AI evidence extraction

The AI turn in science: a reflexive thematic analysis of scientists’ commentaries and educational implications

Ho Yin Chan, Sibel Erduran · Research in Science & Technological Education · 2026

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

9/10

Relevance

3/4

Quality (LMQS)

Evidence

Citations

0.00

FWCI

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.1080/02635143.2026.2669549

Methodology & findings

Study design

Reflexive thematic analysis (RTA) of review commentaries published in Nature and Science journals between 2021 and 2024.

Sample

N = 281, 7 groups

Main result

The study found that "AI is woven into scientific practice in ways that cut across epistemic aims, material infrastructures, professional roles, institutional arrangements, governance practices, and public legitimacy." Across the corpus of 281 commentaries, "AI is rarely discussed as a concern confined to a single FRA-NOS aspect. Instead, it is repeatedly framed through patterns that simultaneously invoke questions of knowledge, authority, infrastructure, labour, coordination, and trust." The analysis identified five cross-cutting themes: AI as reconfiguration of epistemic authority and judgement, AI as infrastructure with materiality and dependency, AI and reconfiguration of scientific labour and professional identity, AI as catalyst for collective governance, and AI as pressure point for legitimacy and public trust.

Reports effect sizes.

Research paradigm

Interpretive/constructivist (qualitative hermeneutic analysis using reflexive thematic analysis within Wittgensteinian philosophical framework)

Author conclusions

"An 'AI turn' has occurred in the sciences and the mediation of the scientific enterprise through AI leads to new epistemological questions for science education. Our approach was based on a Wittgensteinian analysis of scientists' narratives in the top science outlets of Nature and Science, illustrating a range of issues including epistemic authority, scientific labour and public trust. Science curricula need to adapt to the emerging AI-driven scientific landscape so as to ensure that students' understanding of science is consistent with such emerging accounts of science." The authors argue that "responding to the AI turn in science education therefore requires not only curricular innovation, but methodological reflexivity in how the nature of science itself is studied and represented."

Risk of bias

Publication bias: Data source limited to two high-impact journals (Nature and Science), reflecting Western, industrialized scientific perspectives; Selection bias: Commentaries represent invited contributions from established researchers, not diverse scientist voices; Researcher positionality bias: Analysis shaped by authors' positions as European-based cultural geographer and science educator with prior FRA-NOS engagement; Gatekeeping bias: Publishing norms and institutional priorities shape which accounts of scientific change become visible and authoritative; Selection bias: corpus limited to two high-impact journals (Nature and Science), which predominantly represent Western, industrialised perspectives; Representation bias: WEIRD (Western, educated, industrialised, rich, democratic) perspectives overrepresented; Publication bias: commentary genre reflects editorial preferences and institutional priorities rather than comprehensive scientific discourse; Researcher positionality bias: authors are European-based researchers (cultural geographer and science educator) with prior engagement with FRA-NOS framework; Publication bias - only high-impact journals (Nature, Science) included, excluding lower-tier and non-English-language publications; Geographical bias - predominantly Western, industrialised contexts (WEIRD perspectives); Selection bias - commentaries are invited contributions from established researchers, not representative sampling of all scientific discourse; Researcher positionality bias - authors are European-based with prior engagement with FRA-NOS framework, which may shape interpretive lens; Temporal bias - analysis period (2021-2024) may not capture longer-term trajectories in AI-science integration

Limitations

The corpus reflects particular conditions of knowledge production: "Commentaries published in Nature and Science are situated within Western, industrialised, and well-resourced contexts, and predominantly represent WEIRD perspectives." The authors note that "the analysis is based on commentaries from two high-impact journals and therefore does not capture the full diversity of global scientific discourse." Additionally, "the interpretive nature of RTA means that findings reflect theoretically informed readings of how AI is articulated, rather than exhaustive or generalisable claims about scientific practice." The analysis does not evaluate the technical performance of specific AI systems, and findings are situated interpretations intended as analytically productive resources rather than comprehensive representations of AI in science.

Open questions raised

The paper identifies that "relatively little research has examined how scientists themselves articulate AI in relation to the epistemic, institutional, and value-laden dimensions of their work." It notes that "Within science education research, the emphasis in applications of AI and generative AI (GenAI) has been rooted in pedagogical concerns" but "The articulation of how AI is mediating scientific practice itself remains relatively limited."
The authors identify a gap in how AI is articulated in relation to the epistemic, institutional, and value-laden dimensions of scientific work: "Despite growing attention to AI's technical capabilities and ethical risks, relatively little research has examined how scientists themselves articulate AI in relation to the epistemic, institutional, and value-laden dimensions of their work." They also note that "the articulation of how AI is mediating scientific practice itself remains relatively limited." Future research should address how science education can cultivate "judgement, responsibility, and the capacity to go on appropriately within AI-inflected scientific practices."
The study identifies that relatively little research has examined how scientists themselves articulate AI in relation to epistemic, institutional, and value-laden dimensions of their work. Within science education research, emphasis has been on pedagogical applications of GenAI and AI, but the articulation of how AI is mediating scientific practice itself remains relatively limited. The authors call for methodological reflexivity in how the nature of science itself is studied and represented, and for educational responses that go beyond introducing isolated technical skills or stand-alone ethics modules.

Extracted from: pdfAgreement 70%

Explore related topics

Responsible AI Use Multi-Domain Qualitative