INTRODUCTION

The pervasive influence of modern technologies on all aspects of life is gradually transforming society, promoting educational attainment and prioritizing knowledge as key factors in social differentiation and personal well-being (Andriievska, 2018, 3). Among these technologies is generative artificial intelligence (GAI), powered by large language models and the deep machine learning of artificial neural networks, which is unlocking new possibilities in science, business, culture, and society.

Key capabilities of GAI currently include the generation of written and spoken language, whose form and content demonstrate cognitive-like abilities, as well as the creation of graphic, photographic, and video content (Jovanović, Campbell, 2022). This article focuses solely on GAI’s ability to produce text.

At the outset, it is important to note that we do not view using tools to generate encyclopedia articles as a viable alternative to original author-created content (for more details, see the forthcoming article by the authors, “Online encyclopedias and generative artificial intelligence: challenges and opportunities (case of the Encyclopedia of Modern Ukraine)”, to be published in Science and Innovation). However, in many other domains, this function of artificial intelligence — popularized by virtual assistants such as ChatGPT (OpenAI), Gemini (Google LLC), Copilot (Microsoft Corporation), and others — has become the norm.

For example, in e-commerce, many companies have significantly reduced or even abandoned copywriters who were responsible for creating content for online stores — the task of describing products is now entrusted to GAI technologies, which perform this task automatically. According to some studies, despite their shortcomings, virtual assistants are capable, for many online stores, not only of replacing copywriters but even of improving the quality of product descriptions, which increases sales (Serbin, 2023).

We believe that if the texts of encyclopedia articles begin to be generated using artificial intelligence tools, this will not only trivialize the corresponding publications or online resources but also devalue the field as a whole, having a detrimental impact on the development of encyclopedism (the preservation, multiplication, and dissemination of reliable knowledge in a reference format). In the current context of a vast amount of information sources, the value of encyclopedias lies primarily in their authors, who possess high professional authority and not only the relevant knowledge but also experience, and many possess wisdom. At the same time, we consider the potential of artificial text generation at the level of other possible directions of its application, improving the processes of preparing encyclopedic literature and disseminating encyclopedic knowledge in society.

Therefore, the aim of this research is to demonstrate the quality of texts synthesized by artificial intelligence, which can serve as supplementary knowledge-based content alongside the main, author-created content of an encyclopedia article.

In recent times, numerous online encyclopedias worldwide have been developed as multifunctional (interactive) information resources. This implies that they serve not only as a platform of reference information but also as a didactic portal that can be beneficial in the educational process (Zhelezniak & Ishchenko 2021). It is crucial for Ukrainian specialists to adopt this encyclopedic practice, as the construction of effective information systems in education, taking into account global experience, the specifics, and realities of the state of domestic education, is an issue that deserves special attention in modern scientific theory and practice (Blyzniuk 2017, 69). Didactic literature (textbooks, manuals, etc.) typically combines a theoretical component with a practical one aimed at mastering the learning material. A practical component of information accompanying articles would be appropriate in any online encyclopedia, as it would undoubtedly increase its heuristic potential.

Our idea is that this (AI-generated) information panel can present key theses from encyclopedia articles (highlights of encyclopedia articles), contributing to a deeper understanding of the article content and better knowledge acquisition. Such a block of compressed information is particularly important in lengthy articles, as not every average encyclopedia user preffers longreads, while abbreviated versions of large articles in the form of the most important fragments would be useful. Thus, key theses in encyclopedia articles would be beneficial both in terms of material assimilation (e.g., in the educational process) and for quick familiarization with reference information (similar to an article summary or outline).

MATERIALS AND METHODS

The creation of an additional information block based on an encyclopedia article using artificial intelligence involves processing the article’s text via a virtual assistant (through API access to the encyclopedia’s website or database) and generating a result based on its content. To test the potential of generative AI for this task, we conducted a small experiment.

For this purpose, we used two AI assistants — ChatGPT and Gemini. This choice was motivated by the desire to compare their performance on the given task. Using their respective chat interfaces, we provided the following instruction: “Write the key theses of this article (no more than five theses):” and supplied the copied text of the article.

The material for the experiment was taken from the Encyclopedia of Modern Ukraine, specifically the articles “Bourgeois Nationalism”¹ [in Ukrainian] and “Nationalism”² [in Ukrainian]. These articles were selected because they are frequently accessed by internet users. The AI models were provided only with the article texts; details such as the list of recommended literature and the authors’ names and initials were omitted.

