Guide to the Responsible Use of Artificial Intelligence in Health Research

Before using an AI tool, it is essential to understand the fundamental frameworks governing research, particularly in healthcare. These are not mere formalities, but basic prerequisites for building trust and ensuring the quality and integrity of research.

Privacy (GDPR)

The General Data Protection Regulation (GDPR) forms the foundation for all processing of personal data. Such processing requires a legal basis. Key principles, including data minimization (do not collect more data than necessary), purpose limitation (data may only be used for the purpose for which consent was obtained) and storage limitation, must be followed. Take particular care when uploading sensitive data to external AI services.

Consent

Participants’ consent must be informed, voluntary, specific and unambiguous. Consider whether existing consent covers the use of AI analysis. If not, it may be necessary to obtain new consent. Transparency with participants about how their data is processed is essential.

Bias

The quality of an AI model depends on the quality of its training data. Historical or systematic biases in the dataset may be perpetuated and amplified by the model. Critically assess both your own data and the AI model: could biases in the dataset, such as the underrepresentation of certain groups, lead to discriminatory or inaccurate conclusions? Results must always be interpreted in a broader professional context.

Responsibility and Transparency

As a researcher, you always retain ultimate responsibility for your research, regardless of which tools are used. It is essential to document which AI tools are used and how they are used. Be prepared to explain your methods. Although many AI models function as “black boxes,” you must be able to explain the processes surrounding the model, such as data collection, preprocessing and validation of results.

AI can be a valuable resource at many stages of the research process. Below are some practical applications and key considerations.

Using AI for Text and Data

AI tools can help summarize literature, analyze large volumes of text, transcribe interviews or improve the language of scientific articles. They can also identify patterns in large datasets that would be difficult to detect manually.

• Use AI as an assistant for brainstorming, structuring and copyediting, but always write the final text yourself. You are the author and are fully responsible for its content.
• Avoid sharing or uploading sensitive data. Do not paste unpublished research data, personal information or other confidential information into publicly available AI services.
• Verify all information. AI models can “hallucinate,” meaning they can generate false facts, citations and references. Double-check all information against primary sources.
• Be transparent about your methods. State in the methods section which AI tools were used and for what purposes.

Using AI for Images and Illustrations

AI can generate illustrations for presentations, posters or publications, which can be useful for visualizing complex concepts. However, there are important pitfalls to be aware of.

• Clarify copyright. Regulations regarding ownership of AI-generated images are still evolving. Always check the terms of use for the service being used. Many scientific journals have developed their own guidelines on this.
• Avoid misleading content. Do not generate realistic images of “patients” or simulated clinical situations, as this can be ethically problematic and misleading. Use AI primarily to create conceptual figures and diagrams.
• Clearly label all generated content. Always state that an illustration was generated using AI, both in the caption and in the methods section, to ensure scientific transparency.

Choosing tools and taking a critical approach to the results are essential. The quality of AI solutions varies, and free services often come with a hidden cost in the form of using your data.

Choosing Tools

• Assess data security. Take particular care with free tools, as the data you upload may be used to train their models. Always read the privacy policy carefully.
• Prioritize institutional solutions. Check whether your institution offers secure AI tools or has framework agreements in place. Such solutions are often assessed for privacy and security.
• Choose tools suited to the purpose. A general-purpose language model is not necessarily the best choice for a specific bioinformatics analysis.

Evaluating AI-Generated Content

Treat AI-generated results in the same way as information from any other source: with sound professional skepticism. Follow a three-step process:

Generate

Obtain a draft, summary or analysis from the AI tool.

Verify

Check whether the facts are correct, the figures are accurate, and the cited references exist and are relevant.

Contextualize

Place the information in your professional context. Are the conclusions reasonable? Are important nuances missing?

Use this checklist as a summary and reminder before using AI in your research project.

• Have privacy (GDPR) and data security been adequately assessed?
• Does the participants’ consent cover the planned use of AI analysis?
• Have I considered potential bias in the data and model?
• Is the tool I intend to use safe and in line with the institution’s guidelines?
• Do I have a plan to verify all factual information and references from the AI tool?
• Is all AI-generated content (text, images) clearly labeled as such?
• Is AI use transparently documented in the publication’s methods section?

• Artificial Intelligence - General Instructions(internal document)
• Artificial intelligence in the health sector – a guide | Journal of the Norwegian Medical Association