A new scientific paper raises a provocative question: are our favorite AI assistants making us all sound—and think—the same? According to an opinion piece published in *Trends in Cognitive Sciences*, the widespread reliance on large language models (LLMs) like ChatGPT may be eroding individual cognitive diversity.

"People have distinct ways of writing, reasoning, and seeing the world," said Zhivar Sourati, a computer scientist at the University of Southern California and the paper's lead author. "When everyone uses the same few LLMs as mediators, those unique styles and strategies become homogenized. We start producing standardized expressions and thoughts."

The scale of this mediation is vast. Pew Research data shows a third of Americans used ChatGPT in 2025, a figure that has doubled since 2023. Among teenagers, two-thirds report using chatbots. In the workplace, a Stanford study noted 78% of organizations deployed AI tools in 2024.

The researchers argue the problem is baked into the technology. LLMs are trained on datasets that often overrepresent dominant languages and viewpoints. Consequently, their outputs reflect a narrow slice of human experience, which they then propagate. The team's analysis found LLM-generated writing exhibits less variation than human-originated text.

This convergence matters, the authors contend, because it threatens pluralism—the societal benefit of multiple perspectives. "Sound judgment has long required exposure to varied thought," they write. A loss of cognitive diversity could impair our collective problem-solving and adaptability.

The influence extends beyond direct users. Sourati notes a social pressure effect: when many people adopt a standardized, AI-influenced mode of communication, those who think differently may feel compelled to conform to seem credible. The ultimate risk, the paper suggests, is that these models don't just change how we write, but subtly redefine what we consider credible, correct, or reasonable in the first place.

Source: CNET