The Use of Artificial Intelligence in Detecting Disinformation and Ensuring Cybersecurity

Yurii Kryvoruchko; Yevhen Solomin; Tetiana Mazurok; Valentyna Kozak; Volodymyr Shusta

doi:10.5935/jetia.v12i59.3530

Authors

Yurii Kryvoruchko Department of Documentation and Information Activities, Higher Education Institution "Academy of Recreational Technologies and Law", Lutsk, Ukraine.
Yevhen Solomin Department of Journalism, Faculty of Philology, Uzhhorod National University, Uzhhorod, Ukraine. https://orcid.org/0000-0001-6770-5505 (unauthenticated)
Tetiana Mazurok Department of Applied Mathematics and Informatics, the Educational and Scientific institute of Natural and Mathematical Sciences, Informatics and Management, South Ukrainian National Pedagogical University named after K. D. Ushynsky, Odesa, Ukraine. https://orcid.org/0000-0002-7829-4446 (unauthenticated)
Valentyna Kozak Department of Philology and Military Translation, Kyiv Institute of the National Guard of Ukraine, Kyiv, Ukraine. https://orcid.org/0009-0000-0121-4949 (unauthenticated)
Volodymyr Shusta Department of Optics, Faculty of Physics, Uzhhorod National University, Uzhhorod, Ukraine. https://orcid.org/0009-0004-7328-8268 (unauthenticated)

DOI:

https://doi.org/10.5935/jetia.v12i59.3530

Abstract

The article examines the use of artificial intelligence methods to detect disinformation in the context of modern cyber threats in order to determine the limits of their effectiveness and practical applicability for information protection. The study was conducted as an empirical quantitative comparative analysis combining secondary global statistical data on the spread and projected development of artificial intelligence in the world with the results of an analysis of open corpora of disinformation content containing text and multimodal materials, with a focus on changes in performance indicators in basic and modified information scenarios. Analysis of international analytical reports has shown a rapid and uneven growth in the scale of artificial intelligence implementation, which creates a complex information environment for the functioning of automated disinformation detection systems. The main results of the experimental analysis show that, under controlled conditions, algorithmic models demonstrate relatively high overall accuracy values, which on average range from 0.84 to 0.91 depending on the type of content. At the same time, it has been established that the completeness indicator is significantly less stable and, in conditions of modified or distorted input data, decreases to 0.70–0.73, which leads to a significant increase in the proportion of missed disinformation content. Integral quality indicators, in particular summary metrics, decrease by an average of 0.08–0.12 in such scenarios compared to baseline conditions. The quantitative results confirm the contextual dependence of model effectiveness and indicate their vulnerability to adaptive information influences in the context of the widespread use of artificial intelligence in the global digital environment. The conclusions justify that algorithmic methods can be an effective auxiliary tool for countering disinformation, but given the current scale and pace of AI implementation, they are not capable of performing an autonomous protective function without taking into account the dynamics of information threats, regular data updates, and combination with expert control mechanisms.