AI Chatbots May Be Standardizing Human Expression, USC Researchers Warn

Artificial intelligence chatbots are increasingly influencing the way individuals communicate, potentially leading to a homogenization of language and thought, according to a recent opinion paper published in Trends in Cognitive Sciences by researchers from the University of Southern California (USC). The study suggests that this trend could diminish humanity's collective wisdom and adaptability.

The research team, led by Morteza Dehghani, a professor of psychology and computer science at USC Dornsife College, emphasizes the importance of incorporating real-world diversity into the training datasets of large language models (LLMs). They argue that such diversity is crucial not only for preserving human cognitive variety but also for enhancing the reasoning capabilities of AI chatbots.

Zhivar Sourati, the study's first author and a PhD student at USC Viterbi School of Engineering, noted that individual differences in writing, reasoning, and worldviews are being mediated by uniform LLMs, leading to standardized expressions and thoughts among users. This standardization, the researchers argue, could reduce creativity and problem-solving abilities within societies.

The paper also highlights that LLM outputs often reflect the language, values, and reasoning styles of Western, educated, industrialized, rich, and democratic societies, potentially overlooking the diverse perspectives present globally. To address these concerns, the researchers advocate for AI developers to intentionally incorporate a broader range of languages, perspectives, and reasoning styles into their models, ensuring that the diversity mirrors the global human experience.

The study underscores the need for a concerted effort to diversify AI models and adjust human interactions with them to safeguard cognitive diversity and the ideation potential of future generations.