AI evidence extraction

The role of generative AI in academic and scientific authorship: an autopoietic perspective

Steven Watson, Erik Brezovec, Jonathan Romic · AI & Society · 2025

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

9/10

Relevance

1/4

Quality (LMQS)

Evidence

Citations

50.00

FWCI

Top 10%

Impact

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.1007/s00146-024-02174-w

Methodology & findings

Study design

Theoretical analysis using social systems theory (Luhmann's autopoiesis framework).

Main result

The study found that "the relationship between generative AI and human authorship is best understood as a collaborative and integrative process" where "generative AI assists in the production and refinement of text, while human authors provide essential context, critical analysis, and domain-specific knowledge necessary for meaningful scientific communication inside the scientific social system." The authors argue that scientific authorship should be understood as an autopoietic system involving structural coupling between author, scientific system, and technology.

Reports effect sizes.

Research paradigm

Interpretivist/Constructivist (social systems theory, autopoiesis, non-essentialist ontology)

Author conclusions

The authors conclude that "the integration of generative AI, particularly LLMs like ChatGPT, into scientific authorship is a complex phenomenon that requires an understanding beyond traditional humanistically oriented perspectives" and that "scientific authorship is a relational phenomenon which is always lesser then its parts (each and every system involved in the creation) but cannot be reduced to either one of it." They propose that "policy which is directed to the generative AI questions in the manner of authorship should include questioning not only technology or the author but the publishing system (commodification of science) and scientific systems inclusive normatives."

Risk of bias

Selection bias in cited examples: The paper relies on selected case studies (e.g., Guo et al. 2024 retraction, Stokel-Walker 2024 estimates) without systematic sampling; Confirmation bias: The theoretical framework (autopoiesis) is applied selectively to support the argument rather than tested against alternative explanations; No empirical validation: The theoretical claims are not tested against data from actual authorship practices; Limited scope of examples: Cited statistics (e.g., 1% of 2023 articles, 17.5% of computer science papers) are presented without methodology details or quality assessment

Limitations

The paper does not conduct empirical measurement or controlled experiments
It is a theoretical essay that "does not conduct systematic empirical research" on the phenomena
The authors acknowledge that "much of the debate and the ensuing challenges for institutions, publishers, and individual scientists have been overwhelmed by media hyperbole and preoccupations with the idea that LLMs can and will simply 'generate' scientific output-mainly derived from the ignorance about the ways in which these LLMs function." The analysis is primarily conceptual rather than empirical validation of the proposed framework.

Open questions raised

The authors identify that most scientific research has focused on meta-analysis or pros/cons debates rather than structured scientific research or theoretical discussions. They call for a more robust and nuanced approach to considering and regulating the use of generative AI in scientific authorship. They also identify the need for structural approaches addressing the 'publish or perish' culture, power relationships in academia, and commodification of science rather than focusing solely on individual moral behavior or technological regulation.
The authors identify the need for: (1) structured consensus-based guideline development for AI use in academic publishing; (2) theoretical frameworks beyond humanistic approaches to understanding AI in scientific authorship; (3) structural approaches addressing publishing commodification, 'publish or perish' culture, and power relations in academia rather than solely individual or technological regulation; (4) improved detection methods for AI-generated content that avoid false positives/negatives and potential 'academic McCarthyism'; (5) policies that question the publishing system and scientific norms of promotion alongside technology regulation.
The authors identify the need for: (1) theoretical and structural approaches to understanding AI-society relationships beyond moral communication; (2) policy frameworks addressing not only technology but also the publishing system's commodification and scientific normatives; (3) examination of power relationships and 'publish or perish' culture as integral to the problem; (4) guidelines, regulations, and policies for AI visualization and text generation that augment accuracy and integrity in scientific communications.

Extracted from: pdfAgreement 82%

Explore related topics

AI Manuscript Writing Academic Writing Conceptual