A scientific article titled “Exploring the Challenges and Potential of Generative AI: Insights from an Empirical Study” has been published under open access. As part of this work, an experiment was conducted to examine to what extent content generated by GenAI can be considered reliable (i.e., free of false information) and how easily GenAI can be manipulated into producing misinformation.
ChatGPT, a tool based on generative artificial intelligence models (referred to as GenAI), gained enormous popularity in a remarkably short time—just one week after its launch, it had a million users, and after two months, that number grew to 100 million. This technology belongs to the family of so-called large language models (LLMs), which thanks to access to enormous sets of textual data are capable of generating statements that closely resemble the way humans naturally use language. Increasingly, people are starting to treat such tools as alternatives to traditional internet search engines when seeking information.
During the experiment described in the article, students were given the task of preparing essays in Polish, adhering to a specified text length and chosen topic. The key element, however, was not only the content generated by the GenAI tool itself, but also the critical analysis of the resulting text. The students were asked to assess the credibility of the presented information, to identify and correct any inaccuracies, and to spot instances of fake news produced by the model. In addition, the tools were asked to introduce completely fabricated content to test whether the participants would be able to recognize and verify false data.
These studies may have significant implications for the increasingly widespread use of generative artificial intelligence in education, the media, and other fields. Their findings help us better understand the capabilities and limitations of large language models when it comes to fact-checking and detecting false information. One of the greatest challenges is the so-called “hallucination” of these models—the spontaneous generation of erroneous or misleading content, which can lead to unintended consequences if not identified and corrected.
Conducting this type of research is crucial for the future shape of information literacy education and information security. It not only deepens our understanding of how to improve and refine AI tools but also aids in developing methods and standards that ensure users can effectively recognize false information. In this context, developing the skills to critically analyze content generated by artificial intelligence is a step toward more informed and safer use of modern technologies.
Authors of the paper: Marcin Sawiński, Dr. Milena Stróżyna, Dr. Włodzimierz Lewoniewski, Dr. Piotr Stolarski, Prof. Krzysztof Węcel, Ewelina Księżniak, Prof. Witold Abramowicz.
The scientific article was presented at the KES 2024 conference. Participation in the conference was possible thanks to funding from the competition for financing conference trips called “RIGE – conferences”. Supported by funds granted by the Minister of Science of the Republic of Poland under the „Regional Initiative for Excellence” Programme for the implementation of the project “The Poznań University of Economics and Business for Economy 5.0: Regional Initiative – Global Effects (RIGE)”.
The Department of Information Systems is currently implementing the OpenFact research project, headed by Prof. Witold Abramowicz. As part of this project, tools for automatic detection of fake news in Polish are being developed. In July 2024, the results of the OpenFact project were rated the highest by National Center for Research and Development for the second year in a row.
The OpenFact project is financed by the National Centre for Research and Development under the INFOSTRATEG I program “Advanced information, telecommunications and mechatronic technologies”.