AI evidence extraction

Core principles of responsible generative AI usage in research

Tim-Dorian Knöchel, Konrad J. Schweizer, Oguz A. Acar, Atakan M. Akıl, Ali H. Al‐Hoorie, Florian Buehler et al. · AI and Ethics · 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

2.62

FWCI

Top 10%

Impact

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.1007/s43681-025-00768-8

Methodology & findings

Study design

Delphi consensus procedure comprising a panel of 16 international and multidisciplinary experts in AI, social sciences, law, ethics, and scientific publishing. The procedure was preregistered on OSF.

Main result

The study identified eight core principles for responsible generative AI usage in research through expert consensus: "Consensus was reached for eight principles (Fig. 1). All principles are organised in a sequential order, starting with the most general requirements, which should be addressed first, followed by steps that are relevant only if the previous principles are satisfied." These principles address Regulations, Data Security, Quality Control, Originality, Bias Mitigation, Accountability, Transparency, and Broader Impact.

Research paradigm

Normative/prescriptive ethics; consensus-driven framework development

Author conclusions

The authors conclude that "Awareness of these eight principles contributes to responsible GenAI use on both a general and concrete level. They serve as an initial take to achieve a stable guide in an everchanging AI landscape and inform the formulation of further guidelines concerning ongoing AI developments in research."

Risk of bias

Limited to 16 experts - potential selection bias in panel composition; Potential disciplinary bias given multidisciplinary composition may not equally represent all perspectives; Geographic/institutional representation of experts not fully specified

Limitations

The authors note that "Due to response stochasticity and iterative involvement, a complete documentation of GenAI usage may be cumbersome, and certain use cases (e.g., copy-editing) might not require detailed reporting
Field or topic-specific guidelines may be needed to ensure consistency." Additionally, the framework is described as "an initial take to achieve a stable guide in an everchanging AI landscape."

Open questions raised

Authors identify the need for field or topic-specific guidelines to ensure consistency in GenAI reporting. They note that as GenAI will continue evolving, these principles serve as an initial framework requiring ongoing refinement and supplementation with discipline-specific guidance.
The authors identify the need for ongoing development: "They serve as an initial take to achieve a stable guide in an everchanging AI landscape and inform the formulation of further guidelines concerning ongoing AI developments in research." Additionally, they note that "Field or topic-specific guidelines may be needed to ensure consistency."
The authors identify that "Field or topic-specific guidelines may be needed to ensure consistency" and that further guidelines are needed concerning "ongoing AI developments in research." They also note that "The framework distinguishes itself from higher-level ethics codes by translating broad scientific and societal values into concrete action recommendations for scientific use and complements more narrow publishing and discipline-specific guidelines by defining overarching principles through an expert committee."

Code: GitHub repository for Shiny app checklist: https://github.com/marton-balazs-kovacs/CorePrincipleGenAIChecklist; Zenodo archive mentioned for checklist preservation; https://github.com/marton-balazs-kovacs/CorePrincipleGenAIChecklist; Zenodo (archived); https://github.com/marton-balazs-kovacs/CorePrincipleGenAIChecklist (Shiny app for checklist); Archived with ZenodoExtracted from: pdfAgreement 82%

Explore related topics

Responsible AI Use Research Integrity Position Paper