, a descriptive statement about observed interaction patterns in a particular setting. The causal language about LLMs being "optimized for instruction-following" should be clearly separated from the empirical findings, or the study needs an appropriate design to support such a claim.
minor comments (3)
  1. [Abstract] The phrase "Person-residual shaded Mosaic plots" should read "Pearson-residual shaded mosaic plots."
  2. [Abstract] The abstract contains informal or ungrammatical constructions, for example "Oftentimes, students engaged in long threads that showed misalignment between their prompts and AI output that exemplified a lack of synergy" and "compound their capability to act as cognitively stimulating or aligned collaborators." These should be revised for clarity and precision.
  3. [Full text] The provided full text is not legible due to character-encoding corruption; a clean, readable version is needed for review and for readers.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the analysis is observational and self-contained, and the coding concerns raised are validity issues, not circularity.

full rationale

The paper reports a qualitative coding of student–AI interactions followed by transition network analysis, sequence analysis, chi-square tests, and mosaic plots. No equation, fitted parameter, or derived prediction is presented; the findings are descriptive characterizations of observed interaction patterns. The claim that an 'Instructive pattern' dominates is a reading of coded data, not a quantity derived from itself by construction. The null correlations between assignment complexity, prompt length, and grades are reported as empirical results, and their interpretation as indicating 'lack of cognitive depth' may be an inferential overreach, but an over-reading of a null result is a correctness or validity concern, not circularity. No self-citation is invoked to justify a load-bearing premise, and no imported uniqueness theorem or ansatz is present. The absence of inter-rater reliability and codebook details weakens evidentiary support but does not make the argument circular. Accordingly, per the instructions to reserve circularity findings for demonstrable reductions to inputs, the appropriate score is 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The paper introduces no fitted parameters or invented entities. Its load-bearing assumptions are the validity of the qualitative coding scheme, the completeness of the interaction logs, and the construct validity of complexity and grades as measures of cognitive depth. These assumptions are not documented in the abstract.

assumptions (3)
  • domain assumption The qualitative codebook reliably distinguishes instructive, collaborative, and other interaction modes from prompt text.
    The central conclusion about dominance of instructive patterns depends on the validity of these categories, but no inter-rater reliability or codebook validation is mentioned in the abstract.
  • domain assumption The logged student-AI exchanges are a complete and representative record of the collaboration process.
    The abstract does not describe how logs were captured, whether system prompts or hidden context affect coding, or whether all relevant interactional context was visible to coders.
  • domain assumption Assignment complexity can be meaningfully operationalized, and grades are a valid outcome measure for cognitive depth.
    The null correlation between complexity, prompt length, and grades is interpreted as a lack of cognitive depth, which presupposes these variables were measured with sufficient construct validity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Human-AI collaboration or obedient and often clueless AI in instruct, serve, repeat dynamics?." pith.science (2026). https://pith.science/paper/57DBFU3A

@misc{pith2026250810919,
  author       = {Pith},
  title        = {Pith review of: Human-AI collaboration or obedient and often clueless AI in instruct, serve, repeat dynamics?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/57DBFU3A}},
  note         = {Machine review of arXiv:2508.10919}
}
read the original abstract

While research on human-AI collaboration exists, it mainly examined language learning and used traditional counting methods with little attention to evolution and dynamics of collaboration on cognitively demanding tasks. This study examines human-AI interactions while solving a complex problem. Student-AI interactions were qualitatively coded and analyzed with transition network analysis, sequence analysis and partial correlation networks as well as comparison of frequencies using chi-square and Person-residual shaded Mosaic plots to map interaction patterns, their evolution, and their relationship to problem complexity and student performance. Findings reveal a dominant Instructive pattern with interactions characterized by iterative ordering rather than collaborative negotiation. Oftentimes, students engaged in long threads that showed misalignment between their prompts and AI output that exemplified a lack of synergy that challenges the prevailing assumptions about LLMs as collaborative partners. We also found no significant correlations between assignment complexity, prompt length, and student grades suggesting a lack of cognitive depth, or effect of problem difficulty. Our study indicates that the current LLMs, optimized for instruction-following rather than cognitive partnership, compound their capability to act as cognitively stimulating or aligned collaborators. Implications for designing AI systems that prioritize cognitive alignment and collaboration are discussed.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