It is also worth noting that the article “Nationalism” consists of two parts: the first written by Volodymyr Kulyk and the second by Vasyl Lisovyi. For the purpose of this experiment, only the section authored by V. Kulyk was used.

Experiment 1

The results of the experiment 1

Key theses from the article “Bourgeois Nationalism” (by Gemini): [in Ukrainian]

Key theses from the article “Bourgeois Nationalism” (by ChatGPT) [in Ukrainian]:

Experiment 2

The results of the experiment 2

Key theses from the article “Nationalism”  (by Gemini) [in Ukrainian]:

Key theses from the article “Nationalism” (by ChatGPT) [in Ukrainian]:

Discussion of Results

Generative AI, a product of advances in natural language processing, is capable not only of generating text — logically and semantically combining words into sentences, sentences into phrases, and larger linguistic units — but also of presenting facts, expressing arguments, manipulating data, and processing information in diverse ways. This demonstrates an analogue to the interplay of human language and thought. However, this capability is far from perfect. Numerous studies and anecdotal observations have shown that information obtained using ChatGPT or similar tools is not always accurate, and trust in such outputs remains understandably low at present (Jung et al., 2024).

The primary function of AI — generating text — is where it excels most, and it is successfully applied across various fields. We have already noted its widespread use in internet marketing (e-commerce). Notably, AI is also actively employed in journalism, not only for content creation but also for fact-checking, data processing, image generation, and translation. These applications reduce the human workload and enhance the efficiency of journalistic practices (Shi & Sun, 2024). For certain types of reports, AI is a true asset. For instance, some researchers analyzing sports articles based on statistical data have found that readers often cannot distinguish between human-written and AI-generated reports about football matches (Rossner et al., 2024).

However, in many other cases, AI-generated news reports raise concerns about the accuracy of information — a fundamental journalistic standard — as well as journalistic ethics in general. As Peña-Fernández et al. aptly note, “journalism cannot exist without journalists” (2023).

Encyclopedism shares some similarities with journalism. However, while the issue of AI in journalism is widely discussed in academic literature, there is a notable lack of studies on AI’s role in encyclopedic work, apart from those focusing on Wikipedia (e.g., Petroni et al., 2023). Lexicography, which has long dreamed of the advent of e-lexicography (de Schryver, 2003), is even more closely related to encyclopediography (encyclopedism) than journalism. Scholars hold differing views on the future of modern lexicography. Some argue that the field may no longer have a future, as AI now enables the building the dictionary entries without human involvement and allows for the generation of entire dictionaries online without large editorial teams or lengthy production times. Others, however, see AI as a source of new opportunities and prospects for advancing lexicography (de Schryver, 2003).

While initial tests of large language models (LLMs) for automated lexicographic tasks (expanding corpora, searching for phrases and word forms, creating definitions and examples) reveal numerous errors, including the occasional generation of hallucinations (i.e., the AI fabricating non-existent but plausible information), requiring additional oversight to identify and correct mistakes (McKean & Fitzgerald, 2024), other research highlights the LLM’s potential to rival human lexicographers. Consequently, the focus has shifted towards exploring effective implementation strategies. For instance, one study compared the effectiveness of ChatGPT-generated definitions with those from a traditional medical dictionary in assisting medical students with hard to undestand terminology. Results indicated that students provided with ChatGPT-generated definitions performed as well as those given traditional definitions, suggesting that LLMs can generate definitions as effectively as humans (Rees & Lew, 2023).

As aptly noted by some lexicographers, the best way to determine whether ChatGPT or another text generator is useful for a particular field is to test it directly (de Schryver, 2023). This recommendation aligns with findings from other researchers who argue that generative AI requires extensive research across diverse domains to assess its effectiveness and identify key implementation areas (Gupta et al., 2024). In this context, our study is among the first to explore the feasibility of using generative AI as a tool for content preparation in encyclopedic work.

