AI-Tools for Literature Research

There are specialized AI tools for searching scientific literature that can be a useful addition to conventional research tools. These tools work on the basis of a literature database that is searched using AI.

Research tools can be divided into two groups, semantic and citation-based tools:

Semantic tools use an initial question to perform a search. The tool creates a list of literature relevant to the question. These tools usually have functions that evaluate the literature found by creating summaries or answering the initial question based on the documents found. Here, the line between semantic tools and generative AI is fluid! Semantic tools are well suited for getting started with a topic.

Citation based tools work with one or more seed papers, that must already exist in your own literature list. Based on these, further recommendations for suitable articles are provided. These tools are particularly useful if you have already found some articles that are relevant to your topic.

It is important to assess which tool can be helpful for one's own search, as coverage can vary significantly depending on the field of study. A look at the data foundation is essential. Additionally, costs and registration requirements should be considered, although most tools offer usable free versions. 

A comprehensive list of literature research tools is available from the University Library in Tübingen. Here you will also find information on the data foundation or on the privacy policy of the tools.

Traditional literature databases are also increasingly offering AI functionalities. Examples include:

  • Business Source Ultimate: This database allows users to search in natural language and provides AI insights for many documents.
  • Statista: With Statista AI, it offers natural language search and generates a short summary based on the results.
  • HeinOnline: Offers AI summaries for many documents.
  • Web of Science: With Smart Search, it offers natural language search including translation. For many documents, there is also a graphical co-citation map.

 

ChatGPT and Literature Research

ChatGPT is a language model that is trained to create texts and imitate human communication. It is important to know that ChatGPT is not a knowledge model. It does not work with facts and knowledge, but uses statistical probabilities to arrange words in such a way that a meaningful text is created.

What ChatGPT can do: Writing poetry or e-mails, translating text, summarizing or structuring text, collecting ideas, programming

What ChatGPT can do only to a limited extent: Reproduce facts, literature research, assessment/evaluation

Most language models now have a search engine integrated, which enables them to perform an internet search and display the results in a processed form. This makes it appear as if the language model actually ‘knows’ something. But here, too, it is purely a matter of calculations; the model reproduces what it finds, but it does not understand what it is writing.

Usually, answers sound very convincing and are often correct, but not always, as a recent study by the European Broadcasting Union has confirmed. The more specialized a topic is, the greater the risk of convincing-sounding “hallucinations.” It is therefore absolutely necessary to check the facts.

ChatGPT can be a good support for literature research. It can help to identify suitable search terms, find synonyms and link them together. Here, too, it is important to remain attentive and check answers.

 

Prompting

The prompt is the work instruction to the AI. The more precisely it is formulated, the better the result.

There are various ways in which prompts can be designed, as a one-off request or as a conversation. Some helpful tips for successful prompts:

  • Define a role for the AI (According to recent findings, such role assignments no longer improve factual accuracy, but can change the tone of the output).
  • Formulate a task: What exactly is to be done? Brainstorming, a structure for a presentation on topic x, a concept for a lesson or a congratulatory card for passing an exam
  • If necessary, specify individual steps: Create a word list, then use it to formulate a search query ...
  • Provide context and background information: Framework conditions, purpose, target group …
  • Define the form and length of the output: Table with three columns, bullet points, a text with a maximum of 100 words ...
  • Formulate precisely: consistent choice of words, avoid vague formulations (such as, if possible, etc.) and negations.
  • Ask the AI if it needs more information and refine the result if it is not yet satisfactory.

 

When using AI tools, it is important to act responsibly and knowledgeably, and to be aware of the potential negative effects and risks. You can find more information here.