Brief Introduction
Firstly, we invite you to consult the University Policy and the documents available on the dedicated page of the Unibo.it portal.
The introduction to these documents states: 'The University intends to use advanced technology and integrate it with awareness and responsibility, promoting the use of AI to serve people, never to replace them.'
Generative artificial intelligence (AI) refers to technology capable of generating new content, such as text, images, code, audio or video.
AI-based text generators are trained using large amounts of text extracted from the internet. This 'training' is then supplemented by testing and feedback from humans. With a few exceptions, there is little information available about how the models that drive these tools were trained. This means that, in most cases, users have no way of knowing what information the tool had access to.
AI-based technology is already present in many of the tools we use and is constantly evolving.
Some points to consider:
Most generative AI tools capture and reuse a significant amount of user data. Therefore, it is advisable to avoid sharing private information with such tools.
All AI-based text generators can produce plausible but incorrect information, known as 'hallucinations', which can result in distorted results.
- There are ethical and social concerns about generative AI, from how these tools were created, to how they are currently used, and how they will be used in future.
Information discovery with generative AI
Chatbots vs. search engines: what's the difference?
Although generative chatbots can answer questions in a way that resembles Google, these tools actually work very differently to the search engines we are used to. A traditional Google search helps users to find and retrieve text or other information that already exists and has been published on the internet. It does this by identifying the words and phrases in a user's search query, looking for existing sources that match these terms and applying a ranking algorithm to identify the most relevant results. (Google has recently started adding AI summaries to search results, which operate more like ChatGPT than traditional Google searches.)
In contrast, ChatGPT and similar tools generate new text in response to a natural language query, based on the information they have been trained on. In other words, search engines retrieve existing text and connect users to it, while generative chatbots create new text based on highly complex language models that attempt to generate the most probable sequence of words given the information they have been trained on.
However, generative chatbots do not reference, cite or draw information from specific or real sources. While a generative chatbot such as ChatGPT may generate or appear to cite sources, it is important to keep in mind that it does not actually 'read' them.
In some cases, chatbots can invent entirely fictitious quotes. These systems often generate responses containing unsupported claims and inaccurate quotations.
While tools such as ChatGPT, Bing Chat and Google Gemini have a broad scope, there are also AI-powered search tools that specialise in navigating and discovering academic research. These tools can return relevant articles and summaries of key information about them in response to a question, or provide contextual information about citations in a given article. Publishers of major scientific databases, such as JSTOR, Elsevier's Scopus and Clarivate's Web of Science, are also experimenting with AI-powered search tools to help researchers discover sources in new ways. All of these emerging search tools are experimental and should be used with caution. In some cases, publishers will not accept articles that have used these tools.
Using AI in research strategy
The research process is highly iterative, and generative chatbots can facilitate the exploration of ideas, the drafting of research questions and the discovery of new areas of interest. In the early stages, using an AI-based tool can be an effective way to initiate the brainstorming process. As well as well-known chatbots such as ChatGPT, Bing and Google Bard, new and specialised platforms are being developed that can be useful for generating research questions or creating effective search queries using Boolean logical operators.
Several tools are specifically designed to summarise articles or other documents. Many of these are paid tools designed to help students and researchers generate summaries of articles and other reading material without having to read the entire document (TL;DR). While these tools can be useful for managing a large number of articles and providing a basic understanding of their content, the sources should be read and studied in their entirety when a deeper understanding of the material is required.
Chatbots typically have limited memory for chat sessions, so they may struggle to generate summaries of very long documents.
Remember: these tools don't read or understand documents, so the summary will be influenced by the dataset on which they're trained. Test an AI tool with some well-known readings to gauge summary quality. When using any generative AI chatbot for your research, it is essential to develop your fact-checking skills, including: knowing how to identify original sources and verify citations or content “quoted” in the generated response. For academic research, you may need to use library resources if the information "comes from" an academic journal article. Always trace original sources before using them in your work!
Citation and attribution with generative AI tools
Before using generative artificial intelligence (GenAI) tools for your research, you should consult your lecturer or supervisor and take care when citing the output.
As the rules for citing AI-generated content are constantly evolving, it is important to refer to the latest guidelines from the main citation style guides or publishers. If in doubt, describe your use of generative AI in sections such as 'Appendix' or 'Methods'. For example: 'I acknowledge that I used Microsoft Copilot (version GPT-4, Microsoft, https://copilot.microsoft.com/) to summarize my initial notes and proofread the final draft.'
Remember that sources are cited for two reasons: to give credit to the author and to allow verification of the sources used. Bear these principles in mind when deciding how to use and cite AI-generated content. The use of generative AI should be acknowledged in any substantial use of it in academic work, for example in the drafting, planning or organization of content. Whether you need to declare this depends on the nature of the task, but some general principles apply.
The use of generative AI may not require acknowledgement when it is used for preliminary or technical purposes, such as familiarizing oneself with a topic, reading summaries of research findings, simplifying complex content for review, or improving grammar and text structure.
While some citation styles recommend citing generative AI systems like any other source, this raises several critical issues. Firstly, AI cannot be considered an author as it cannot take responsibility or generate original ideas; in fact, it reproduces pre-existing content. Furthermore, it is difficult to trace it back to a verifiable source. However, if the publisher or reviewer requires the use of AI to be cited and a specific style to be followed, it is necessary to be aware of and adhere to the relevant guidelines, which may include listing the tool as the 'author' in the bibliography.
Below are some examples and direct links.
According to APA style, AI-generated content should be cited as the output of an algorithm, with authorship attributed to the company that developed the model (e.g. OpenAI for ChatGPT).
According to the Chicago Manual of Style, the AI tool is considered the author and the developer company is considered the publisher (e.g. ChatGPT is the author and OpenAI is the publisher). If possible, the prompt used should also be reported, preferably in a note. The citation date corresponds to the date of content generation.
MLA style is more flexible than APA or Chicago style and recommends citing AI when incorporating its output, such as text, images, data or paraphrases, into your work. For indirect uses, such as translation or revision, an explanatory note in the appendix or methods section is sufficient. As AI-generated content is considered an authorless source, a descriptive title based on the prompt should be cited in both the text and the bibliography. If available, include a shareable link showing the content generated during the chat instead of the tool's generic URL.
[Credits: BROWN University Library]