Study: LLMs may be standardizing human expression — and subtly nudging how we think

It has been reported that artificial-intelligence chatbots are doing more than tidy prose: researchers at USC Dornsife warn they may be quietly narrowing the range of how we write, speak and even reason. The claim comes in an opinion paper in Trends in Cognitive Sciences led by Morteza Dehghani and first author Zhivar Sourati, who argue that when billions lean on the same handful of large language models (LLMs), individual voices — and the cognitive diversity those voices reflect — can get smoothed into a single, familiar tone.

What the researchers found

The team points to multiple studies showing LLM outputs are less varied than human writing and skew toward the language, values and reasoning styles of WEIRD (Western, educated, industrialized, rich, democratic) populations. When people use LLMs to polish text or brainstorm, individual stylistic quirks fade. Groups using LLMs can generate fewer and less creative ideas than groups that pool their own, diverse thinking. It has been reported that interaction with biased LLMs can also pull users’ opinions toward the model’s framing — allegedly reshaping not just words but judgments.

Why this matters

Why care? Because cognitive diversity is the grease in the gears of creativity, problem-solving and social resilience. If everyone starts to sound and think alike, collective wisdom suffers. Think of it like streaming algorithms for music: they can surface hits, but they also risk flattening an ecosystem. The emotional sting here is real — people report feeling less ownership over creations they’ve LLM-assisted, and that loss of voice matters.

What the researchers recommend

Dehghani and colleagues urge AI developers to train models on broader, real-world diversity to preserve different reasoning styles and cultural perspectives — and to improve model reasoning. More than a technical tweak, it’s a design choice about what kind of society we want to shape. So here's the question: do we want convenient, homogenized clarity, or a messier chorus of human voices? Choices now will echo into how we think tomorrow.

Sources: usc.edu, Hacker News