3a. Limitations of AI and Other Sources#
Introduction#
In academic activity and industry, good decisions and solid designs depend on accurate, up-to-date, and trustworthy information. Throughout your thesis, you’ll rely on information from articles, websites, reports, and maybe even AI tools. But not all of these sources are equally reliable, and using weak or incorrect information can easily lead to poor results.
Why Evaluating Information Matters
What are main reasons to evaluate sources?
Considerations for GenAI Use
What are some major concerns for genAI use?
Step 1: Why Evaluating Information Matters#
Evaluating information matters because:
It prevents technical errors. Using outdated data, incorrect formulas, or unverified claims can lead to serious mistakes. For example, choosing the wrong material specifications, misinterpreting a dataset, or designing an inefficient algorithm.
It saves you time and effort. When you learn to spot low-quality or irrelevant sources early, you can focus on the material that actually supports your project.
It improves the credibility of your work. Projects or reports that are based on reliable, well-chosen sources are more convincing and show that you understand your topic deeply.
It strengthens your critical thinking. Engineers and scientists constantly have to assess whether data, methods, or solutions are valid. Evaluating information sources develops that same analytical mindset.
It prepares you for real-world problem solving. In professional settings, you’ll often need to judge the reliability of information, from technical manuals to software documentation or industry reports, before making decisions that have real consequences.
In short, evaluating information helps you build accurate, efficient, and credible work both in your studies and your future career.
Step 2: Considerations and Limitation of GenAI use#
GenAI tools can be helpful during the research process if they are used correctly. At the start of your project, you should check whether you are allowed to use these tools, and in what way. For more information on acknowledging your use of AI you can also have a look at this section.
THESIS SUPERVISOR
It is strongly suggested that you discuss your planned AI use with your thesis supervisor and check what is allowed.
GenAI use comes with a lot of considerations, and this means you should critically evaluate the output of a tool, as well as the need to use a tool.
It is often unclear how the input (prompts) from users are processed in GenAI tools. These could also be (and often are) used for training purposes. It is therefore important to never insert personal details or confidential information in GenAI tools. If you are allowed to use AI, and are handling confidential information, consider using a tool that wil run AI models locally on your laptop, like Ollama.
AI tools can produce incorrect or misleading information. The output of AI systems can, for example, include citations that look realistic, but don’t really exist. You should always verify the output of AI with reliable sources.
GenAI tools are trained with real-life data. The output can therefore reflect and amplify biases from human thinking. For example, research has shown that outputs of GenAI tools often include biases against women and people of colour. Other biases, for example confirmation bias (the tendency to search for and favour information that confirms your beliefs), can also affect the data that is used for training the GenAI tool, and therefore also affect the output. So, also from the perspective of potential biases, it is important to always critically check the GenAI output.
GenAI tools are trained with data that is publicly available. However, the authors of the texts that are used for training purposes did not give permission for this. Although GenAI output does not typically include directly copied passages, often it does include extracted patterns and styles from the works of others. This leads to a grey area of potential copyright infringement.
Completely outsourcing tasks to GenAI tools is not smart, not only because of the possible inaccuracies and biases in the output, but also because it leads to overreliance on GenAI. Overreliance, in turn, can lead to loss of creativity and critical thinking skills.
The environmental impact of GenAI use is significant: great amounts of water and energy are necessary for running the models (i.e., responding to prompts) and training them. Although this is a personal consideration, you could take this into account when you are considering using GenAI tools.
Five principles for working with AI
For effective and responsible GenAI use, keep the following five principles in mind:
Prompt: Effective GenAI use starts with an effective prompt. Make your prompt specific, using (relevant elements from) the format: clear task-description + persona + context + format + tone + exemplars
Proof: Always check the output for inaccuracies and biases. Critically evaluate the quality of the output, using your own critical thinking skills.
Privacy (and other considerations): There are many unknowns about how GenAI tools use your input. Never feed personal details or confidential information to GenAI tools. Also consider the environmental impact of GenAI before using it.
Presentation: The output from GenAI tools may include grammatically correct sentences, but it is often generic and flavourless. Don’t forget about your own voice when you are writing a text.
Property: In the end, always remember that you are responsible for your work, also when you use GenAI.
Adapted from “Summary Part 1: Effective and Responsible GenAI Use”. Walma, L., & Looij, M. AI for Literature Review is licensed CC-BY