On overview of important considerations when using generative artificial intelligence in research data management.

What is Generative Artificial Intelligence?

Generative artificial intelligence (GenAI) enables users to generate new materials and resources using descriptive prompts. GenAI tools are trained and learn language patterns from large collections of text, images, code, and other data, which may include harvested content from public websites and repositories. Training data can contain errors and biases that can impact the quality of the output.

GenAI and research data

Common applications of GenAI in research typically entail text synthesis and generation, data processing, and production of research software. This is often seen as: making automated summaries of primary literature, multi-repository searches for specific types of data, as well as assistance with software development and debugging.

Important considerations when using GenAI in research data management

GenAI is evolving rapidly, with new tools introduced almost daily. GenAI should not be used without thoughtful consideration of its ethical, legal, societal, and environmental impacts. Additionally, maintaining transparency on the use of GenAI and its potential pitfalls is key to ensuring this valuable tool does not impact the integrity of your research. Researchers must remain compliant with data privacy, security, and relevant legal requirements when using GenAI.

Data processing

Before you use GenAI tools to assist with processing, you should consider potential conflicts regarding data ownership. Who owns the research data you are providing to the GenAI service? And does providing this data mean ownership is transferred to the GenAI company? In most cases, researchers do not hold the rights to their research data and thus cannot agree to transfer ownership to another party. Carefully review the terms of use of any GenAI company, tool or service, as well as the terms of use of the data, and when in doubt, consult your legal officer (Internal knowledge - Contracts & Authorizations).

Do not enter confidential, sensitive, or personal data into GenAI platforms that do not meet WUR’s technical, legal and contractual requirements. This includes data received from partners or third parties, unpublished research data, personal or patient data, and any other information that is not freely and publicly available.

The use of GenAI in data processing should be included in your Data Management Plan, Software Management Plan and/or Data Sharing Agreement. For any questions, see WUR’s Research Data Management portal and contact your group’s Data Steward or your Legal Department.

Sensitive data

For sensitive data, you need to adhere to stricter sharing guidelines. Providing it to a GenAI service is data sharing and may not be lawful. The terms of use for these services often require users to agree that any data they provide can be used for further training, and in some cases, other applications. Any GenAI applications that are online and not locally hosted run the risk of a data breach and are not suitable for sensitive data. Any projects processing personal data must adhere to the GDPR and thus should be cautious about sharing data with a GenAI service. Always consult your privacy and/or information security officer (Internal knowledge - Privacy & Information Security Officer of your administration) before sharing sensitive data.

For questions about personal data, contact your science group’s Privacy Officer; for issues related to data security, contact your science group’s Information Security Officer; and for guidance on responsible data sharing and ethical considerations, consult Data Sharing at WUR.

Unpublished data

Similar to sensitive data, uploading unpublished research data to a GenAI service is not allowed since it is a form of distribution. GenAI services do not securely store files provided by users and cannot guarantee the safeguarding of unpublished data. Do not provide your unpublished research data to a GenAI tool unless you are prepared for the data to be distributed and used to train later versions of the model. To avoid this risk, consider first publishing your data and defining the terms of use through a Creative Commons licence (External knowledge - Choosing a licence for your research data).

Data scraping

GenAI tools can efficiently scrape data from multiple sources via a single prompt. Though this saves time, most GenAI services do not train their models to check the terms of use of the data they scrape and provide to users. Uncritical use of scraped datasets risks legal repercussions if the data is protected by copyright or if reuse is restricted. To avoid this, carefully check your data sources and ensure that the data is appropriately licensed for reuse. Do not rely on the GenAI service to ensure that the reuse of scraped data is legal.

Data mining

GenAI services can be used to automate data mining and synthesis. However, data mining results can be impacted by training errors and biases in GenAI models. The quality of the output of a GenAI service depends on both the quality and specificity of the prompt, as well as the quality and relevancy of the data the model was trained on. Be highly critical of outputs produced via a data mining prompt and carefully consider if you could replicate the findings manually.

Documentation of GenAI use

If you are using a GenAI service to assist with your research, it is important to transparently document this. Many journals require researchers to disclose this when submitting a research output, and some funders may require researchers to declare this in their data management plan. Export your chat logs when using GenAI to assist with research and preserve these as a form of research data documentation. When using GenAI to assist with the production of research software, use version control to keep track of edits made to generated scripts. Like laboratory equipment, treat GenAI services as a research instrument, and provide detailed reports of the platform, version, and any other program specifications required to reproduce the conditions under which you provided a prompt.

Support

For questions about personal data, contact your science group’s Privacy Officer; for issues related to data security, contact your science group’s Information Security Officer; and for guidance on responsible data sharing and ethical considerations, consult Data Sharing at WUR.