AI in Research: An Ethical Decision Framework
To maintain the trust that underpins scientific progress, researchers using AI must adhere to a framework of core responsibilities. This guide synthesizes the requirements set forth by major international health and publishing ethics organizations into a set of actionable principles for preparing a manuscript.
The central pillar of all guidance is that authorship confers both credit and responsibility. Because an AI cannot be held accountable, it cannot be an author. According to the ICMJE, authorship is based not only on substantial contribution but also on being accountable for the work. Researchers must therefore:
- Assume Full Responsibility: Scientists must take complete responsibility for any errors, omissions, or ethical breaches in content produced by AI/ML systems. This includes the ability to explain why errors occurred and to be transparent about the limitations of their knowledge.
- Uphold Authorship Criteria: Acknowledge that AI tools do not meet the criteria for authorship. They cannot approve final manuscripts or agree to be accountable for the research, which are fundamental requirements for being listed as an author.
Transparency & Disclosure: Building Trust
Undisclosed use of AI in submissions is a rising concern for journal editors. To ensure the integrity of the scientific record, transparency is not optional. The ICMJE has issued an influential statement, adopted by thousands of journals, that mandates clear disclosure.
- Disclose All Use: At the time of submission, authors are required to disclose whether they used AI-assisted technologies like large language models or chatbots in the production of their work.
- Specify in Detail: This disclosure must be made in both the cover letter and the body of the submitted manuscript. It should describe precisely how the technology was used and which tools were employed.
Data Integrity: Upholding Scientific Standards
The use of AI introduces new challenges to maintaining the integrity of data and the originality of the text. Researchers have a duty to safeguard the reliability of their work.
- Prevent Plagiarism: Authors must be able to assert that their paper contains no plagiarism, including in text and images produced by AI. This requires ensuring appropriate attribution and full citations for all quoted material, regardless of how it was generated.
- Label Synthetic Data: The WHO and other bodies emphasize that when AI is used to generate synthetic data, researchers must clearly indicate which parts of the data are synthetic. They should also describe how the data were created and explain the rationale for their use to avoid blurring the line between real and fabricated information.