47 extracted references · 39 canonical work pages

  1. [1]

    The Turing transformation: Artificial intelligence, intelligence augmentation, and skill premiums

    Agrawal, A., Gans, J., and Goldfarb, A.. The Turing transformation: Artificial intelligence, intelligence augmentation, and skill premiums. In Special Issue 5: Grappling With the Generative AI Revolution, Special5. 2024. https://doi.org/10.1162/99608f92.35a2f3ff

    work page doi:10.1162/99608f92.35a2f3ff 2024

  2. [2]

    A research agenda for hybrid intelligence: Augmenting human intellect with collaborative, adaptive, responsible, and explainable artificial intelligence

    Akata, Z., Balliet, D., de Rijke, M., Dignum, F., Dignum, V., Eiben, G., Fokkens, A., Grossi, D., Hindriks, K., Hoos, H., Hung, H., Jonker, C., Monz, C., Neerincx, M., Oliehoek, F., Prakken, H., Schlobach, S., van der Gaag, L., van Harmelen, F., … Welling, M.. A research agenda for hybrid intelligence: Augmenting human intellect with collaborative, adapti...

    work page 2020

  3. [3]

    and Sidan, E

    Akçapınar, G. and Sidan, E.. AI chatbots in programming education: guiding success or encouraging plagiarism. In Discover Artificial Intelligence, 4(1). 2024. https://doi.org/10.1007/s44163-024-00203-7

    doi:10.1007/s44163-024-00203-7 2024

  4. [4]

    Mean difference, standardized mean difference (SMD), and their use in meta-analysis: As simple as it gets

    Andrade, C.. Mean difference, standardized mean difference (SMD), and their use in meta-analysis: As simple as it gets. In The Journal of Clinical Psychiatry, 81(5). 2020. https://doi.org/10.4088/JCP.20f13681

    work page doi:10.4088/jcp.20f13681 2020

  5. [5]

    TheBigPromptLibrary: A collection of prompts, system prompts and LLM instructions

    Bachaalany, E.. TheBigPromptLibrary: A collection of prompts, system prompts and LLM instructions. In Github. 2024. https://github.com/0xeb/TheBigPromptLibrary

    work page 2024

  6. [6]

    and Kim, S

    Bae, J. and Kim, S.. Identifying and ranking influential spreaders in complex networks by neighborhood coreness. In Physica A: Statistical Mechanics and Its Applications, 395, 549–559. 2014

    work page 2014

  7. [7]

    Two decades of artificial intelligence in education: Contributors, collaborations, research Topics, challenges, and future directions

    Chen, X., Zou, D., Xie, H., Cheng, G., and Liu, C.. Two decades of artificial intelligence in education: Contributors, collaborations, research Topics, challenges, and future directions. In Journal of Educational Technology & Society, 25(1), 28–47. 2022

    work page 2022

  8. [8]

    W., Oudah, M., Tennom, Ishowo-Oloko, F., Abdallah, S., Bonnefon, J.-F., Cebrian, M., Shariff, A., Goodrich, M

    Crandall, J. W., Oudah, M., Tennom, Ishowo-Oloko, F., Abdallah, S., Bonnefon, J.-F., Cebrian, M., Shariff, A., Goodrich, M. A., and Rahwan, I.. Cooperating with machines. In Nature Communications, 9(1), 233. 2018

    work page 2018

Show all 47 references
  1. [9]

    and Burke, D

    Crompton, H. and Burke, D.. Artificial intelligence in higher education: the state of the field. In International Journal of Educational Technology in Higher Education, 20(1). 2023. https://doi.org/10.1186/s41239-023-00392-8

    2023 doi

  2. [10]

    Epskamp, S., Borsboom, D., and Fried, E. I.. Estimating psychological networks and their accuracy: A tutorial paper. In Behavior Research Methods, 50(1), 195–212. 2018

