pith. machine review for the scientific record. sign in

arxiv: 2605.08486 · v1 · submitted 2026-05-08 · 💻 cs.CY

Recognition: 1 theorem link

· Lean Theorem

Teachers' Perceived Benefits and Risks of AI Across Fifty-Five Countries: An Audit of LLM Alignment and Steerability

Yan Tao , Olga Viberg , Deepak Varuvel Dennison , Zhikun Wu , Ren\'e F. Kizilcec
Authors on Pith no claims yet

Pith reviewed 2026-05-12 01:19 UTC · model grok-4.3

classification 💻 cs.CY
keywords AI in educationteacher perceptionsLLM alignmentcross-national variationTALIS surveyAI benefits and risksmodel steerabilityeducational technology policy
0
0 comments X

The pith

Large language models compress cross-national differences in how teachers perceive AI benefits and risks and overestimate both.

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

Teachers in different countries balance the benefits and risks of AI in education in measurably distinct ways, according to representative data from the OECD TALIS survey covering 55 countries and territories. The paper audits whether eight current LLMs can reproduce those country-level patterns when prompted in general or country-specific ways. The models instead flatten the observed variation, inflate both positive and negative views, and gain little from added reasoning steps or identity cues. This gap matters because LLM outputs increasingly shape professional discussions and policy ideas about AI adoption in schools. The authors therefore caution against substituting model-generated guidance for direct collection of teacher perspectives when setting global directions for educational AI.

Core claim

Using TALIS data, teachers report substantial country-to-country differences in the perceived benefits and risks of AI; when eight state-of-the-art LLMs are asked to answer the same items under general or country-specific prompting, their responses compress those differences, overestimate both benefits and risks, and show only limited improvement from higher-reasoning variants or identity prompts, with partial exceptions such as Gemini 3 Fast preserving some ranking patterns.

What carries the argument

Direct benchmarking of LLM survey responses against the TALIS international teacher sample under controlled prompting conditions.

If this is right

  • LLM outputs should not be used as substitutes for direct teacher input when designing global AI-in-education policies.
  • Some models retain partial ability to recover cross-national ranking patterns and may still support exploratory hypothesis generation.
  • Professional development materials and policy briefs that rely on LLMs risk transmitting flattened or exaggerated views of AI to teachers.
  • Steerability techniques such as country-identity prompting produce only marginal gains in matching real teacher distributions.

Where Pith is reading between the lines

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

  • Model developers could fine-tune on disaggregated TALIS-style items to reduce compression of country differences.
  • Policy bodies may need separate mechanisms for collecting localized teacher feedback rather than relying on LLM synthesis.
  • Experimental studies could test whether exposure to aligned versus misaligned LLM content changes teachers' actual classroom AI use.
  • The same audit method could be applied to other stakeholder groups such as school leaders or parents to map broader perception gaps.

Load-bearing premise

That LLM answers to survey-style questions about AI can be read as meaningful indicators of alignment with teachers' real-world perceptions across cultures.

What would settle it

A new round of surveys in which teachers from high- and low-variation countries rate LLM-generated summaries of AI benefits and risks and show no systematic difference in trust or adoption intent compared with summaries drawn from their own country's TALIS responses.

Figures

Figures reproduced from arXiv: 2605.08486 by Deepak Varuvel Dennison, Olga Viberg, Ren\'e F. Kizilcec, Yan Tao, Zhikun Wu.

