Librarian's guide to Artificial Intelligence

The University of Turku has published guideline on using artificial intelligence in research on 4 January 2024. The guideline applies to applications or services utilizing generative artificial intelligence.

The University of Turku encourages researchers to use AI applications both in research and in the related processes. However, in using AI applications, the following perspectives have to be taken into consideration:

• Follow the guideline on the responsible conduct of research by the Finnish Advisory Board on Research Integrity and the requirements of processing personal data stated in the EU General Data Protection Regulation.
• Describe the use of artificial intelligence in research, including technical details, openly.
• Check the copyright terms of the application before utilizing material produced with AI, for example, publicly or with partners. Ensure that the terms allow the use of the materials as planned and do not conflict with the research agreement.
• Do not enter research data that contains confidential or classified data or personal information into online AI applications. Personal data and research materials containing such information may only be entered into an AI application for which the university has made a data processing agreement with the application provider. In such cases, inform the research subjects about the processing of personal data using artificial intelligence.
• Do not upload research plans, research materials, or unpublished studies to online AI applications without careful consideration. Be aware of the risks that uploaded material may become training data for AI applications or be transferred to third parties.
• Artificial intelligence cannot be considered the author of a scientific output. As the user of artificial intelligence, you are responsible for the content and accuracy of the scientific output.
• The ownership of the research’s background material and research plans is retained either by the original owner or in accordance with the terms of the research agreement regardless of the use of an AI application.
• Take into account guidelines on confidentiality when using AI.
• Check the funder's terms for using AI applications. Ensure that the copyright terms of the AI application do not conflict with the funding conditions. If necessary, separately negotiate the use of AI with the funder.
• If you use an AI application to generate code, consider the licenses for utilizing the code. Publish your own software as open-source, if possible.
• Take into account the guidelines of publishers and journals regarding the use of artificial intelligence and the disclosure of its use in publications.

Artificial Intelligence applications can be employed throughout various stages of the research process, such as data collection, coding, literature reviews, mapping relevant materials, categorizing data, textual analysis, data analytics, summarization, and brainstorming. However, AI doesn't replace critical thinking, expertise, or ethical decision-making.

Consider Data Privacy and Security:

• Avoid processing personal, confidential, sensitive (e.g., personal or health information), or inappropriate data in AI applications. All shared data with these applications is retained and utilized for further training of the AI. This particularly applies to AI applications directed at large user bases, which anyone can register for, either for free or through a monthly subscription. In paid applications or versions, it may be possible to limit usage to local operations, in which case the entered data is not used as training material. It is recommended that when inputting the aforementioned types of information, one should use the organization's own AI-powered applications in accordance with their usage instructions and limitations..
• Implement encryption, user management, and anonymization.
• Inform research participants about the use of AI applications.

Tips for using AI applications:

• Always validate and verify the outputs of AI tools. Seek out errors and ensure that the operation is appropriate.
• AI application operations aren't transparent, and they can make inferential errors. Do not use their results directly for conclusions. However, they serve as valuable research assistants for forming preliminary insights or detecting correlations.
• Familiarize yourself with effective AI application use and prompt design. Improving results/responses can involve asking follow-up questions, querying differently, rectifying errors in answers, refining queries, reinforcing desired responses, coding in small segments, initially providing context or setting, and offering examples of the desired type of responses. Beware that AI applications might have been manipulated maliciously or fed "poisoned" data.
• Only use high-quality data in standardized formats. Clean and well-structured data yields more reliable results. AI tools won't work well with incomplete, inconsistent, or poorly organized data.
• Answers produced by AI can be inaccurate, outdated, biased, prejudiced, offensive, against common sense, or unoriginal.
• Verify facts, data, and links from original sources. AI tools might hallucinate or fabricate sources. Not all AI tools reveal their sources for answers. In such cases, verify the accuracy of the information through traditional information seeking methods.
• Remember, some AI tools might lack the latest or essential information (data might not have been publicly available in written form during AI's training phase). Use multiple tools or methods to ensure a more comprehensive understanding.
• Take into account the limitations imposed by the language model used in relation to your data.

Document AI usage. AI tool operations can't be directly replicated, so thorough documentation throughout the process is vital. Record the prompts you use and the responses/results you receive.

Include the use of AI in your data management plan as well. Plan your AI utilization and consider if the data management plan should describe:

• How and when in your research you'll use AI.
• How you'll depict AI's use in research results.
• How you'll protect participants' personal data.
• How you'll address ethical and data protection concerns related to AI usage.
• The terms of use, restrictions, transparency, and reusability of the AI application you're using.
• Verifiability and replicability of research results.

References:

NYU Libraries. (2023). Machines and Society. https://guides.nyu.edu/data/home

OpenAI. (2023). ChatGPT.

University of Queensland. (2023). Artificial Intelligence. https://uq.pressbooks.pub/digital-essentials-artificial-intelligence/chapter/using-ai-tools-in-your-studies/

Amanda Wheatley and Sandy Hervieux. Separating Artificial Intelligence from Science Fiction: Creating an Academic Library Workshop Series on AI Literacy https://escholarship.mcgill.ca/concern/books/0r9678471

Chubb, J., Cowling, P. & Reed, D. Speeding up to keep up: exploring the use of AI in the research process. AI & Soc 37, 1439–1457 (2022). https://doi.org/10.1007/s00146-021-01259-0

-----------------------------------------------------------

Page updated on September 2, 2024.