When conducting a literature search, we recommend that you start by using the library resources, including Heritage, our online catalogue and our Journals and Databases LibGuide.
Here are some examples of how you could use AI alongside the library's resources to support your searching.
However, as with any use of AI, there are limitations to be aware of and these include:
Haman & Skolnik’s 2023 study illustrates why caution is needed. They asked ChatGPT to list 10 seminal academic articles in medicine and to provide the DOI for each. The prompt was run 5 times, creating 5 unique chats, each with 10 publications. In total, it generated 48 unique articles. However, only 8 out of the 50 DOIs and only 17 out of the 50 articles existed in databases. In 66% of cases, ChatGPT gave a non-existent paper.
Best practice would be to check all information and references created by Generative AI in trusted, academic resources such as those provided by the library.
Haman, M & Skolnik, M. (2023). Using ChatGPT to conduct a literature review. Accountability in Research: Ethics, Integrity and Policy, 31(8), 1244 - 1246. https://www.tandfonline.com/doi/full/10.1080/08989621.2023.2185514
A prompt is the input given to an AI model to elicit a response. It can be a question, statement, or any text that directs the AI on the desired output. The quality and clarity of the prompt can greatly affect the relevance and accuracy of the AI's response.
Here are some general principles to use when prompt writing to help ensure that the AI you are using generates accurate, relevant and useful responses.
Be Clear and Specific: Clearly define the task or question and avoid ambiguous language
Provide Context: Include relevant background information to guide the AI
Set Expectations: Indicate the format or style of response you want and mention any specific requirements
Keep It Concise: Avoid long or complex prompts by focusing on essential information
Iterate and Refine: Test and adjust prompts based on the AI’s responses
Consider the Audience: Tailor the prompt to the intended audience’s knowledge level using appropriate vocabulary
Prompt Frameworks provide a structured approach to writing effective AI prompts and help the AI understand your request. Here are some examples of frameworks you could use.
Google 5-step method (TCREI)
Task: What do you want the AI to do. Specify a persona or format.
Context: Provide necessary details to refine the output
References: Include examples to guide the AI's response
Evaluate: Did the input you provided give you the output you needed?
Iterate: Experiment, try again, adjust your prompt, add more information
The CO-STAR framework
Context: Why do I want the model to do that?
Objective: What do I want it to do?
Style: In what style should it write?
Tone: How is the emotion and sentiment conveyed?
Audience: Who is the recipient?
Response: How should the model's answer be structured?
Generative AI tools, such as ChatGPT and Copilot, can be useful for brainstorming a topic at the beginning of the research process. However, you will need to check the accuracy and reliability of any information provided.
Tools such as Connected Papers, Research Rabbit and Lit Maps can be useful to show connections between research papers, to visualise the literature on the topic and to overcome any limitations caused by having to know key words and synonyms.
Generative AI such as ChatGPT and Copilot can be useful for generating keywords and synonyms to use when searching. They can also be useful for formatting search queries that are appropriate for different databases.
Remember that any results will need scrutiny and refinement.
Note: ChatGPT's training doesn't include PubMed, so it might not always find relevant medical studies during a search.
Semantic Scholar is a search and discovery tool with partnerships with some academic journal publishers.
Connected Papers, Research Rabbit and Consensus can be useful when looking for research papers with similar themes.
AI tools such as Rayyan AI could be useful for the screening process as they provide consistency in interpretation.
Tools such as Elicit can extract data from PDFs into an organised table.
However, it is unable to access full texts behind paywalls, and it is unable to carry out advanced cognitive tasks necessary for comprehending and synthesizing literature.
Whitfield & Hofman. (2023) describe Elicit as 'a supplement to traditional library database searching for advanced searchers.'
Whitfield, S & Hofman, M. (2023). Elicit: AI literature review research assistant. Public Services Quarterly, 19(3), 201 - 207. https://www.tandfonline.com/doi/full/10.1080/15228959.2023.2224125
SCITE AI looks for publications that cite the title of the paper entered into the search and it can tell you if there is any supporting or contrasting evidence.
Research carried out by Qureshi et al.(2023) concluded that while ChatGPT has potential to create summaries from abstracts, which would help when synthesising research, 'there were errors that suggest the technology is not yet ready for such a task.'
Qureshi, R., Shaughnessy, K., Gill, K., Robinson, K., Li, T.,& Agai, E. (2023). Are ChatGPT and large language models "the answer" to bringing us closer to systematic review automation? Systematic Reviews 12, Article 72.https://systematicreviewsjournal.biomedcentral.com/articles/10.1186/s13643-023-02243-z
The following ethical considerations need to be remembered when using any AI tool:
Robust and Secure
Fair