Figure 1
Figure 1. Figure 1: Teacher benefit and risk perceptions of AI for education across 55 countries/territories. Black points represent [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Perceived benefits (left) and risks (right) of AI in education across countries/territories. Black circles show the mean [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Distribution of differences between LLM and human responses across countries for perceived benefits (top) and risks [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
read the original abstract

Teachers' trust in artificial intelligence (AI) in education depends on how they balance its perceived benefits and risks. Yet global discussions about scaling AI in education rely on fragmented evidence, as most studies of teachers' perceptions focus on single countries or small samples. This lack of representative cross-national evidence limits both theory building and policy development. At the same time, large language models (LLMs) are increasingly used in research, policy, and teachers' professional workflows, despite limited validation in education. To address these gaps, we conduct a large-scale audit of LLM alignment with teachers' perceptions of AI by combining representative international survey data with systematic model evaluation. Using OECD TALIS data from 55 countries and territories, we measure cross-national variation in teachers' perceived benefits and risks of AI. We then benchmark responses from eight state-of-the-art LLMs across four providers under both general and country-specific prompting, comparing higher- and lower-reasoning models. Results reveal substantial cross-national variation in teacher perceptions that is not reliably reflected in LLM outputs. Models compress country differences, overestimate both benefits and risks, and show limited gains from identity prompting or enhanced reasoning. This misalignment matters because LLM-generated guidance and professional discourse increasingly shape how teachers learn about and discuss AI, potentially influencing trust and future adoption decisions. Our findings caution against treating LLM outputs as substitutes for direct engagement with teachers when informing global AI-in-education initiatives. At the same time, some models (e.g., Gemini 3 Fast) partially capture cross-national ranking patterns, suggesting a complementary role in hypothesis generation and exploratory comparative analysis.

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 / 2 minor

Summary. The paper audits LLM alignment with teachers' perceptions of AI benefits and risks by combining OECD TALIS survey data from 55 countries/territories with systematic evaluations of eight state-of-the-art LLMs. It reports substantial cross-national variation in teacher perceptions that LLMs fail to capture reliably: models compress country differences, overestimate both benefits and risks, and exhibit only limited improvements under country-specific identity prompting or enhanced reasoning. The work concludes that LLM outputs should not substitute for direct teacher engagement in global AI-in-education initiatives, while noting partial capture of ranking patterns by some models (e.g., Gemini 3 Fast).

Significance. If the empirical comparisons hold after methodological clarification, the study offers a valuable large-scale benchmark (55 countries, multiple providers and reasoning levels) that cautions against over-reliance on LLMs for cross-cultural perception modeling in education. It provides falsifiable, quantitative evidence of misalignment that can inform both theory on AI trust and practical guidelines for using LLMs in policy or professional development. The scale and direct comparison to representative survey data are strengths that distinguish it from smaller single-country studies.

major comments (2)
  1. [Methods] Methods section: the manuscript provides no details on statistical methods for quantifying overestimation or compression of country differences, TALIS response rates, exact prompting templates (including country-specific identity prompts), model versions, temperature settings, or how Likert-scale equivalence was established between human and LLM responses. These omissions are load-bearing for the central claim that observed misalignment is substantive rather than methodological.
  2. [Results] Results and Discussion: the claims of limited gains from identity prompting and enhanced reasoning, and of scale compression, rest on the unvalidated assumption that LLM outputs can be directly compared to TALIS Likert items as alignment proxies. Without reported checks for prompt artifacts, response anchoring, or cultural response-style controls, the reported patterns could arise from how LLMs interpret survey-style prompts rather than gaps in model knowledge.
minor comments (2)
  1. [Abstract] Abstract: the distinction between 'higher- and lower-reasoning models' is stated but not defined; the main text should explicitly list which models belong to each category and the criteria used.
  2. [Methods] The paper would benefit from a table summarizing the eight LLMs, their providers, and the four prompting conditions to improve readability of the experimental design.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback, which has helped us strengthen the methodological transparency and robustness of our claims. We have revised the manuscript to address both major comments and provide the requested details and validations. Our point-by-point responses follow.

read point-by-point responses

  1. Referee: [Methods] Methods section: the manuscript provides no details on statistical methods for quantifying overestimation or compression of country differences, TALIS response rates, exact prompting templates (including country-specific identity prompts), model versions, temperature settings, or how Likert-scale equivalence was established between human and LLM responses. These omissions are load-bearing for the central claim that observed misalignment is substantive rather than methodological.

