CoChat
Sign Up
Cited by 238Artificial Intelligence in EducationEthics and Social Impacts of AIArtificial Intelligence in Healthcare and Education

Artificial Intelligence in Education: Implications for Policymakers, Researchers, and Practitioners

Open access
Ifenthaler, Dirk, Majumdar, Rwitajit, Gorissen, Pierre, Judge, Miriam, Mishra, Shitanshu, Raffaghelli, Juliana, Shimada, Atsushi·2024·Technology, Knowledge and Learning
DOI 10.1007/s10758-024-09747-0
Open this paper in CoChat — free accountOpen PDF

Abstract

One trending theme within research on learning and teaching is an emphasis on artificial intelligence (AI). While AI offers opportunities in the educational arena, blindly replacing human involvement is not the answer. Instead, current research suggests that the key lies in harnessing the strengths of both humans and AI to create a more effective and beneficial learning and teaching experience. Thus, the importance of ‘humans in the loop’ is becoming a central tenet of educational AI. As AI technology advances at breakneck speed, every area of society, including education, needs to engage with and explore the implications of this phenomenon. Therefore, this paper aims to assist in this process by examining the impact of AI on education from researchers’ and practitioners' perspectives. The authors conducted a Delphi study involving a survey administered to N = 33 international professionals followed by in-depth face-to-face discussions with a panel of international researchers to identify key trends and challenges for deploying AI in education. The results indicate that the three most important and impactful trends were (1) privacy and ethical use of AI; (2) the importance of trustworthy algorithms; and (3) equity and fairness. Unsurprisingly, these were also identified as the three key challenges. Based on these findings, the paper outlines policy recommendations for AI in education and suggests a research agenda for closing identified research gaps.

AI snapshot

Written from the paper’s full text.

TL;DR

This Delphi study with N = 33 international professionals and a face-to-face researcher/policymaker panel identifies privacy and ethical use of AI and big data, trustworthy algorithms, and fairness/equity as the top trends and challenges for AI in education. It outlines policy recommendations and a research agenda for policymakers, researchers, and practitioners.

Key findings

  • The three most important trends were privacy and ethical use of AI and big data in education (M = 8.7, SD = 1.286), trustworthy algorithms for supporting education (M = 8.3, SD = 1.608), and fairness & equity of AI in education (M = 8.2, SD = 1.674); the same three were top for impact and challenges.
  • Challenges overall were rated significantly higher than impact and importance: challenges M = 7.68, SD = 0.315; impact M = 7.05, SD = 0.593; importance M = 7.28, SD = 0.829; F(2, 57) = 3.512, p < 0.05, Eta² = 0.110 (medium effect).
  • Top challenges were privacy and ethical use of AI and big data in education (M = 8.8, SD = 1.455), trustworthy algorithms for supporting education (M = 8.3, SD = 1.804), and fairness & equity of AI in education (M = 8.3, SD = 1.855).
  • The panel also identified emerging realities: new roles of stakeholders in education, human-AI alliance in education, and precautionary pre-emptive policies preceding practice for AI in education; the paper recommends actions for policymakers, researchers, and practitioners.

Methods

  • Literature review plus a mixed Delphi design: a computer-based survey of N = 33 international professionals (26 from research institutions, 5 from industry, 2 from government; average age M = 46.64, SD = 9.83; average experience M = 17.8 years, SD = 9.4) rated 20 consolidated topics—reduced from an initial list of 70—on importance, impact, and key challenges using a 1–10 scale.
  • The second phase was a 3-day face-to-face workshop with international researchers and policymakers using a discussion protocol; discussion protocols were transcribed and analyzed with QCAmap using inductive and deductive coding, while quantitative data were analyzed in R at the 0.05 significance level.

Limitations

  • Stated in the paper: reaching consensus in the Delphi discussion may lead to a biased perspective; the sample size was limited; the convenience sample could have included more participants and further differentiated experience levels in AI in education; possible overlaps between identified constructs were monitored through face-to-face discussion.
  • Evident from the text: the sample is small and heavily weighted toward research institutions (n = 26 of 33), with uneven country representation, which limits generalizability.

