Making the most of Generative AI

Finding information

• Using Generative AI can benefit literature searching, particularly at the early stages of a search.
• It can provide quick summaries and new areas of investigation for different topics.
• An AI search isn’t suited for in-depth academic research on its own and needs to be used alongside more traditional search tools, such as Library Search.

Using AI to find information

Summaries and definitions

Generative AI can provide quick summaries of topics. If you are starting to investigate a subject, especially quite an unusual or specific one, AI can provide a useful brief overview or set of definitions to help you familiarise yourself with a new topic. Read Getting started with Generative AI to help you generate prompts for finding information.

The examples below have been generated using ChatGPT:

Generative AIs can also provide summaries of books, paragraphs and sections of text so that you can understand the main key points by generating a plain English explanation.

Build a keyword vocabulary list

To effectively use Library Search, you could use AI to help you generate a list of keywords and alternatives. Having a list of additional key words in your search for literature can lead to finding relevant books and articles you may not have discovered using the words you initially considered.

The following example was generated using Google Bard. Each term has also been provided with a definition.You could enter the suggested search terms in Library Search.

Ideas and inspiration

AI can provide a useful ‘second opinion’ when it comes to deciding how to begin your literature search, suggesting possible avenues for investigation which you may not have thought of. Even if the tool’s response itself isn’t especially detailed, it may well provide ideas for new search terms and directions you could use for carrying out further literature searching.

Save time

Often we find ourselves spending a long time in databases, or in the library search, trying to find a book or article which contains something simple. However, asking the same question in AI can provide quicker results. Many AI tools are capable of responding to questions and prompts using ordinary, everyday language. Reading this information can give us the ideas and search terms we need to find information in our search.

Please also be aware that AI tools often have a variety of settings and controls that might affect the answer they give you: fine-tuning an AI tool to get the best out of it as a search tool can be every bit as complex as using any academic database!

Limitations for finding information

Purpose

Generative AIs may not specifically be designed as search tools; even ones that are may be works-in-progress or beta versions. Therefore, they may not always produce the best search results. Always check for any further information when using an AI tool to understand its current strengths and weaknesses.

Coverage

Generative AI needs to be trained on a set body of data: they may struggle with any information that falls outside that body of data. Answers may also become less reliable for more unusual or specific questions for which there are fewer sources to draw on, because there are fewer existing sources on that subject for them to mimic.

Accuracy

Generative AIs are essentially driven by predictive models rather than a genuine understanding of a subject, and so may generate vague, inaccurate or outright false information, a process referred to as 'hallucinating'. And, as with all software, user error can also be a factor: the quality of the answer they give may also be very dependent on the exact prompt you give them.

Hard to assess quality of information

Generative AI may not always tell you where they are drawing their information from, which makes it very hard to judge the reliability and accuracy of what they tell you. Some AI tools, particularly those specifically designed with searching in mind, are getting better at providing references to their sources so you can confirm their suitability and reliability; and it may be possible to prompt an AI to list its sources.  But even when tools appear to name their sources of evidence, they aren't necessarily telling the truth. AI tools are notorious for inventing fictitious references to non-existent sources, so always check the references actually go to real sources!

Starting point only

AI search tools usually collate different sources to create a short summary of a topic, which can make a useful starting point for research into an unfamiliar topic, much like Wikipedia. However, academic and professional research often involves questions which don't have a single, neat answer, because the topic is complex, disputed, under-researched, or hard to investigate (or all four!). So also like Wikipedia,  an AI summary usually won't have the detail and nuance you need to work out an answer in these cases, and you will need to refer to original sources.

Unsuitable for systematic searching

Many of the protocols and checklists used for carrying out systematic research have not yet been adapted for AI: and many AIs do not produce the consistent, reproducible response to a set input required for systematic searches. As a result, at the present time AI tools generally are not suitable for systematic searches.

Need for original sources

Generative AI can be very good at summarising information, generating ideas, or analysing data: what they can’t do is go off and carry out new research or experiments themselves! So you should still use the original sources in which the research was first published, as that will give you the fullest account of the research’s methods and conclusions, which you will need for critical reading and writing about the subject. Always use Library Searchor the subject guides, as this will give you full access to high quality academic literature.