    Authors: We agree these details are essential. In the revised manuscript we have expanded the Methods section with a new subsection on 'Quantification of Misalignment and Prompting Protocol.' This now specifies: (1) statistical methods, including mean absolute differences and Cohen's d for overestimation, and country-level variance ratios plus range compression metrics for cross-national differences; (2) TALIS response rates, drawn from the official OECD documentation (overall participation >75% in 52 of 55 entities, with country-specific figures now tabulated in Appendix A); (3) verbatim prompting templates, including the exact country-identity phrasing (e.g., 'You are a teacher in [Country] responding to the following survey...'), provided in full in Appendix B; (4) model versions (GPT-4o-2024-08, Claude-3.5-Sonnet-20240620, Gemini-1.5-Pro, etc.) and temperature settings (0.0 for all deterministic runs, 0.7 for exploratory checks); and (5) Likert equivalence, achieved by instructing models to output integer scores on the identical 1-5 scale with the same verbal anchors used in TALIS. These additions directly substantiate that the reported misalignment is not an artifact of omitted protocol details. revision: yes

  2. Referee: [Results] Results and Discussion: the claims of limited gains from identity prompting and enhanced reasoning, and of scale compression, rest on the unvalidated assumption that LLM outputs can be directly compared to TALIS Likert items as alignment proxies. Without reported checks for prompt artifacts, response anchoring, or cultural response-style controls, the reported patterns could arise from how LLMs interpret survey-style prompts rather than gaps in model knowledge.

    Authors: We acknowledge the need for explicit validation of the proxy assumption. The original manuscript already contained robustness tests across prompt variations, but we have now added a dedicated 'Validation of Survey-Style Prompting' paragraph in Results. This reports: (a) re-runs with three alternative phrasings (neutral, role-play, and forced-choice) showing stable overestimation and compression patterns (correlation of country rankings >0.85 across phrasings); (b) checks for anchoring by comparing mean responses under different scale instructions; and (c) discussion of cultural response-style limitations, noting that while full controls are infeasible without matched human experiments, the partial capture of known TALIS rankings by Gemini 3 Fast and the consistency across eight independent models support that the misalignment reflects knowledge gaps rather than prompt artifacts alone. We have also added this as an explicit limitation in the Discussion. The revision is partial because core robustness evidence was already present but has been expanded and foregrounded. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical audit relies on external TALIS data and model outputs

full rationale

The paper performs a direct empirical comparison between representative OECD TALIS survey responses from 55 countries and LLM outputs under varied prompting conditions. No equations, fitted parameters, self-definitional constructs, or load-bearing self-citations reduce any central claim (e.g., compression of country differences or overestimation of benefits/risks) to a tautology or input-derived result. The derivation chain consists of independent data collection, prompting protocols, and statistical benchmarking that remain falsifiable against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the assumption that survey items validly measure perceived benefits and risks and that LLM prompt responses constitute a comparable signal; no new free parameters or invented entities are introduced.

axioms (2)
  • domain assumption TALIS survey responses accurately reflect teachers' genuine perceptions of AI benefits and risks
    Invoked when interpreting cross-national variation as real differences in teacher views.
  • domain assumption LLM outputs under the tested prompts can be scored for alignment against human survey distributions
    Required to treat model responses as comparable to teacher data.

pith-pipeline@v0.9.0 · 5613 in / 1355 out tokens · 41707 ms · 2026-05-12T01:19:45.576886+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

62 extracted references · 62 canonical work pages

  1. [1]

    Saleh Afroogh, Ali Akbari, Emmie Malone, Mohammadali Kargar, and Hananeh Alambeigi. 2024. Trust in AI: progress, challenges, and future directions.Hu- manities and Social Sciences Communications11, 1 (2024), 1–30

    work page 2024

  2. [2]

    Guilherme FCF Almeida, José Luiz Nunes, Neele Engelmann, Alex Wiegmann, and Marcelo De Araújo. 2024. Exploring the psychology of LLMs’ moral and legal reasoning.Artificial Intelligence333 (2024), 104145

    work page 2024

  3. [3]