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

Integrity signals

Registry

Crossref · 10.1007/s10758-024-09747-0

Cites

74 of 74 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

0 of 21 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    Privacy and ethical use of AI and big data in education was rated as the most impactful trend (M = 8.2; SD = 1.608).

    “Table 2 shows the most impactful trends, including (1) Privacy and ethical use of AI and big data in education (M = 8.2; SD = 1.608)”
    Quote in the paperNot a substantial claim
  • Findingown data

    Trustworthy algorithms for supporting education was rated as the second most important trend (M = 8.3; SD = 1.608).

    “As shown in Table 1, the most important trends included (1) Privacy and ethical use of AI and big data in education (M = 8.7; SD = 1.286), ( 2) Trustworthy algorithms for supporting education (M = 8.3; SD = 1.608), and Fairness & equity of AI in education (M = 8.2; SD = 1.674).”
    Quote in the paperNot a substantial claim
  • Findingown data

    Fairness & equity of AI in education was rated as the third most important trend (M = 8.2; SD = 1.674).

    “As shown in Table 1, the most important trends included (1) Privacy and ethical use of AI and big data in education (M = 8.7; SD = 1.286), ( 2) Trustworthy algorithms for supporting education (M = 8.3; SD = 1.608), and Fairness & equity of AI in education (M = 8.2; SD = 1.674).”
    Quote in the paperNot a substantial claim
  • Findingown data

    Privacy and ethical use of AI and big data in education was rated as the most important trend (M = 8.7; SD = 1.286).

    “As shown in Table 1, the most important trends included (1) Privacy and ethical use of AI and big data in education (M = 8.7; SD = 1.286)”
    Quote in the paperSubstantial claim
  • Findingown data

    Trustworthy algorithms for supporting education was rated as the second most impactful trend (M = 7.7; SD = 2.268).

    “Table 2 shows the most impactful trends, including (1) Privacy and ethical use of AI and big data in education (M = 8.2; SD = 1.608), ( 2) Trustworthy algorithms for supporting education (M = 7.7; SD = 2.268), and (3) Fairness & equity of AI in education (M = 7.7; SD = 1.736).”
    Quote in the paperNot a substantial claim
  • May combine two claims. 7
  • Classified as probably not a substantial claim. 14

  • May combine two claims.

    Findingown data

    Trustworthy algorithms for supporting education had mean importance 8.7, impact 8.8, and challenge 8.2 (N = 33).

    “| Trustworthy algorithms for supporting education 8.7 | 0.823 8.8 | 1.135 8.2 | 1.476 | | |”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Fairness & equity of AI in education had mean importance 8.3, impact 7.7, and challenge 8.2 (N = 33).

    “| Fairness & equity of Al in education | 8.3 | 0.949 7.7 | 1.829 8.2 | 1.476 | | |”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Privacy and ethical use of AI and big data in education had mean importance 8.4, impact 7.8, and challenge 8.5 (N = 33).

    “| Privacy and ethical use of Al and big data in | 8.4 | 1.35 7.8 | 2.201 8.5 | 1.354 | | |”
    Quote in the paperSubstantial claim

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

    Findingown data

    The panel of international researchers and policymakers identified precautionary pre-emptive policies preceding practice for AI in education as an emerging educational reality with AI.

    “Further, the panel of international researchers and policymakers identified emerging educational realities with AI, including (d) new roles of stakeholders in education, (e) human-AI-alliance in education, and (f) precautionary pre-emptive policies preceding practice for AI in education.”
    Quote in the paperNot a substantial claim

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

    Study design and conductnone given

    All procedures performed in studies involving human participants were under the ethical standards of the institutional and national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.

    “All procedures performed in studies involving human participants were under the ethical standards of the institutional and national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.”
    Quote in the paperNot a substantial claim

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

    Study design and conductnone given

    All effects were tested at the 0.05 significance level, and effect size measures were computed where relevant.

    “All effects were tested at the 0.05 significance level, and effect size measures were computed where relevant.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Bias can occur at all stages of the machine learning life cycle and encompasses historical, representation, measurement, aggregation, evaluation and deployment bias.

    “Bias, which can occur at all stages of the machine learning life cycle, is a multilayered phenomenon encompassing historical bias, representation bias, measurement bias, aggregation bias, evaluation bias and deployment bias (Suresh & Guttag, 2021).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Suresh & Guttag, 2021

      Not checked in the free run.

