Scientific research with AI tools
What are AI tools?
AI tools are applications in which “artificial intelligence” is integrated. Generative AI is particularly in focus at the moment. A generative language model specifies the probabilities of how different words relate to each other (e.g. in which contexts they occur, which word follows another, etc.). The major further development NLP (Natural Language Processing) allows input requests (so-called prompts) to be formulated in natural spoken or written language.
Limitations and risks of AI tools
Common AI tools can supplement literature searches in catalogs and databases, but cannot replace them.
- AI tools are currently no substitute for research in specialist databases with scientifically sound and extensive content, especially for complex literature searches.
- AI tools usually do not display any sources, can provide incorrect data and only access content that is available free of charge online.
- Scientifically highly relevant and up-to-date information behind the so-called “paywall” cannot be evaluated, e.g. from licensed databases such as Scopus, MLA or Web of Science.
There is a wide range of AI tools that are constantly changing. Achieving good results requires precise knowledge of the data sources as well as the functions, advantages, limitations and characteristics of the tools.
Data basis / paywall
Common AI tools such as Semantic Scholar are currently unable to evaluate content behind the so-called paywall. The results lack high-quality, up-to-date scientific information from paid e-books, e-journals and databases such as Web of Science, MLA or Scopus.
Fake News
The results of searches with AI tools show probabilities of word sequences. These are not necessarily facts. Specialist knowledge is required to assess the results. A fact check is advisable because the following problems can occur, among others:
- Output of false, misleading, incomplete or one-sided information
- Data hallucinations (e.g. indication of non-existent publications)
- No or incorrect indication of data sources
Data security
Personal registration is usually required. In addition, the operators regularly use the communication with the AI tools to improve their systems by viewing and evaluating prompts and result texts.
Costs
The difference in quality between the free and paid versions, for example, is considerable. Free versions are often limited in terms of data. Paid versions with monthly costs are required to integrate the currently available database.
Can I use AI tools in my studies?
Keep yourself regularly informed about the decisions of your institute, your faculty and the RUB regarding the use of AI tools for student work. If in doubt, ask your lecturer or academic advisor.
What does the use of an AI tool mean for me and my studies?
The use of AI tools has both advantages and disadvantages. For example, AI tools can support and facilitate learning and work processes. At the same time, their use can reduce your own skills acquisition, e.g. in the area of research or writing skills.
You should therefore consider this:
- What are the advantages and disadvantages of working with an AI tool for me?
- Do I want to use an AI tool and for what?
- Is the AI tool suitable for scientific work?
- Do I have the necessary knowledge to use the tool?
- How can I check my results from working with the AI tool for correctness and reliability?
How do I find the right AI tool for my scientific research?
The AI models are constantly being developed further, so that new functions are added practically on a weekly basis
You can find the latest information here:
Prompting
What is Prompting?
The input commands / questions to the AI are called prompts. Prompting is the way in which humans can interact with AIs. The quality of an AI's response is heavily dependent on the formulation of the prompt. The topic of prompting is also developing rapidly.
Clarity and precision:
Formulate your request as clearly and precisely as possible. Avoid ambiguities so that the AI model understands exactly what you want to know. Short, easy-to-understand sentences are helpful. Avoid filler words.
Specifity:
Be specific in your request. General or overly broad questions can lead to imprecise or overly general answers.
Phrasing:
Formulate your prompts neutrally and avoid biased or suggestive wording to get unbiased and objective answers. Experiment with different wording and structures to see what kind of prompts generate the best responses. Use the feedback from the AI model to refine your prompts. Formulating the query in English can lead to more concrete results.
Contextuality:
Give sufficient context to your query. A good prompt contains all the necessary information that the AI model needs to generate an accurate response. For example, assign a specific role to the AI tool, state the goal of your task, specify a form for the result (e.g. text, a table, a graphic...)
Structuring:
If necessary, structure your prompt with bullets, numbering or paragraphs. This helps the AI model to recognize the different parts of your request and answer them accordingly. Ask for step-by-step processing to get a well-structured result.
Use of examples:
If necessary, add examples or analogies to clarify your query. This can help the AI model to better understand your query.
Adaptation to the AI model:
Consider the specific capabilities and limitations of the AI model you are using. Different models can react differently to the same prompts.
Focus on the target:
Keep your goal in mind and make sure your prompt is directly aligned with it. Avoid unnecessary information that could confuse the AI model.
Consider topicality:
If your request requires up-to-date information, specify this. However, please note that AI models may not always have access to the latest data.