AI: Library tools and resources

You must document your use of AI, whether it is for your academic work as a student or for publishing in a scientific journal. You should record all information for transparency and proper referencing of your work. You should also document all the output you generate.

Are you a student? Read “How to document your GenAI use?” from WUR.

Are you a WUR researcher, PhD candidate or postdoc? Read the policy of the journal in which you want to publish.

In addition, do not cite AI as if it were a source or an author. AI generated text, images or documents are not (primary) sources of information. You should always read and cite the primary source of information.

You may be tempted to quickly get scientific information on a topic by simply typing a prompt in genAI tools, such as ChatGPT, Copilot or Gemini. However, the underlying data and parameters in these chatbots are unknown, and the answer you get may be biased or incorrect (hallucination). These tools have not been developed to search for scientific information.

Some AI tools can help you search for scientific information. These fall into two categories: tools that provide a list of references based on your prompt, such as Elicit or Consensus, and tools that provide a map of interconnected and linked papers, such as Connected Papers or Litmaps. A curated list of these tools with information on the underlying database, privacy policy, limitations, and costs is offered by the Tübingen University Library under “Literature Research with AI – Tools and Introduction”.

WUR Library strongly advises to use these AI tools only in addition to Boolean searching for literature in bibliographic databases, such as Scopus, CAB Abstracts, or PubMed. The scope of the AI-based tools is limited and may be restricted to a specific field of study or to open access publications only. You may also miss relevant literature because these tools suggest only a limited number of articles.

You can use tools like ChatGPT to help you construct your Boolean search query for a bibliographic databases. Note that you must know how your chosen database works and how to carefully choose prompts to get a good output. Realise that you still may miss important search terms, that the syntax of the genAI-generated query may not be correct and that you will not use all functionalities that a bibliographic database offers. You can find examples of prompts you can use to help you create search queries for bibliographic databases in the Libguide from Birmingham City University.

AI tools can assist you to screen and select publications faster. These include ASReview LAB, Rayyan, EPPI-Reviewer, or Covidence.

ASReview LAB is a free-open source machine learning tool that screens and systematically labels a large collection of textual data. The tool is developed and coordinated by Utrecht University. You can install the software locally on your device and keep full control over your data.

ASReview LAB uses active learning. It learns from the selections you make based on your inclusion/exclusion criteria. At the start of your search, you have to provide the tool with a training set of at least one relevant and one irrelevant record/publication. After each decision, the tool re-ranks the publications that you haven’t seen yet and moves the publications that are probably more relevant to the top. When you continue screening, you’ll find more and more irrelevant publications. At a certain point, you can decide to stop screening and thus save time.

Rayyan also uses artificial intelligence to help you with reviewing. You can label individual references with the reason for inclusion or exclusion or other useful terms related to the topic, population, or geography. The basic version of the tool is free.

AI tools can help you to automatically extract data from publications and to structure the information you found, for example, in a spreadsheet, database or data lake(house) with information, such as the studied population, setting, geography, the number of test subjects, the methodology, statistical results, etc.

Elicit is an AI tool that can search, summarise, and extract data from scientific papers. You can also upload papers yourself if copyright allows it. Read more on data extraction with Elicit in this libguide from Birmingham City University.

DistillerSR is an AI tool that can be used in all steps of a systematic review, including data extraction. Note that this is not a free tool.

Alternatively, you could use models (out of the box or custom built) based on (gen)AI. If you are interested to hear more about these, you can email mdt.library@wur.nl.

Copyright notice: Be careful with uploading copyright protected materials in AI tools. You are allowed to do text and datamining on copyright protected materials for non-commercial research purposes. For education or any other purpose, you have to be careful as some publishers do not allow you to upload their licensed materials into AI tools. Publishers can request an ‘opt-out’, meaning that you cannot use their materials for text or datamining. Most publishers actually do not (yet) have such an ‘opt-out’ for AI training, meaning that you are allowed to use their materials for text and data mining with AI tools. Ask the Copyright Information Point for help on whether you can use copyright protected materials in AI tools.

