AI and information retrieval 2.0

Generative AI, ChatGPT and information retrieval

Using AI technologies in a responsible and strategic way can support your studies. Generative AI tools such as ChatGPT can help you quickly gain an overview of a new subject area. For example, you can explore subject specific terminology, request summaries from different perspectives, or test ideas. However, it is important to remember that you are ultimately responsible for your own learning and search process.

Generative AI is based on probabilities. It predicts how a sentence is likely to continue based on patterns in the data on which it has been trained. Earlier generations of AI tools were unable to provide sources for their claims, but AI-generated responses today often include links. Nevertheless, the generated content can still be inaccurate and the links provided may not correspond to the information presented.

When using AI tools to identify articles or other sources for your assignments, it is important to verify the accuracy of the information and ensure that you can determine where it originates from. Since this can be difficult to ascertain in AI tools it can be better to use a scientific database instead.

Information retrieval tools

There are dedicated literature discovery tools that use AI to help you identify relevant articles and other sources when you already have a few key publications to start from. These tools can use both citation analysis and semantic similarity to suggest additional articles. In some cases, the results are presented as an interactive graph, allowing you to explore connections and discover more relevant material. There are also tools that use AI to interpret questions expressed in natural language.

In citation analysis, further articles are suggested if they use the same source/s cited by the article you found.

In semantic searches, a language model is used to attempt to understand the meaning of a document or research question and to match it with similar content that does not necessarily use the same formulations.

Although general-purpose chatbots may provide links to websites or articles, they are not designed specifically for information retrieval. It is important to use the right tool for the right task.

Various AI technologies are also integrated into many of the traditional databases available through the library's database list. Many database providers are also planning to expand the use of AI within their services. AI can be used to broaden search results, improve relevance ranking, or generate summaries.

Which materials are searchable using the tool?

This is a question you should always ask when searching for information, but it is particularly important when using new AI tools. Some tools only include freely available (open access) content. Contents subscribed to by the library will not be included. A tool may provide broad coverage across a range of subject areas or be suitable only for specific disciplines. The number of included publications also varies between tools. The type of content can differ as well, with some material being scholarly and some not. As with any information search conducted for your studies, it is important to evaluate the sources you find.

On which data were the incorporated AI models trained?

To identify similarities between articles, a model is trained on large amounts of data. Although these datasets are extensive (hence the term “large” in large language models) they are not unlimited and have constraints in terms of size, scope, and currency. The data used to train these models is rarely disclosed.

Imbalances in the training data may be reflected in the model and risk generating biases. If training was done on data from certain sources during a particular period, generalisations from this period and the sources will be seen in the delivered results. A model trained on technical literature may for example be less accurate when searching for related articles in the humanities and social sciences.

How is the information you enter used?

The information you enter into a tool, or any documents you upload, may be stored and used in ways that are beyond your control. Some tools continuously improve their models by incorporating data provided by users. You should never enter sensitive data, personal information, or content that that may not be freely distributed, such as copyrighted materials including books and articles that are not openly accessible.

What transparency and replicability criteria should the search fulfil?

Transparency, openly accounting for how a result was achieved, is one of the fundamental principles of research and science and should always be applied in your studies.

Replicability is another fundamental principle entailing that someone else should be able to repeat your process and achieve the same results.

All AI tools have tissues with transparency and reproducibility because they are so-called “black boxes”. You provide the tool with input in the form of a query (often referred to as a “prompt”) or a set of articles, and it produces a result. However, it is difficult to explain exactly why a particular result was generated. Using AI tools is therefore fundamentally different from using subject headings and search terms in traditional scholarly databases.

AI models are trained using probability theories, which means that the same input can produce different results. This affects both the transparency of the tools and the reproducibility of their outputs. Regardless of which tool you use, it is important to be transparent about the tools you have used and the results they have generated. You can consult the library’s referenc guides for information on how to cite a chatbot correctly.

There are many AI companies that provide chatbots today, for example OpenAI (ChatGPT), Google (Gemini), and Anthropic (Claude). There are also many tools that use generative AI in other ways. These include a range of AI-based tools designed to help users find scholarly articles, for example Semantic Scholar, Perplexity, and Undermind.

Please note that the library neither subscribes to nor provides support for any of these tools.