    Arnav Arora, Lucie-aimée Kaffee, and Isabelle Augenstein. 2023. Probing Pre- Trained Language Models for Cross-Cultural Differences in Values. InProceedings of the First Workshop on Cross-Cultural Considerations in NLP (C3NLP), Sunipa Dev, Vinodkumar Prabhakaran, David Ifeoluwa Adelani, Dirk Hovy, and Luciana Benotti (Eds.). Association for Computational ...

    work page doi:10.18653/v1/2023.c3nlp-1.12 2023

  4. [4]

    Mohammad Atari, Mona J Xue, Peter S Park, Damián E Blasi, and Joseph Henrich

    work page

  5. [5]

    PsyArXiv preprint , year=

    Which Humans? PsyArXiv. doi:10.31234/osf.io/5b26t

    work page doi:10.31234/osf.io/5b26t

  6. [6]

    Ryan S Baker, Amy E Ogan, Michael Madaio, and Erin Walker. 2019. Culture in Computer-Based Learning Systems: Challenges and Opportunities.Computer- Based Learning in Context1, 1 (2019), 1–13

    work page 2019

  7. [7]

    Duncan Wadsworth, and Hanna Wallach

    Solon Barocas, Anhong Guo, Ece Kamar, Jacquelyn Krones, Meredith Ringel Morris, Jennifer Wortman Vaughan, W. Duncan Wadsworth, and Hanna Wallach

    work page

  8. [8]

    In Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society

    Designing Disaggregated Evaluations of AI Systems: Choices, Consider- ations, and Tradeoffs. InProceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society(Virtual Event, USA)(AIES ’21). Association for Computing Machinery, New York, NY, USA, 368–378. doi:10.1145/3461702.3462610

    work page doi:10.1145/3461702.3462610 2021

  9. [9]

    James Calleja and Patrick Camilleri. 2025. Primary school teachers’ perceptions towards the use of generative AI in teaching using lesson study.International Journal for Lesson and Learning Studies14, 3 (2025), 237–252. doi:10.1108/IJLLS- 11-2024-0268

    work page doi:10.1108/ijlls- 2025

  10. [10]

    Yong Cao, Li Zhou, Seolhwa Lee, Laura Cabello, Min Chen, and Daniel Hersh- covich. 2023. Assessing Cross-Cultural Alignment between ChatGPT and Human Societies: An Empirical Study. InProceedings of the First Workshop on Cross- Cultural Considerations in NLP (C3NLP), Sunipa Dev, Vinodkumar Prabhakaran, David Ifeoluwa Adelani, Dirk Hovy, and Luciana Benotti...

    work page doi:10.18653/v1/2023 2023

  11. [11]

    Ismail Celik, Muhterem Dindar, Hanni Muukkonen, and Sanna Järvelä. 2022. The promises and challenges of artificial intelligence for teachers: A systematic review of research.TechTrends66, 4 (2022), 616–630. doi:10.1007/s11528-022-00715-y

    work page doi:10.1007/s11528-022-00715-y 2022

  12. [12]

    Ismail Celik, Sini Kontkanen, Jari Laru, and Alanur Ahsen Dalyanci. 2026. Co- constructing adaptive lesson plans with GenAI: Pre-service teachers’ Intelligent- TPACK and prompt engineering strategies.Computers & Education241 (2026), 105485. doi:10.1016/j.compedu.2025.105485

    work page doi:10.1016/j.compedu.2025.105485 2026

  13. [13]

    Shu-Yang Chien, Michael Lewis, Katia Sycara, Jui-Shiang Liu, and Asli Kumru

    work page

  14. [14]

    ACM Transactions on Interactive Intelligent Systems8, 4 (2018), 1–31

    The effect of culture on trust in automation: reliability and workload. ACM Transactions on Interactive Intelligent Systems8, 4 (2018), 1–31. doi:10.1145/ 3230736

    work page 2018

  15. [15]

