This guide is aimed at students who wish to know how generative AI tools may be used to help with academic work. It currently covers using AI to help with literature searching and summarising. It also covers how to use AI ethically and responsibly.
In this guide we use three tools (CoPilot/ChatGPT, Claude, and Elicit) to conduct several different tasks related to literature searching.
Getting the best results from any AI tool requires you to write effective prompts. The guidance below from the University of Edinburgh's Bayes Centre provides some tips on how to write good prompts.
ChatGPT can help you get started with your research by suggesting research questions and search strategies you might use.
For example, you may know the broad topic area you are interested in but may not be sure how to make it more specific or write it as a research question. In the example below, the prompt I use is ‘I am an undergraduate student writing a dissertation on ways AI can help with the climate crisis. Suggest some research questions I could use’. I made my prompt as specific as possible by telling ChatGPT that a) I am an undergraduate student and that b) I am writing a dissertation. This gives the chatbot context and the level at which I am working.
It suggests a range of different questions such as ‘How can AI technologies be leveraged to optimize energy consumption and improve efficiency in various sectors such as transportation, manufacturing, and agriculture to mitigate climate change?’ and ‘What role can AI play in enhancing the accuracy of climate modeling and prediction, thereby enabling better adaptation and mitigation strategies?’.
If I added further context, such as the subject area I am working in, I get even more targeted questions. In this second example, I changed my prompt to ‘I am an undergraduate student writing a dissertation on ways AI can help with the climate crisis. I am working in the field of agriculture. Suggest some research questions I could use.’
The results I got have now changed to be more specifically about how AI can help with climate change in relation to agricultural topics. Example questions the chatbot suggests are ‘How can AI be utilized to optimize agricultural resource management (water, fertilizers, pesticides) to minimize environmental impact and maximize crop yield in the face of climate change?’ and ‘What role can AI play in predicting and mitigating the impact of extreme weather events on crop production and food security?’.
Do not feel you have to use any of these questions exactly as they appear in the chat. Rather, they can help you get a better idea of the questions you may want to ask and you can change them to better suit your area of interest.
If you know a question you need to answer, but you're struggling to develop a search strategy that finds relevant academic literature, ChatGPT can help suggest search terms you might use.
In this example, I ask ChatGPT to ‘Suggest some search terms I could use to search for academic literature on the topic of: What role can AI play in enhancing the accuracy of climate modelling and prediction, thereby enabling better adaptation and mitigation strategies?’
It responds with a list of search terms including ‘climate modeling accuracy’, ‘machine learning in climate modeling’ and ‘Climate forecasting using AI’. It also suggests using Boolean operators to refine my search.
At this point I could try some of these search terms or I can ask further questions. If you get any responses you are not happy with, either because they are incorrect, not very good, or you don’t understand what it means, ask further questions or tell ChatGPT that there is a mistake in its response. Do not just take what it says as correct in the first instance.
In this case I am not sure what Boolean operators are so I ask ChatGPT to clarify what they are and how I use them in a search.
I then ask ChatGPT to create an example search strategy using Boolean operators so I can see what it looks like in practice:
ChatGPT identifes the main concepts in my question (The role of AI, enhancing the accuracy of climate modeling and prediction, and better adaptation and mitigation strategies). It identifies search terms for each concept such as 'machine learning' and 'deep learning' as well as 'artificial intelligence', and 'climate forecasting' as well as 'climate prediction'. It then combines all of these into a search string that reads:
(AI OR "Artificial Intelligence" OR "Machine Learning" OR "Deep Learning") AND ("Climate Modeling" OR "Climate Prediction" OR "Weather Prediction" OR "Climate Forecasting" OR "Weather Forecasting" OR "Predictive Modeling" OR "Data-driven Modeling" OR "Numerical Modeling") AND ("Adaptation Strategies" OR "Mitigation Strategies" OR "Climate Change Adaptation" OR "Climate Change Mitigation" OR "Resilience Strategies").
In this way I have used ChatGPT to help me create an effective search strategy that I can use when searching library databases.
ChatGPT can be used to recommend scholarly journals across different academic subject areas.
For example, following the guidance on prompting in the AI Guidance for Staff and Students page, I construct a prompt in the form of "Acting as a (ROLE), perform (TASK) in (FORMAT)":
Take on the persona of a research support librarian. List 5 peer-reviewed scholarly journals I can use to find research in the area of political philosophy. The output should include a brief description of the scope of each journal.
ChatGPT responds by listing 5 journals: Political Theory, Ethics, the Journal of Political Philosophy, the Critical Review of International Social and Political Philosophy, and Contemporary Political Theory.
This is a useful list, as all five titles are well-cited, peer-reviewed journals. I can now go to DiscoverEd and perform a Journals Search to check the Library's access to these journals.
It is important to remember that though generative AI tools, such as ChatGPT, can be useful in suggesting scholarly journals in your subject area, it is not necessarily suggesting the most cited or top journals or a diverse range of journals taking in global perspectives (for example, journals published in the Global South). An excellent way to familiarise yourself with the scholarly journals in your subject area is to go to the Library subject guide for your area and look for information about recommended journals or the databases that enable you to search key journals for your subject. You can also ask your Academic Support Librarian for advice.
When asking a generative AI tool to make recommendations, it can be useful to provide examples with your question to provide additional context. When asking Gemini to recommend academic databases that I can use to search for academic journal articles in a specific subject area, I can name a database or databases I have found useful in the past. This will help Gemini to understand the kinds of database I am interested in.
It can also be useful to remind the generative AI tool to use relaible sources and to cite them, so you can track where its getting its information from.
For example:
I am doing research for my dissertation. I want to find peer-reviewed articles on the subject of business information systems. I have found the databases Business Source Complete and Science Direct useful for this. What other databases would you recommend? Answer using reliable sources only and cite those sources.
Gemini responds by listing 5 databases: EBSCOhost, JSTOR, ACM Digital Library, IEEE Xplore, and ABI/INFORM.
This list can provide you with a useful starting point. For four of the databases, Gemini provides a link to the database website, which you can check to find out more about the subject areas covered. You can check the remaining database on the Web. As full text access to articles in many academic databases is available only to subscribers, it is important that you check the Library's Databases A-Z to see access is available. You can then use the links to the databases on the A-Z list to access the database and run searches.
Though AI tools can suggest useful databases for literature searches, it is important to familiarise yourself with the range of databases in your subject area. Consult the Library's Databases by Subject page and/or your subject guide.
Elicit is an AI research assistant. Elicit uses language models to help you automate research workflows, such as undertaking parts of literature reviews.
Elicit can find relevant papers without perfect keyword match, summarize takeaways from the paper[s] specific to your question, and extract key information from the papers. [@elicit-research].
*Elicit Basic: essentially a free trial. You can't export results or transfer unused credits. Credits are deducted when you search, run workflows, add columns to tables, etc. For example, a search uses approx. 200 credits. You can only add more credits if you subscribe.
*Note: Elicit pre-defined columns currently have a biomedical focus.