The Impact of Artificial Intelligence on Human Expression
The increasing prevalence of artificial intelligence (AI) systems has sparked discussions about their potential to not only standardise human expression but also to homogenise culture and cognition. Researchers are examining how tools like ChatGPT can actively shape information, potentially leading to uniform patterns in language, creativity, and cultural outputs that may ultimately influence the behaviours, judgments, and decision-making processes of both content creators and consumers.
Since the introduction of OpenAI’s highly popular chatbot and its widespread adoption, various studies have been conducted to investigate the technology’s possible impact on human cognitive processes. These investigations have delved into the effects of frequent engagement with AI tools on creativity, concentration, judgement formation, and reasoning, as well as on the perception and interpretation of experiences. Some findings indicate that regular use of AI could blur individual distinctions in these areas, thereby diminishing cultural and cognitive diversity across societies.
Comparative Analysis of Human and AI-generated Narratives
Recent unpeer-reviewed analyses have compared stories written by individuals with similar texts produced by five large language models (LLMs). The results revealed that the content generated by these AI systems tends to exhibit narrative characteristics more commonly than those created entirely by humans. In a related study published earlier this year, researchers assessed the performance of 22 AI models against over 100 human participants in a series of creative challenges. The findings indicated that while AI outputs were marginally more original, the responses among algorithms were notably more homogeneous.
Emily Wenger, a computer scientist at Duke University and co-author of this study, emphasises that this unification raises existential questions about the individuality of society’s members. She poses a thought-provoking query: “If we all use AI to draft our emails or anything else, what will we become as a species? Our society has suffered when we silence or ignore dissenting voices,” as highlighted in her statements.
AI’s Influence on Academic Publishing
The phenomenon of homogenisation extends beyond controlled experiments. Last year, researchers from the School of Psychological and Cognitive Sciences at Peking University examined over 400,000 articles published on the Web of Science before and after the release of ChatGPT-3.5. They found that while the number of publications per author increased following the AI’s introduction, there was a significant rise in content and linguistic style similarities. The authors suggest that this content uniformity could persist for months, leading to what they term a “creative scar”: “Although generative AI can enhance creative performance, users do not genuinely acquire the ability to create and easily lose it once generative AI is no longer accessible,” they explain.
Loss of Individuality in AI-modified Texts
AI-modified content presents additional challenges that go beyond mere linguistic uniformity. A study conducted by Zhivar Sourati, a doctoral student at the University of Southern California, analysed over 880,000 texts published on local media, preprint platforms like arXiv, and Reddit. The findings revealed that using LLMs to correct grammar in human-generated posts dilutes indicators related to personality, moral values, and demographic characteristics of the authors. These conclusions support the notion that AI may contribute to a loss of cultural diversity, a premise echoed by other research.
A 2025 study explored the differences and similarities in texts where participants from India and the United States shared information about their cultural rituals, national heroes, and symbols. In both groups, some participants received assistance from an AI-based autocomplete tool. The results indicated that this assistance not only made the writing styles within each group highly similar but also resulted in resemblances between the two cultures. For instance, Indian participants produced texts that resembled American writing styles and omitted specific details about culturally significant elements such as the Diwali festival.
Cognitive Implications of Linguistic Homogeneity
Sourati contends that the linguistic similarities fostered by AI, coupled with the blurring of cultural traits, could easily extend to cognitive processes. He articulates his hypothesis by referencing George Orwell’s 1984: “The way we speak influences the way we think.” He argues that if AI is causing humans to express themselves in increasingly similar ways, it is likely that their reasoning will also converge.
Some studies provide evidence that could substantiate this hypothesis. For instance, research conducted prior to the 2024 US elections identified certain LLMs that exhibited a bias favouring the Democratic Party’s initial candidate, Joe Biden, over Donald Trump. The study concluded that after interacting with AI tools based on these language models, supporters of the current US president adjusted their views regarding the Republican politician.
Addressing the Risks of AI Homogenisation
Experts suggest that the reasons behind this phenomenon stem from both technical factors and inherent human psychology. From a technical standpoint, the limited diversity of data used in AI training, the reward systems that favour the prediction of likely patterns over divergent ideas, and human evaluation processes that often prioritise satisfactory responses over novelty contribute to a “modal collapse,” leading systems to produce excessively similar results.
From a psychological perspective, specialists argue that individuals may adapt their ideas, reasoning, and behaviours to align with AI responses, perceiving that these models reflect general knowledge or societal opinions. Consequently, these outcomes could serve as anchors that shape users’ actions and thoughts over the long term.
Experts highlighted the necessity for addressing this issue through the utilisation of more diverse training data, adjustments to learning processes (including reward mechanisms), and the incorporation of additional sociocultural references in language models. While researchers acknowledge that robust evidence of widespread loss of individuality, cultural diversity, or cognitive variation due to the extensive use of AI has yet to emerge, they caution that the phenomenon may gain societal significance as more individuals engage with this next-generation technology.
“It is not a total collapse. It is not as if everything suddenly appears identical. However, considering the extent of AI adoption, the homogenising effect of AI does have consequences,” concluded Alwin de Rooij, a creativity researcher at Tilburg University.