Searching for information with AI
On this page, you will learn how to use AI to search for information, such as scientific articles, for your studies. You will get hands-on advice as well as an understanding of how the tools work.
How do AI search tools work?
The generative AI technology used in chatbots such as ChatGPT, Copilot and other AI search tools is called Large Language Models (LLMs). A large language model is a statistical model of human language and of what human-written texts usually look like. Chatbots and other AI search tools base their answers on probability calculations. In simple terms, this means that AI tools take your prompt (the text or question you typed) and generate an answer based on what is likely to be an answer. It is often said that models "hallucinate" when the answer does not match reality, that is, when they give incorrect or untrue answers. However, there is really no difference between what models do when they produce true versus untrue responses. In both instances, the AI tool has created text that matches what text usually looks like and what words usually appear together.
As a consequence, large language models may reproduce factual errors, stereotypes, or bias. They may also generate sources that sound credible but are, in fact, fabricated. Furthermore, different chatbots have been trained on, and have access to, different data. Chatbots such as ChatGPT and Copilot do not generate answers solely based on scientific material. AI search tools such as Elicit, Consensus, and Scite, on the other hand, have been designed to provide answers based on scientific publications, which they also provide references to. Some large language models have a built-in search tool that allows them to retrieve sources from, for example, Google, but not all. Therefore, you need to pay attention to how the AI tool you use works, and determine how you can best use it to generate the information you are looking for.
Example of question to KIB from researchers:
We read in a news article on a website about an interesting new research paper allegedly written by A and B. It is supposedly titled X, but we cannot find the article anywhere. Did A and B write this article, or could it be an AI fabrication?
KIB's answer:
Generative AI created this reference. The two researchers, A and B, exist and are both authorities in this field, but they have not written anything together. The reason AI generated this reference is that many AI models are based on probability. : A and B are two prominent researchers in the same field and it is probable that they could have written an article together. And if they had done so, it is likely that the article would have been titled X, based on what they had written before. The article, however, does not actually exist.
AI search tools that provide sources
There are AI search tools designed to retrieve real sources. These tools are often based on a large language model that is linked to a database or search engine through a technology called retrieval augmented generation (RAG). The technology allows the tool to retrieve real sources based on the prompt you have entered.
AI search tools can be connected both to more general search engines, such as Google, and to more scientifically orientated search engines, such as Semantic Scholar. In both instances, the large language model generates a text in response to your prompt using real sources. Please note, however, that AI search tools usually cannot access the full text. Instead, the generated text is based on abstracts.
Agent-based AI search tools
Agent-based AI search tools are helpful for tasks that require several steps. One example is the agent-based AI search tool Undermind, where a search for information is conducted in several steps. Simply put, you first chat with a chatbot that asks you to clarify exactly what you are looking for. Then, the tool generates prompts that are used to conduct more in-depth searches.
Snowball searching AI search tool
Some search tools use AI without being generative. These tools base their search on one or more articles that you have already identified to find similar articles. Snowball or citation searching is a search technique where you start from one relevant article to find more via, for example, reference lists or citations, or with AI search tools showing similar articles.
Quick facts
- Large language model is the generative AI technology found in chatbots and AI search tools such as ChatGPT, Copilot, Elicit and SciSpace.
- Large language models are statistical models based on probability calculations. Based on your prompt, they generate a probable answer based on the data they have access to. As a consequence, they may reproduce factual errors, stereotypes and biases.
- A large language model has no concept of what is true or false; it is only trained to generate language. It may therefore generate text that is consistent with reality, but it can also generate content that is completely incorrect.
- A large language model cannot summarise a text but only shorten it, because the model can neither evaluate nor judge what is important in the text, nor assess sources, nor apply critical thinking. The summary you get may even be incorrect. Therefore, you always need to read the original sources carefully.
- ChatGPT and Copilot provide answers that have not been generated based solely on scientific texts. If you need scientific information, use AI search tools specifically based on scientific material.
- Some large language models are linked to a database or search engine that uses a technology called RAG (retrieval augmented generation). These tools may provide real references.
- AI search tools usually do not have access to the full text, but base their answers on abstract s.
- AI search tools do not clarify why they have highlighted particular articles; the selection process is not transparent. Further, these tools cannot think critically or evaluate sources and the quality of the selected sources may therefore vary. You always need to check all sources carefully.
What AI search tools can I use?
If you need scientific articles, avoid using ChatGPT or Copilot. These tools are large language models not intended solely for academic use. As a result, you may end up with sources that are either fabricated or not scientific. Instead, you should use AI search tools that are designed to retrieve scientific articles. They work in slightly different ways and generate their answers based on scientific material. However, since AI search tools will search among a large amount of material, often hundreds of millions of articles, the results may include problematic studies or articles published in predatory journals. It is therefore important that you always review the articles you get from AI search tools to ensure that they are credible.
Questions to ask yourself when choosing a tool:
- How does it work?
- What types of search queries or prompts work best for this particular tool?
- Where does the tool search?
- Are the articles credible?
When can an AI search tool be used?
AI search tools are often perceived as more intuitive than traditional databases, as you may use natural language instead of keywords and Boolean operators. However, remember that AI search tools often conceal complex processes involved in searching, making your search less transparent.
Where you should search for information – in course literature and books, via Google, via a scientific database or via an AI search tool – depends on the purpose of your search and your requirements for the task or the answer. Before you start searching, you need to identify why you are searching for information to determine which approach to take. AI search tools may be suitable at certain stages of your search process. Below is a brief description of three different phases of the search process.
- Explorative phase of your search process? AI search tools may be useful.
- Complementary and snowball searching? AI search tools may be useful.
- Systematic and exhaustive search? Using AI search tools is NOT sufficient. A structured, transparent and replicable search in databases is needed. Read more about structured and systematic searching in scientific databases.
Differences between traditional databases and AI search tools
Traditional databases and AI search tools differ in several ways: how they handle information, how they present results and what they require from you as a user. The table below shows some key differences that may help you decide which tool to use for different purposes.
Quick facts
- Which tool you should choose depends on context. You always need to consider why you are searching and then decide where and how it is most appropriate to search.
- There are significant differences between traditional databases and AI search tools. For example, where they search, how they search, and what sources they have access to.
- AI search tools cannot summarise text or assess if an article is good, as they do not understand and cannot interpret the content of a text. They can only shorten the text.
- AI search tools can sometimes give wrong answers even if the source it refers to exists. Again, they do not understand and cannot interpret the content of the text.
- Chatbots such as ChatGPT and Copilot generate texts that are not based solely on scientific material. If you need scientific information, it is better to use AI search tools based on scientific material.
Prompt tips for AI search tools
When using AI search tools, you need to formulate a clear request to the tool (that is, prompt). The quality of the prompt determines the quality of the output. If you need scientific information, you need information that is both relevant and neutral. Ensure that your search does not introduce bias. Below, you will find less effective prompts for finding information, as well as more effective ones.
Keep in mind!
You are always responsible for your own learning and what you produce in your studies.
Make sure you do so with academic integrity, that is, be transparent about how you use AI tools and do not use them more than is permitted for your course.
Do not share personal information, sensitive data or copyrighted material with the tools.
If you would like us to get back to you, please submit your contact information in the form below along with your feeback.