Submitting a paper:

Using AI tools for writing & publishing raises concerns on the publication’s quality and reliability, academic integrity, and authorship and attribution. Biases and inaccuracies generated by the AI tool undermine academic integrity. Who can claim ownership of AI-generated content? Current intellectual property laws do not yet provide clear answers to this. AI tools may lead to plagiarism.

WUR Library urges you to be transparent in your use of AI tools and to always check a publisher’s policy before including AI-generated content in your publication. Most publishers require that you explicitly mention your use of AI, while others forbid its use for certain purposes, such as creating figures. In addition, WUR has guidelines on the use of genAI tools in PhD research.

The following is a non-exhaustive list of links to publisher AI policies :

• Elsevier
• Springer (incl. BMC and Nature portfolio)
• Wiley
• Taylor & Francis
• Oxford University Press
• Sage
• Brill

Peer-review:

For confidentiality and proprietary reasons, WUR Library advises reviewers never to upload a non-published manuscript into any external hosted tool, even if you only want to improve spelling or grammar. By doing so, you breach your peer-reviewer's confidentiality and could be uploading sensitive data. Moreover, the peer-review process often requires a high-level of understanding of the topic, which one could argue can only be performed by a fellow human. Some publishers already have policies in place for reviewers.

• Elsevier
• ASM
• Springer

When selecting an AI tool, you need to consider many aspects. Ask yourself critical questions when selecting or rejecting a tool:

• Purpose. How do you intend to use the tool? What is your (information) need?
• Developers. Where are the headquarters or the servers physically located? Who are the funders? Are there potential conflicts of interest?
• Accuracy and reliability. Is the tool you selected prone to hallucinations (incorrect results)?
• Bias. Are you aware of the potential bias the tool may give?
• -Transparency. How was it trained? What is the underlying database? What data does it use? How does the tool make decisions and produce specific results?
• Data privacy and security. What kind of data is collected by the tool? What are they used for (for example: training)? With whom are they shared? Does it comply with data protection regulations (GDPR)?
• Access. Is it a paid tool? Note that WUR Library currently doesn’t offer access to any paid AI tool.
• Your AI literacy level. Do you have sufficient knowledge to use the tool effectively and safely?
• Your subject matter expertise. Do you have sufficient knowledge to evaluate the results generated by the tool?
• Repeatability. Do you or somebody else need to be able to repeat your work, e.g., for a systematic review?

WUR has also rated different tools based on accuracy and quality, flexibility and features, and data security and privacy (students – staff).

AI outputs may be incorrect (hallucination), incomplete, outdated, or biased. Users always need to evaluate the answer the tool gives. This applies to all AI outputs, including generated texts, data extracted from a publication, provided references, generated images, etc.

Ask the tool to provide you with the sources to support its claim.

Read the sources! Are they real? Are they trustworthy? Are they of good quality? Are they unbiased? Are the sources supporting what the AI output claims? Why were these sources chosen? Are there better sources, or missing sources that impact the information the AI tool gives you?

Lateral reading

You can fact-check the information provided by a genAI tool by using a method called lateral reading. This method is designed to evaluate the credibility of an information source by comparing it with other sources. This is not about evaluating the sources provided by the genAI tool. This is about searching for information on the same topic via different search systems, for example, through a bibliographic database, a Google search, or in the media. You can then decide whether or not the information provided by the genAI tool is true, false, misleading, biased, etc.

Comprehensiveness

Does the tool give you the level of detail you need or does it just give you a small set of publications? Are these publications the most relevant ones for you and do they fit within your aim? Or do you miss publications that are important to find?

A note on paper mills

Paper mills are companies that publish poor quality articles or fake articles. Some paper mills use genAI to generate fake articles and publish them in their journals. You can read more on how to identify papers coming from paper mills on retraction watch.