    Iris Delikoura, Yi R Fung, and Pan Hui. 2025. From superficial out- puts to superficial learning: Risks of large language models in education. arXiv:2509.21972 [cs.CY] https://doi.org/10.48550/arXiv.2509.21972

    work page doi:10.48550/arxiv.2509.21972 2025

  16. [16]

    Haoxiang Fan, Guanzheng Chen, Xingbo Wang, and Zhenhui Peng. 2024. Lesson- Planner: Assisting Novice Teachers to Prepare Pedagogy-Driven Lesson Plans with Large Language Models. InProceedings of the 37th Annual ACM Symposium on User Interface Software and Technology(Pittsburgh, PA, USA)(UIST ’24). As- sociation for Computing Machinery, New York, NY, USA, ...

    work page doi:10.1145/3654777.3676390 2024

  17. [17]

    Ozan Filiz, Mehmet Haldun Kaya, and Tufan Adiguzel. 2025. Teachers and AI: Understanding the factors influencing AI integration in K-12 education.Education and Information Technologies30 (2025), 17931–17967. doi:10.1007/s10639-025- 13463-2

    work page doi:10.1007/s10639-025- 2025

  18. [18]

    Yao Fu, Litu Ou, Mingyu Chen, Yuhao Wan, Hao Peng, and Tushar Khot. 2023. Chain-of-Thought Hub: A Continuous Effort to Measure Large Language Models’ Reasoning Performance. arXiv:2305.17306 [cs.CL] https://arxiv.org/abs/2305. 17306

    work page arXiv 2023

  19. [19]

    Yuan Gao, Dokyun Lee, Gordon Burtch, and Sina Fazelpour. 2025. Take caution in using LLMs as human surrogates.Proceedings of the National Academy of Sciences122, 24 (2025), e2501660122

    work page 2025

  20. [20]

    Olga Arranz Garcia, María del Carmen Romero García, and Vidal Alonso-Secades

    work page

  21. [21]

    Perceptions, strategies, and challenges of teachers in the integration of artificial intelligence in primary education: A systematic review.Journal of Information Technology Education: Research24 (2025), 006

    work page 2025

  22. [22]

    Anastasia Gouseti, Fiona James, Lee Fallin, and Kevin Burden. 2025. The ethics of using AI in K-12 education: A systematic literature review.Technology, Pedagogy and Education34, 2 (2025), 161–182

    work page 2025

  23. [23]

    2023.Guidance for generative AI in education and research

    Wayne Holmes, Fengchun Miao, et al. 2023.Guidance for generative AI in education and research. Unesco Publishing

    work page 2023

  24. [24]

    Wayne Holmes and Ilkka Tuomi. 2022. State of the art and practice in AI in education.European journal of education57, 4 (2022), 542–570

    work page 2022

  25. [25]

    Jie Huang and Kevin Chen-Chuan Chang. 2023. Towards Reasoning in Large Language Models: A Survey. InFindings of the Association for Computational Linguistics: ACL 2023, Anna Rogers, Jordan Boyd-Graber, and Naoaki Okazaki (Eds.). Association for Computational Linguistics, Toronto, Canada, 1049–1065. doi:10.18653/v1/2023.findings-acl.67

    work page doi:10.18653/v1/2023.findings-acl.67 2023

  26. [26]

    Ruiping Huang, Yue Yin, Na Zhou, and Fei Lang. 2025. Artificial intelligence in K- 12 education: An umbrella review.Computers and Education: Artificial Intelligence (2025), 100519

    work page 2025

  27. [27]

    Ben Hutchinson, Negar Rostamzadeh, Christina Greer, Katherine Heller, and Vinodkumar Prabhakaran. 2022. Evaluation Gaps in Machine Learning Prac- tice. InProceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency(Seoul, Republic of Korea)(FAccT ’22). Association for Computing Machinery, New York, NY, USA, 1859–1876. doi:10.1145/3...

