Artificial Intelligence in the Academic Health Sciences Guide

There are hundreds of thousands of AI tools available online. In this guide page are just a few tools you can explore using for your research. On this page you will also find links to resources on how AI has been used in health sciences research.

Note that many AI tools use the data you enter to train their systems. Be cautious of what information you input into AI tools, and make sure to review UC's policies on generative AI use.

GenAI is great for brainstorming or as a creative partner. Possible uses include:

• Ask for help coming up with or refining a research topic. For example, ask it to generate potential questions for a broad topic.
• Prompt it to help you brainstorm keyterms and search strategies for your topic, then plug these into a library database.
• Use it to help you outline your paper or give you a project template you can refine.

GenAI is not as good coming up with specific resources or citations. Do not assume it is accurate, as GenAI tools can create plausible but incorrect information about any topic. 

Artificial Intelligence has generated much interest among researchers conducting systematic reviews and other evidence synthesis projects.  The information below provides links to guidance, tools, and articles on the use of AI in the systematic review process.

Guidance for the use of AI in systematic reviews:

• Cochrane Handbook 4.6.6.2: Automating the Selection Process
  The linked guidance from Cochrane discusses the use of automated tools/artificial intelligence in three areas of the systematic review process. They give examples of where it is currently being used, as well as areas where it is not currently recommended. Covidence has integrated a few of the tools--Cochrane RCT classifier, relevancy ranking, and text highlighting--into their product. See Automated tools for systematic reviews below for more details. 
• Artificial intelligence (AI) technologies in Cochrane
  View the recent recording of a Cochrane web clinic on the use of AI technologies in Cochrane reviews. Learn how Cochrane currently uses machine learning, the challenges and opportunities of generative AI use in evidence syntheses, and their focus on responsible use of AI in systematic reviews. 
• 2020 Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA)
  PRISMA 2020 guidance states that authors should describe how automation tools were used in the review process, including: search strategy design or translation, study selection process, data collection process, study risk of bias assessment, etc. See the PRISMA 2020 Expanded Checklist and Explanation and Elaboration (E&E) documents for more details.

Automated tools for systematic reviews

• Covidence Cochrane RCT Classifier
  Learn more about using the Cochrane RCT classifier when screening records in Covidence.
• Covidence Relevancy Ranking
  Learn more about how Covidence uses machine learning to speed up the title/abstract screening process through relevancy ranking.
• Covidence Text Highlighting
  Learn how to create and manage text highlighting in Covidence to speed up title/abstract screening.
• PubMed PubReMiner
  Use PubMed PubReMiner to harvest common keywords and subject headings (MeSH) terms. Copy and paste PubMed unique identifiers (PMIDs) and click 'Start PubMed PubReMiner' to see how frequently keywords and MeSH terms are used in those articles. This tool can also be used to identify potential journals for publication or key authors in the field.
• SR Accelerator
  The SR Accelerator is a collection of automation tools to help make the process of evidence synthesis projects easier and faster. Tools include: The Methods Wizard, Word Frequency Analysis, a Deduplicator and more.

Articles about the use of AI for systematic reviews:

• Using artificial intelligence methods for systematic review in health sciences: A systematic review
  "This review delineated automated tools and platforms that employ artificial intelligence (AI) approaches and evaluated the reported benefits and challenges in using such methods."
• Semi-automated Tools for Systematic Searches
  Provides an overview of semi-automated tools used in the systematic review search process.
• Can ChatGPT Write a Good Boolean Query for Systematic Review Literature Search?

​​​​​​​"In this paper, we investigate ChatGPT as a means for automatically formulating and refining complex Boolean queries for systematic review literature search."

NotebookLM

• Some Strengths
  • Notebook app that allows you to upload sources and ask Chatbot questions based on those uploaded documents
  • Chatbot responses are grounded in the information you provide
  • Chatbot provides citations for its responses (i.e., it highlights where in the documents you uploaded it drew its information)
  • Great for summarizing uploaded notes and extracting key concepts
• Some Limitations
  • Chatbot cannot search the web (it's limited to searching the documents you upload)
  • Can only upload PDFs or Google Docs, or copy and paste text
  • Limited to 20 document uploads

Research Rabbit

• Some Strengths
  • Finds related articles based on citation networks, and provides visualizations of these networks
  • Works best if you already have at least one relevant article in mind (i.e., a "seed article")
  • Can upload an RIS or BibTeX of relevant articles to analyze
  • All features are free!
• Some Limitations
  • Metadata for articles can have errors (e.g., listed authors are sometimes incorrect)
  • Not all articles are scholarly (e.g., it sometimes retrieves things like libguides!)
  • Limited to articles available in Semantic Scholar or PubMed

NLM's LitSuggest

• Some Strengths:
  • Model you can train to alert you to new records in PubMed, ranked by relevancy to your topic
  • Allows you to tag records as relevant or irrelevant to better train the model
  • Allows you to download records marked as relevant and/or irrelevant
  • Great for helping you stay up to date on topics of interest
• Some Limitations:
  • Takes a long time to load
  • Need to have already compiled at least 10 relevant studies to train the model
  • Limited to records in PubMed

Microsoft Copilot, ChatGPT, and Gemini

• Some Strengths:
  • Chatbots answer questions using information from the web or inputted data
  • Great for summarizing and outlining inputted information
  • Great for brainstorming ideas
  • Can be used to identify search terms
• Some Limitations:
  • Susceptible to hallucinations
  • No quality control in the sources from which it extracts information for its responses (unless you tell it to only base its responses on information you input)
  • Doesn't always cite the sources it uses for its responses (or sometimes makes up citations)
  • Information in its responses may be outdated
• Note: The University of Cincinnati has a subscription to Microsoft Copilot Chat. You can learn more about the subscription here.

Elicit

• Some Strengths
  • Retrieves and summarizes literature on a query
  • Provides a summary that integrates the top 4 retrieved papers (for the free version), and extracts information from individual articles (e.g., main findings, methodology, intervention, etc.)
  • Can extract information from uploaded PDFs
  • Has a chatbot which you can ask questions about a paper
• Some Limitations
  • Not all articles are scholarly (e.g., it sometimes retrieves things like libguides!)
  • Isn't always reliable when extracting information from articles (especially for articles where it only has access to the abstract)
  • Searches are not transparent (searches wouldn't be reproducible or comprehensive enough for something like a systematic or scoping review)

SciSpace

• Some Strengths
  • Retrieves and summarizes literature on a query
  • Provides a summary that integrates the top 5 retrieved papers (for the free version), and extracts information from individual articles (e.g., limitations, conclusions, methods, etc.)
  • Can extract information from uploaded PDFs
  • Has a chatbot where you can ask questions about a paper
• Some Limitations
  • Limited usage for a free account (e.g., only 5 searches per day)
  • Metadata for articles can have errors (e.g., listed authors are sometimes incorrect)
  • Searches are not transparent (searches wouldn't be reproducible or comprehensive enough for something like a systematic or scoping review)

Perplexity

• Some Strengths:
  • Chatbot that answers questions using sources from the web
  • Provides links to sources it uses in the query
  • Great for finding webpages about a topic
• Some Limitations:
  • Sources are typically non-scholarly
  • Only provides a few sources per query

For a list of additional AI tools, see Ithaka S+R's Generative AI Product Tracker.