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

    Study design and conductcitation

    The face-to-face discussion followed a discussion protocol made available before the meeting.

    “During the face-to-face meeting, the panel of international researchers and policymakers followed a discussion protocol made available before the meeting (Beiderbeck et al., 2021).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Beiderbeck et al., 2021

      Not checked in the free run.

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

    Study design and conductcitation

    Reaching a consensus in the discussion may lead to a biased perspective on the research topic.

    “However, reaching a consensus in the discussion may also lead to a biased perspective on the research topic (Beiderbeck et al., 2021).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Beiderbeck et al., 2021

      Not checked in the free run.

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

    Study design and conductnone given

    Possible overlaps between the identified constructs are a limitation.

    “Further, a limitation may be seen in possible overlaps between the identified constructs during the Delphi study.”
    Quote in the paperNot a substantial claim

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

    Study design and conductnone given

    Additional informed consent was obtained from all individual participants for whom identifying information is included in this article.

    “Additional informed consent was obtained from all individual participants for whom identifying information is included in this article.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    AI tools augment the role of educators by automating administrative tasks, enabling them to allocate more time to mentoring, fostering creativity, and critical thinking.

    “Moreover, AI tools augment the role of educators by automating administrative tasks, enabling them to allocate more time to mentoring, fostering creativity, and critical thinking (Ames et al., 2021).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Ames et al., 2021

      Not checked in the free run.

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

    Study design and conductnone given

    The constructs were constantly monitored concerning their content validity and refined through in-depth face-to-face discussion.

    “However, through the indepth face-to-face discussion of the panel of international researchers, the constructs were constantly monitored concerning their content validity and refined accordingly.”
    Quote in the paperNot a substantial claim

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

    Study design and conductcitation

    Both inductive and deductive coding techniques were used.

    “Both inductive and deductive coding techniques were used (Mayring, 2015).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Mayring, 2015

      Not checked in the free run.

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

    Premisecitation

    The reciprocal collaboration of humans and AI might be a suitable approach for enhancing the capacities of both.

    “Due to the remaining challenges of implementing meaningful AI in educational contexts, especially for more sophisticated tasks, the reciprocal collaboration of humans and AI might be a suitable approach for enhancing the capacities of both (Baker, 2016).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Baker, 2016 source

      Not checked in the free run.

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

    Study design and conductnone given

    Regular researcher debriefing was conducted during data analysis to enhance the reliability and validity of the quantitative and qualitative analysis.

    “Regular researcher debriefing was conducted during data analysis to enhance the reliability and validity of the quantitative and qualitative analysis.”
    Quote in the paperNot a substantial claim

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

    Study design and conductnone given

    Informed consent was obtained from all individual participants included in the study.

    “Informed consent was obtained from all individual participants included in the study.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    The proliferation of AI in education raises pertinent ethical concerns, including data privacy, algorithmic biases, and the digital divide.

    “However, the proliferation of AI in education also raises pertinent ethical concerns, including data privacy, algorithmic biases, and the digital divide (Baker & Hawn, 2021;Ifenthaler, 2023).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Baker & Hawn, 2021 source

      Not checked in the free run.

    • cites Ifenthaler, 2023

      Not checked in the free run.

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

    Study design and conductnone given

    All data were saved and analysed using an anonymized process as per conventional research data protection procedures.

    “All data were saved and analysed using an anonymized process as per conventional research data protection procedures.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    This customisability of education fosters a more inclusive and effective learning environment, accommodating diverse learning needs and regulations.

    “This customisability of education fosters a more inclusive and effective learning environment, accommodating diverse learning needs and regulations.”
    Quote in the paperNot a substantial claim
20 of 20

  • The paper relies on this without support.

    AI will be omnipresent in education.

    “New roles of stakeholders in education is related to the phenomena that AI will be omnipresent in education, which inevitably involves stakeholders interacting with AI systems in an educational context.”

  • The paper relies on this without support.

    AI has far more potential to transform the education arena than any technology before it.

    “Despite its hesitant implementation, AI has far more potential to transform the education arena than any technology before it.”

