AI for scientific integrity: detecting ethical breaches, errors, and misconduct in manuscripts

Area of Science:

• Scholarly Publishing
• Artificial Intelligence
• Research Integrity

Background:

• Generative AI (GenAI) is increasingly used in scientific writing for drafting and analysis.
• Risks include undisclosed AI authorship, content manipulation, and AI-driven papermills.
• Maintaining research integrity in the GenAI era is a growing concern.

Purpose of the Study:

• To examine strategies for detecting unethical GenAI use in manuscripts.
• To investigate AI tools for identifying errors in scientific literature.
• To assess the current state of AI detection and error identification technologies.

Main Methods:

• Reviewed capabilities and limitations of AI detectors for human-written vs. machine-generated text.
• Investigated AI-powered systems for detecting statistical errors, image manipulation, and citation inaccuracies.
• Discussed publishing industry initiatives against AI-driven papermills.

Main Results:

• Existing AI detectors show performance gaps, genre sensitivity, and vulnerability to adversarial attacks.
• AI tools for error detection in research are emerging but not yet sufficiently accurate or reliable.
• Current AI developments represent early but promising steps toward AI-assisted quality control.

Conclusions:

• Current AI detection and error identification tools are not ready for academic assessment.
• Further development is needed to ensure accuracy and reliability for scholarly publishing.
• AI-assisted quality control in scientific publishing shows future potential.