    work page doi:10.1145/3531146.3533233 2022

  28. [28]

    Zhuoren Jiang, Biao Huang, Jianan Ge, Chenxi Lin, Yueqian Xu, and Jianxing Yu

    work page

  29. [29]

    Simulating social perception with large language models: perceptions of China’s common prosperity.Journal of Chinese Governance(2025), 1–29

    work page 2025

  30. [30]

    Alexander John Karran, Patrick Charland, Joé Trempe-Martineau, Ana Ortiz de Guinea Lopez de Arana, Anne-Marie Lesage, Sylvain Senecal, and Pierre- Majorique Leger. 2025. Multi-stakeholder perspective on responsible artificial intelligence and acceptability in education.npj Science of Learning10, 1 (2025), 44

    work page 2025

  31. [31]

    Mehdi Khamassi, Marceau Nahon, and Raja Chatila. 2024. Strong and weak alignment of large language models with human values.Scientific Reports14, 1 (2024), 19399

    work page 2024

  32. [32]

    Ariba Khan, Stephen Casper, and Dylan Hadfield-Menell. 2025. Randomness, Not Representation: The Unreliability of Evaluating Cultural Alignment in LLMs. InProceedings of the 2025 ACM Conference on Fairness, Accountability, and Trans- parency (FAccT ’25). Association for Computing Machinery, New York, NY, USA, 2151–2165. doi:10.1145/3715275.3732147

    work page doi:10.1145/3715275.3732147 2025

  33. [33]

    Aidan Kierans and Shiri Dori-Hacohen. 2026. Position: Aligning AI requires automating reasoning norms. SSRN. doi:10.2139/ssrn.5399451 Available at SSRN

    work page doi:10.2139/ssrn.5399451 2026

  34. [34]

    Juhee Kim. 2025. Perceptions and preparedness of K-12 educators in adopting generative AI.Research in Learning Technology33 (2025), 3448. doi:10.25304/rlt. v33.3448

    work page doi:10.25304/rlt 2025

  35. [35]

    René F Kizilcec. 2024. To advance AI use in education, focus on understanding educators.International Journal of Artificial Intelligence in Education34, 1 (2024), 12–19

    work page 2024

  36. [36]

    Leidner and Timothy Kayworth

    Dorothy E. Leidner and Timothy Kayworth. 2006. A review of culture in infor- mation systems research: Toward a theory of information technology culture conflict.MIS Quarterly30, 2 (2006), 357–399. doi:10.2307/25148735

    work page doi:10.2307/25148735 2006

  37. [37]

    Zhuoran Lu, Gionnieve Lim, and Ming Yin. 2025. Understanding the Effects of Large Language Model (LLM)-driven Adversarial Social Influences in Online Information Spread. InProceedings of the Extended Abstracts of the CHI Conference on Human Factors in Computing Systems (CHI EA ’25). Association for Computing Machinery, New York, NY, USA, Article 555, 7 pa...

    work page doi:10.1145/3706599.3720019 2025

  38. [38]

    Sandra C Matz, Jacob D Teeny, Sumer S Vaid, Heinrich Peters, Gabriella M Harari, and Moran Cerf. 2024. The potential of generative AI for personalized persuasion at scale.Scientific Reports14, 1 (2024), 4692

    work page 2024

  39. [39]

    Timothy R McIntosh, Tong Liu, Teo Susnjak, Paul Watters, and Malka N Hal- gamuge. 2024. A reasoning and value alignment test to assess advanced gpt reasoning.ACM Transactions on Interactive Intelligent Systems14, 3 (2024), 1–37

    work page 2024

  40. [40]

    Ryan, Alan Ritter, and Wei Xu

    Tarek Naous, Michael J. Ryan, Alan Ritter, and Wei Xu. 2024. Having beer after prayer? Measuring cultural bias in large language models. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Association for Computational Linguistics, Bangkok, Thailand, 16366–16393

    work page 2024

  41. [41]