    2018

  3. [11]

    J., Mõttus, R., and Borsboom, D

    Epskamp, S., Waldorp, L. J., Mõttus, R., and Borsboom, D.. The Gaussian Graphical Model in Cross-Sectional and Time-Series Data. In Multivariate Behavioral Research, 53(4), 453–480. 2018

    2018

  4. [12]

    W., and Gašević, D

    Fan, Y., Tan, Y., Raković, M., Wang, Y., Cai, Z., Shaffer, D. W., and Gašević, D.. Dissecting learning tactics in MOOC using ordered network analysis. In Journal of Computer Assisted Learning. 2022. https://doi.org/10.1111/jcal.12735

    2022 doi

  5. [13]

    Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance

    Fan, Y., Tang, L., Le, H., Shen, K., Tan, S., Zhao, Y., Shen, Y., Li, X., and Gašević, D.. Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance. In BJET. 2024. https://doi.org/10.1111/bjet.13544

    2024 doi

  6. [14]

    S., and Studer, M

    Gabadinho, A., Ritschard, G., Müller, N. S., and Studer, M.. Analyzing and Visualizing State Sequences in R with TraMineR. In Journal of Statistical Software, 40(4). 2011. https://doi.org/10.18637/jss.v040.i04

    2011 doi

  7. [15]

    A., Choo, K

    Gao, J., Gebreegziabher, S. A., Choo, K. T. W., Li, T. J.-J., Perrault, S. T., and Malone, T. W.. A Taxonomy for Human-LLM Interaction Modes: An Initial Exploration. In Extended Abstracts of the CHI Conference on Human Factors in Computing Systems, 1–11. 2024

    2024

  8. [16]

    Prompt Engineering with ChatGPT: A Guide for Academic Writers

    Giray, L.. Prompt Engineering with ChatGPT: A Guide for Academic Writers. In Annals of Biomedical Engineering, 51(12), 2629–2633. 2023

    2023

  9. [17]

    A modern approach to transition analysis and process mining with Markov models: A tutorial with R

    Helske, J., Helske, S., Saqr, M., López-Pernas, S., and Murphy, K.. A modern approach to transition analysis and process mining with Markov models: A tutorial with R. In Learning analytics methods and tutorials: A practical guide using R, in–press. 2024

    2024

  10. [18]

    Human-AI Complementarity in Hybrid Intelligence Systems: A Structured Literature Review

    Hemmer, P., Schemmer, M., Vössing, M., and Kühl, N.. Human-AI Complementarity in Hybrid Intelligence Systems: A Structured Literature Review. In PACIS 2021. 2021. https://aisel.aisnet.org/pacis2021/78/

    2021

  11. [19]

    Hybrid intelligence: Human–AI coevolution and learning

    Järvelä, S., Zhao, G., Nguyen, A., and Chen, H.. Hybrid intelligence: Human–AI coevolution and learning. In British Journal of Educational Technology: Journal of the Council for Educational Technology. 2025. https://doi.org/10.1111/bjet.13560

    2025 doi

  12. [20]

    Differences in student-AI interaction process on a drawing task: Focusing on students’ attitude towards AI and the level of drawing skills

    Kim, J., Ham, Y., and Lee, S.-S.. Differences in student-AI interaction process on a drawing task: Focusing on students’ attitude towards AI and the level of drawing skills. In Australasian Journal of Educational Technology, 40(1), 19–41. 2024

    2024

  13. [21]

    Kim, J., Lee, H., and Cho, Y. H.. Learning design to support student-AI collaboration: perspectives of leading teachers for AI in education. In Education and Information Technologies, 27(5), 6069–6104. 2022

    2022

  14. [22]

    Students’ prompt patterns and its effects in AI-assisted academic writing: Focusing on students’ level of AI literacy

    Kim, J., Yu, S., Lee, S.-S., and Detrick, R.. Students’ prompt patterns and its effects in AI-assisted academic writing: Focusing on students’ level of AI literacy. In Journal of Research on Technology in Education, 1–18. 2025

    2025

  15. [23]