We now turn to the results of our experiment. After reviewing the texts generated by artificial intelligence and comparing them to the content of encyclopedia articles, we observed the following: generative AI can effectively identify critical information within encyclopedic texts, structure it appropriately, and present it as coherent, logically constructed prose. Notably, the AI does not incorporate external information on the topic, relying exclusively on the provided articles. Based on the selected texts and the task at hand, we believe the AI performed competently. In some cases, ChatGPT produced better results, while in others, Gemini performed better. We did not observe a clear advantage of one chatbot over the other; both demonstrated comparable performance. This lack of significant difference between virtual assistants has been noted by many researchers studying AI applications across various domains (e.g., Rane et al., 2024).

The quality of generated text depends heavily on the quality of the source material: the more informative the original encyclopedia article, the more substantial the derived theses. In our experiment, the two articles demonstrated significant differences not only in length but also in content quality (informativeness). The article on bourgeois nationalism could have been more comprehensive, incorporating a greater amount of meaningful data and engaging facts to better illustrate this phenomenon. These shortcomings can be attributed to both editorial missteps in selecting the author and the limited editorial experience in handling such work at the time.

Regarding this issue (specifically, the process of editorial teams gaining experience), the researchers’ observations are particularly insightful: work on the initial volumes of encyclopedias tends to be more carefully planned compared to subsequent volumes. For example, the experience of compiling 19th-century European encyclopedias shows that the early volumes, starting with the initial letters, were prepared with greater care in terms of article composition (Loveland, 2019). The same researcher also points out a notable imbalance in the register of entries between the initial letters — where the greatest diversity of topics is concentrated — and the later ones. For instance, in the Encyclopédie nouvelle (1834–1842), the first volume is devoted exclusively to the letter “A”, while the final volume covers a range of letters from “S” to “Z” (Loveland, 2019, p. 130).

The experience of assembling a Ukrainian encyclopedia bears some similarities to that of Europe, as there are cases where multi-volume encyclopedic projects remain unfinished or are started with only one or two volumes published. It is not uncommon for the final volumes to be produced hastily. Given that the realization of multi-volume encyclopedic projects spans one or even several decades, various factors contribute to the disparities between the initial and final volumes, such as the number of articles, the authors involved, and often, their professional qualifications. The creation of the Encyclopedia of Modern Ukraine exemplifies this trend, with significantly more authors contributing to the initial volumes compared to the later ones. Among the reasons for this decline in author participation are the lack of remuneration for writing articles and the long waiting periods for publication (typically, one volume was published annually).

The ongoing digitalization of society, open access to information, the rise of Wikipedia, and other factors collectively reduce the motivation for experts to engage in traditional encyclopedic projects. This suggests that editorial boards of encyclopedias may begin turning to generative AI for preparing articles. Notably, there are currently no established guidelines for the responsible use of generative AI in the field of encyclopedias. As a result, this issue requires dedicated, in-depth scholarly discussion to develop a consensus among encyclopedists regarding the ethical implications of using AI to produce encyclopedic content instead of original authorial work. Essentially, this would shift the role of encyclopedias from providing knowledge to generating knowledge, potentially undermining the credibility of encyclopedias as reliable sources and impacting the "information ecosystem and knowledge infrastructure" (Bentzen 2023).

CONCLUSIONS

The research explored the ability of generative AI to generalize information from encyclopedia articles, identify key elements, and produce concise summaries.

Such text blocks could serve as additional knowledge-based content to enhance online encyclopedia articles. This would increase interest in encyclopedias, boost user engagement, and improve comprehension and retention of information. It would be particularly valuable for lengthy or hard to understand articles that might pose challenges for the average reader.

The findings suggest that generative AI is capable of performing these tasks. This conclusion is based on experiments conducted on two articles from the Encyclopedia of Modern Ukraine. The virtual assistants ChatGPT and Gemini identified key theses for these articles, which were positively evaluated by the authors of this study, who have experience in building the Encyclopedia of Modern Ukraine and other encyclopedic and lexicographic works.

This positive evaluation hints at the potential of generative AI to effectively writing encyclopedia articles. The challenge of recruiting authors for encyclopedia articles has always been a significant concern and is now more pressing than ever. However, it is important to note that such practices (writing encyclopedia articles by AI rather than humans) would be considered a misuse in encyclopedism, potentially diminishing the value of encyclopedias. As a result, there is a growing need for discourse among encyclopedists to establish ethical guidelines for applying generative AI in this field.

FOOTNOTES

https://esu.com.ua/article-38057

https://esu.com.ua/article-71080