  • The paper relies on this without support.

    AI in education has shifted from being narrowly focused on automation-based tasks to augmenting human capabilities linked to learning and teaching.

    “Human-AI-alliance in education emphasises that AI in education shifted from being narrowly focused on automation-based tasks to augmenting human capabilities linked to learning and teaching.”

  • The paper relies on this without support.

    Precautionary pre-emptive policies precede practice for AI in education, and decision-makers tend to introduce restrictive policies in reaction to initial societal concerns with emerging AI developments.

    “Precautionary pre-emptive policies precede practice for AI in education, underlining that, overwhelmed by the rapid change in the technology landscape, decision-makers tend to introduce restrictive policies in reaction to initial societal concerns with emerging AI developments.”
4 of 4

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Statistics recomputed

Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.

  • ConsistentF(2, 57) = 3.512reported p < .05 · recomputed p = .036

    “F(2, 57) = 3.512, p < .05”

1 of 1

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.

  • Privacy trend has conflicting ratings in Table 4

    Major

    The results report importance, impact, and challenge means of 8.7, 8.2, and 8.8 for the privacy trend. Table 4 gives the same trend means of 8.4, 7.8, and 8.5, respectively, despite identifying both as ratings from N = 33.

  • Trustworthy-algorithms ratings conflict with Table 4

    Major

    The results give trustworthy algorithms an importance mean of 8.3 and an impact mean of 7.7. Table 4 instead gives means of 8.7 and 8.8; it also makes this trend, rather than privacy, the highest-rated of the three on both measures.

2 of 2

Cites

Flagged first, then by citations

Research works printed but not linked: 5

  1. Buckingham Shum, S., & McKay, T. A. (2018). Architecting for learning analytics. Innovating for sustainable impact. EDUCAUSE Review, 53(2), 25–37. https://er.educause.edu/articles/2018/3/architecting-for-learning-analytics-innovating-for-sustainable-impact
  2. Daugherty, P. R., & Wilson, H. J. (2018). Human + machine: Reimagining work in the age of AI. Harvard Business Review Press.
  3. Jimerson, J. B., & Childs, J. (2017). Signal and symbol: How state and local policies address data-informed practice. Educational Policy, 31(5), 584–614. https://doi.org/10.1177/089590481561344 · DOI
  4. Jones, M., & Winne, P. H. (Eds.). (1992). Adaptive learning environments. Springer.
  5. Zeide, E. (2019). Artificial intelligence in higher education: Applications, promise and perils, and ethical questions. EDUCAUSE Review, 54(3), 21–39.

Cited by

Sorted by citations

Related

Authors
Ifenthaler, D., Majumdar, R., Gorissen, P., Judge, M., Mishra, S., Raffaghelli, J., & Shimada, A.
Affiliations
  • University of Mannheim (DE)
  • Kumamoto University (JP)
  • HAN University of Applied Sciences (NL)
  • Dublin City University (IE)
  • UNESCO (FR)
  • University of Padua (IT)
  • Kyushu University (JP)
Corresponding author
Ifenthaler, Dirk
ORCID
Venue
Technology, Knowledge and Learning
Publisher
Springer Science and Business Media LLC
Published
2024-06-04
Volume
29
Issue
4
Pages
1693-1710
ISSN
  • 2211-1662 (ISSN-L)
  • 2211-1670
DOI
10.1007/s10758-024-09747-0
OpenAlex
W4399332765
Cited by
238
References
74
Citations by year
2026: 124 · 2025: 93 · 2024: 21
FWCI
89.14
Type
journal article
Language
en
Open access
Licence
Locations
  • Technology Knowledge and Learning (publishedVersion, cc-by)
  • MADOC (University of Mannheim) (acceptedVersion, cc-by)
  • Padua Research Archive (University of Padova) (submittedVersion, cc-by)
Funders
Universität Mannheim
Topics
  • Physical Sciences › Computer Science › Information Systems › Artificial Intelligence in Education
  • Social Sciences › Safety Research › Ethics and Social Impacts of AI
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
Keywords
artificial intelligence in education, human-in-the-loop, AI ethics, privacy, equity and fairness, AI policy, research agenda, Delphi method, human-AI collaboration

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 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

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