    A map of exploring human interaction patterns with LLM: Insights into collaboration and creativity

    Li, J., Li, J., and Su, Y.. A map of exploring human interaction patterns with LLM: Insights into collaboration and creativity. In Artificial Intelligence in HCI, 60–85. 2024

    2024

  16. [24]

    H., and Shah, C

    Liao, L., Yang, G. H., and Shah, C.. Proactive conversational agents in the post-ChatGPT world. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval. 2023. https://doi.org/10.1145/3539618.3594250

    2023

  17. [25]

    M., Yang, S., Furze, L., and Dawson, P

    Lodge, J. M., Yang, S., Furze, L., and Dawson, P.. It’s not like a calculator, so what is the relationship between learners and generative artificial intelligence? In Learning Research and Practice, 9(2), 117–124. 2023

    2023

  18. [26]

    tna: Transition Network Analysis (TNA) [Data set]

    López-Pernas, S., Tikka, S., and Saqr, M.. tna: Transition Network Analysis (TNA) [Data set]. In CRAN: Contributed Packages. 2024. https://doi.org/10.32614/cran.package.tna

    2024 doi

  19. [27]

    Understanding student perceptions of artificial intelligence as a teammate

    Marrone, R., Zamecnik, A., Joksimovic, S., Johnson, J., and De Laat, M.. Understanding student perceptions of artificial intelligence as a teammate. In Technology Knowledge and Learning. 2024. https://doi.org/10.1007/s10758-024-09780-z

    2024 doi

  20. [28]

    Detection of Learning Strategies: A Comparison of Process, Sequence and Network Analytic Approaches

    Matcha, W., Gašević, D., Ahmad Uzir, N., Jovanović, J., Pardo, A., Maldonado-Mahauad, J., and Pérez-Sanagustín, M.. Detection of Learning Strategies: A Comparison of Process, Sequence and Network Analytic Approaches. In Transforming Learning with Meaningful Technologies, 525–540. 2019

    2019

  21. [29]

    The Strucplot Framework: Visualizing Multi-way Contingency Tables with vcd

    Meyer, D., Zeileis, A., and Hornik, K.. The Strucplot Framework: Visualizing Multi-way Contingency Tables with vcd. In Journal of Statistical Software, 17, 1–48. 2007

    2007

  22. [30]

    and Wise, A

    Molenaar, I. and Wise, A. F.. Temporal aspects of learning analytics-grounding analyses in concepts of time. In The Handbook of Learning Analytics, 66–76. 2022

    2022

  23. [31]

    Co-intelligence

    Mollick, E.. Co-intelligence. In W H Allen. 2024

    2024

  24. [32]

    Networks (2nd ed.)

    Newman, M.. Networks (2nd ed.). In Oxford University Press. 2018

    2018

  25. [33]

    S., Shawe-Taylor, J., and Vespignani, A

    Pedreschi, D., Pappalardo, L., Ferragina, E., Baeza-Yates, R., Barabási, A.-L., Dignum, F., Dignum, V., Eliassi-Rad, T., Giannotti, F., Kertész, J., Knott, A., Ioannidis, Y., Lukowicz, P., Passarella, A., Pentland, A. S., Shawe-Taylor, J., and Vespignani, A.. Human-AI coevolut...

    2025

  26. [34]

    Ray, P. P.. ChatGPT: A comprehensive review on background, applications, key challenges, bias, ethics, limitations and future scope. In Internet of Things and Cyber-Physical Systems, 3, 121–154. 2023

    2023

  27. [35]

    Risko, E. F. and Gilbert, S. J.. Cognitive offloading. In Trends in Cognitive Sciences, 20(9), 676–688. 2016

    2016

  28. [36]

    Transition Network Analysis: A novel framework for modeling, visualizing, and identifying the temporal patterns of learners and learning processes

    Saqr, M., López-Pernas, S., Törmänen, T., Kaliisa, R., Misiejuk, K., and Tikka, S.. Transition Network Analysis: A novel framework for modeling, visualizing, and identifying the temporal patterns of learners and learning processes. In arXiv [cs.SI]. 2024. http://arxiv.org/abs/...

