Об идеологических предвзятостях генеративного ИИ: Российско-украинский конфликт в репрезентации ChatGPT

Российская школа связей с общественностью. 2026. № 40. С. 171–191.

Baysha O., Trofimov V.

A growing number of scholars are warning about the dangers of the reproduction by generative AI of socio-political and ideological biases absorbed by models from the texts on which they were trained. If a given model was trained on Western media texts, it may generate narratives that reproduce West centric views of world events. This will manifest itself in the reproduction of definitions of global problems normalized in Western hegemonic discourses. To identify how AI reproduces ideological biases towards Russia, and how they can be detected in generated texts, this article examines an analytical report generated by ChatGPT, a chatbot developed by the US company OpenAI. The topic was the Ukrainian-Russian conflict, which began with the (Euro)Maidan – a protest movement in Ukraine (2013–2014) that led to the overthrow of the government. Text analysis drew on the discourse theories of Ernesto Laclau and Chantal Mouffe, as well as Norman Fairclough. Its results showed that ChatGPT, despite its promise of neutrality, generated a black-and-white narrative, which missed all the important nuances that could help its users understand the situation. The model replicated a negative image of Russia that had been normalized in Western political discourse long before the advent of the Generative AI in our everyday lives. This confirmed the concerns of other researchers that AI-generated texts could exhibit ideological bias and reproduce West-centrism, while doing so with a claim to authority and neutrality.

Keywords: Euromaidan, Евромайдан, российско-украинский конфликт, ChatGPT, ChatGPT, генеративный ИИ, большие языковые модели, generative AI, Large language models (LLM), AI biases, West-centrism, Russia-Ukraine conflict, предвзятости ИИ, западноцентризм