    Tanya Nazaretsky, Moriah Ariely, Mutlu Cukurova, and Giora Alexandron. 2022. Teachers’ trust in AI-powered educational technology and a professional devel- opment program to improve it.British journal of educational technology53, 4 (2022), 914–931

    work page 2022

  42. [42]

    OECD. 2024. . Technical Report. OECD Publishing, Paris. https://www.oecd. org/en/publications/results-from-talis-2024_90df6235-en/full-report

    work page 2024

  43. [43]

    2025.Teaching and Learning International Survey (TALIS) 2024 Conceptual Framework

    OECD. 2025.Teaching and Learning International Survey (TALIS) 2024 Conceptual Framework. Technical Report. OECD Publishing, Paris. doi:10.1787/7b8f85d4-en

    work page doi:10.1787/7b8f85d4-en 2025

  44. [44]

    OpenAI. 2024. Learning to Reason with LLMs. https://openai.com/index/learning- to-reason-with-llms/. Accessed: 2026-02-12

    work page 2024

  45. [45]

    Vinodkumar Prabhakaran, Rida Qadri, and Ben Hutchinson. 2022. Cultural Incongruencies in Artificial Intelligence. arXiv:2211.13069 [cs.CY] https://arxiv. org/abs/2211.13069

    work page arXiv 2022

  46. [46]

    Data and Discrimination: Converting Critical Concerns into Productive Inquiry,

    Christian Sandvig, Kevin Hamilton, Karrie Karahalios, and Cedric Langbort. 2014. Auditing algorithms: research methods for detecting discrimination on internet platforms. (May 2014). Paper presented at "Data and Discrimination: Converting Critical Concerns into Productive Inquiry, " preconference at the 64th Annual Meeting of the International Communicati...

    work page 2014

  47. [47]

    Shibani Santurkar, Esin Durmus, Faisal Ladhak, Cinoo Lee, Percy Liang, and Tatsunori Hashimoto. 2023. Whose opinions do language models reflect?. In Proceedings of the 40th International Conference on Machine Learning(Honolulu, Hawaii, USA)(ICML’23). JMLR.org, Article 1244, 34 pages

    work page 2023

  48. [48]

    Agrima Seth, Monojit Choudhury, Sunayana Sitaram, Kentaro Toyama, Aditya Vashistha, and Kalika Bali. 2025. How deep is representational bias in llms? the cases of caste and religion.Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society8, 3 (2025), 2319–2330

    work page 2025

  49. [49]

    Lijuan Shen, Nengsheng Qiu, and Zhanfeng Wang. 2025. Psychological safety and trust as drivers of teachers’ continued use of AI tools in classrooms.Scientific Reports15, 1 (2025), 31426

    work page 2025

  50. [50]

    Rana Taheri, Neda Nazemi, Sarah E Pennington, Jason A Clark, and Faraz Dad- gostari. 2025. Factors Influencing Educators’ AI Adoption: A Grounded Meta- Analysis Review.Computers and Education: Artificial Intelligence(2025), 100464. doi:10.1080/23812346.2025.2603822

    work page doi:10.1080/23812346.2025.2603822 2025

  51. [51]

    Xiao Tan, Gary Cheng, and Man Ho Ling. 2025. Artificial intelligence in teaching and teacher professional development: A systematic review.Computers and L@S ’26, June 29-July 03, 2026, Seoul, Republic of Korea Yan Tao, Olga Viberg, Deepak Varuvel Dennison, Zhikun Wu, René F. Kizilcec Education: Artificial Intelligence8 (2025), 100355

    work page 2025

  52. [52]

    Baker, and René F

    Yan Tao, Olga Viberg, Ryan S. Baker, and René F. Kizilcec. 2024. Cultural bias and cultural alignment of large language models.PNAS Nexus3, 9 (2024), pgae346. doi:10.1093/pnasnexus/pgae346

    work page doi:10.1093/pnasnexus/pgae346 2024

  53. [53]

