Challenges of Using Generative Artificial Intelligence (GAI) in the "Research" of Personas from National History
Aleksandra Fostikov
The Institute of History Belgrade, Belgrade, Serbia, Senior Research Associate, aleksandra.fostikov@ iib.ac.rs.
Boban Petrovski
Faculty of Philosophy, Ss Cyriland Methodius Universityin Skopje, Full Professor, boban@fzf.ukim.edu.mk.
CHALLENGES OF USING GENERATIVE ARTIFICIAL INTELLIGENCE (GAI) IN THE “RESEARCH” OF PERSONAS FROM NATIONAL HISTORY
The recent breakthrough in the development of GAI, along with open access to some of its models through chatbot interfaces, has created numerous opportunity for research. However, despite a series of positive changes such as accelerating research, synthesizing data, speeding up code writing, or searching for new medications, it is essential to highlight the creation of misinformation as a particular problem, which is increasing precisely through the use of AI in research. One should also bear in mind that a GAI can offer vastly different perspectives on the world depending on its datasets, algorithm, and even the manifestation of biases or determinations of what is good or bad.
To ascertain whether GAI is capable of distinguishing between misinformation and valid data, as well as its ability to critically analyze the information at hand, we tested the three most popular freely available LLM models: ChatGPT 3.5, Perplexity, and Bing. To conduct this test, we chose two historically significant personas from the history of the medieval Balkans, as well as from the national histories of more contemporary states, around whose legacy debates still rage in the historiography of multiple nations, in line with ongoing political events. Regarding the answers that we got it must be pointed out that the use of GAI and its most popular models in historical “research” has proven to be quite challenging and thus far highly debatable. The concerns we have had, primarily regarding the essential question of their validity, considering that they are not trained with a specific set of analytical skills characteristic of every historian researcher, have emerged as justified.The presence of inaccurate data and potential disinformation emerges as a particularly serious problem, which, essentially, can create more negative consequences than benefits.
Our general overview is that GAIs are still in a so-called “gray zone” or not completely reliable, and therefore not entirely relevant. When using them, a considerable degree of caution is required or, rather, mandatory rechecking of the given responses. In addition, our analyses have shown that given their current insufficient development, the provided responses with the assertion of a stance may generate a one-sided and even imaginary truth, as well as inaccurate data presented as facts, which could lead to deep contradictions with the views of contemporary scientific achievements on a given historical issue. Therefore, it must be emphasized that in the era of GAI, especially when researching historical topics, with an emphasis on those that form the backbone of contemporary political relations, it is necessary to approach with utmost critical scrutiny during both the search and application of GAI. This same approach should be taught to future generations as well.
For the full article see links below:
https://works.hcommons.org/records/fz291-jnk05
https://www.researchgate.net/publication/391645761_Challenges_of_Using_…
Fostikov, Aleksandra and Petrovski, Boban (2024) Challenges of Using Generative Artificial Intelligence (GAI) in the “Research” of Personas from National History. In: ROLE of Historical Figures in History and Collective Memory. Institute of National History, Skopje, pp. 141-159. ISBN 978-608-4981-28-2