    2024 arXiv

  29. [37]

    The Impact of Student-AI Collaborative Feedback Generation on Learning Outcomes

    Singh, A., Brooks, C., and Wang, X.. The Impact of Student-AI Collaborative Feedback Generation on Learning Outcomes. In AI for Education: Bridging Innovation and Responsibility at the 38th AAAI Annual Conference on AI. 2024. https://openreview.net/pdf?id=IQZ2dcsVq0

    2024

  30. [38]

    Cognitive ease at a cost: LLMs reduce mental effort but compromise depth in student scientific inquiry

    Stadler, M., Bannert, M., and Sailer, M.. Cognitive ease at a cost: LLMs reduce mental effort but compromise depth in student scientific inquiry. In Computers in Human Behavior, 160, 108386. 2024

    2024

  31. [39]

    and Weinberger, A

    Strijbos, J.-W. and Weinberger, A.. Emerging and scripted roles in computer-supported collaborative learning. In Computers in Human Behavior, 26(4), 491–494. 2010

    2010

  32. [40]

    Bridging the gulf of envisioning: Cognitive challenges in prompt based interactions with LLMs

    Subramonyam, H., Pea, R., Pondoc, C., Agrawala, M., and Seifert, C.. Bridging the gulf of envisioning: Cognitive challenges in prompt based interactions with LLMs. In Proceedings of the CHI Conference on Human Factors in Computing Systems, 31, 1–19. 2024

    2024

  33. [41]

    Exploring human-generative AI interaction in L2 learners’ source use practices: Issues, trials, and critical reflections

    Sun, Q.. Exploring human-generative AI interaction in L2 learners’ source use practices: Issues, trials, and critical reflections. In Journal of Academic Writing, 14(1), 24–42. 2024

    2024

  34. [42]

    Investigating the effect of artificial intelligence in education (AIEd) on learning achievement: A meta-analysis and research synthesis

    Tlili, A., Saqer, K., Salha, S., and Huang, R.. Investigating the effect of artificial intelligence in education (AIEd) on learning achievement: A meta-analysis and research synthesis. In Information Development. 2025. https://doi.org/10.1177/02666669241304407

    2025 doi

  35. [43]

    D., van Bork, R., Boschloo, L., Kossakowski, J

    van Borkulo, C. D., van Bork, R., Boschloo, L., Kossakowski, J. J., Tio, P., Schoevers, R. A., Borsboom, D., and Waldorp, L. J.. Comparing network structures on three aspects: A permutation test. In Psychological Methods. 2022. https://doi.org/10.1037/met0000476

    2022 doi

  36. [44]

    Towards mutual theory of mind in human-AI interaction: How language reflects what students perceive about a virtual teaching assistant

    Wang, Q., Saha, K., Gregori, E., Joyner, D., and Goel, A.. Towards mutual theory of mind in human-AI interaction: How language reflects what students perceive about a virtual teaching assistant. In Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems. 2...

    2021

  37. [45]

    Strategic Chain-of-Thought: Guiding accurate reasoning in LLMs through strategy elicitation

    Wang, Y., Zhao, S., Wang, Z., Huang, H., Fan, M., Zhang, Y., Wang, Z., Wang, H., and Liu, T.. Strategic Chain-of-Thought: Guiding accurate reasoning in LLMs through strategy elicitation. In arXiv [cs.AI]. 2024. http://arxiv.org/abs/2409.03271

    2024 arXiv

  38. [46]

    J., Ge, L., and Gao, Z

    Xu, W., Dainoff, M. J., Ge, L., and Gao, Z.. Transitioning to human interaction with AI systems: New challenges and opportunities for HCI professionals to enable human-centered AI. In International Journal of Human-Computer Interaction, 1–25. 2022

    2022

  39. [47]

    Large language models as Markov chains

    Zekri, O., Odonnat, A., Benechehab, A., Bleistein, L., Boullé, N., and Redko, I.. Large language models as Markov chains. In arXiv [stat.ML]. 2024. http://arxiv.org/abs/2410.02724

    2024 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.