    Tarang Tripathi, Smriti R Sharma, Vatsala Singh, Palaash Bhargava, and Chan- draditya Raj. 2025. Teaching and learning with AI: a qualitative study on K-12 teachers’ use and engagement with artificial intelligence. InFrontiers in Education, Vol. 10. Frontiers, 1651217

    work page 2025

  54. [54]

    Johanna Velander, M. A. Taiye, N. Otero, and Marcelo Milrad. 2024. Artificial intelligence in K-12 education: eliciting and reflecting on Swedish teachers’ understanding of AI and its implications for teaching & learning.Education and Information Technologies29 (2024), 4085–4105. doi:10.1007/s10639-023-11990-4

    work page doi:10.1007/s10639-023-11990-4 2024

  55. [55]

    Kizilcec

    Olga Viberg, Mutlu Cukurova, Yael Feldman-Maggor, Giora Alexandron, Shizuka Shirai, Susumu Kanemune, Barbara Wasson, Cathrine Tømte, Daniel Spikol, Marcelo Milrad, Raquel Coelho, and René F. Kizilcec. 2025. What Explains Teachers’ Trust of AI in Education across Six Countries?International Journal of Artificial Intelligence in Education35, 3 (2025), 1288–1316

    work page 2025

  56. [56]

    Angelina Wang, Jamie Morgenstern, and John P Dickerson. 2025. Large language models that replace human participants can harmfully misportray and flatten identity groups.Nature Machine Intelligence7 (2025), 400–411. doi:10.1038/s42256- 025-00986-z

    work page doi:10.1038/s42256- 2025

  57. [57]

    Rose Wang and Dorottya Demszky. 2023. Is ChatGPT a Good Teacher Coach? Measuring Zero-Shot Performance For Scoring and Providing Actionable Insights on Classroom Instruction. InProceedings of the 18th Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2023), Ekaterina Kochmar, Jill Burstein, Andrea Horbach, Ronja Laarmann-Quante,...

    work page doi:10.18653/v1/2023.bea- 2023

  58. [58]

    Jing Yao, Xiaoyuan Yi, Yifan Gong, Xiting Wang, and Xing Xie. 2024. Value FULCRA: Mapping Large Language Models to the Multidimensional Spectrum of Basic Human Value. InProceedings of the 2024 Conference of the North Ameri- can Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), Kevin Duh, Helena ...

    work page doi:10.18653/v1/2024.naacl-long.486 2024

  59. [59]

    Yaxuan Yin, Shamya Karumbaiah, and Shona Acquaye. 2025. Responsible AI in Education: Understanding Teachers’ Priorities and Contextual Challenges. In Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Trans- parency (FAccT ’25). Association for Computing Machinery, New York, NY, USA, 2705–2727. doi:10.1145/3715275.3732176

    work page doi:10.1145/3715275.3732176 2025

  60. [60]

    Marín, Melissa Bond, and Franziska Gouverneur

    Olaf Zawacki-Richter, Victoria I. Marín, Melissa Bond, and Franziska Gouverneur

    work page

  61. [61]

    Systematic review of research on artificial intelligence applications in higher education,

    Systematic review of research on artificial intelligence applications in higher education—where are the educators?International Journal of Educational Technology in Higher Education16, 1 (2019), 1–27. doi:10.1186/s41239-019-0171-0

    work page doi:10.1186/s41239-019-0171-0 2019

  62. [62]

    Self-discover: Large language models self-compose reasoning structures

    Pei Zhou, Jay Pujara, Xiang Ren, Xinyun Chen, Heng-Tze Cheng, Quoc V. Le, Ed H. Chi, Denny Zhou, Swaroop Mishra, and Huaixiu Steven Zheng. 2024. Self-Discover: Large Language Models Self-Compose Reasoning Structures. arXiv:2402.03620 [cs.AI] https://arxiv.org/abs/2402.03620

    work page arXiv 2024