Are AI Chatbots Making Us All Think Alike? | CNET

The unique ways humans think and solve problems may be at risk of erosion, according to a fresh opinion paper co-authored by scientists and psychologists. The research suggests that frequent use of large language models (LLMs) like ChatGPT could be leading to a homogenization of thought and communication patterns.

“Individuals differ in how they write, reason, and view the world,” says Zhivar Sourati, a computer scientist at the University of Southern California and the paper’s first author, in a statement. “When these differences are mediated by the same LLMs, their distinct linguistic style, perspective and reasoning strategies become homogenized, producing standardized expressions and thoughts across users.”

The Rise of Standardized Thought

The paper, published Wednesday in the journal Trends in Cognitive Sciences, examines the implications of hundreds of millions of people globally relying on a limited number of chatbots. This isn’t a hypothetical concern; adoption rates are climbing rapidly. Pew Research found that 34% of U.S. Adults used ChatGPT in 2024, double the percentage from the previous year. Among teenagers, chatbot use is even more prevalent, with two-thirds reporting usage and nearly a third using them daily. Businesses are also increasingly integrating AI into their operations; a Stanford report showed that 78% of organizations reported using AI in 2024, up from 55% in 2023.

The core issue, according to the researchers, is that LLMs tend to generate writing that exhibits less variation than human-authored text. This isn’t necessarily a flaw in the technology itself, but a consequence of how these models are trained. LLMs are designed to identify and reproduce statistical patterns within their training data. Sourati explains that because this data “often overrepresent dominant languages and ideologies, their outputs often mirror a narrow and skewed slice of human experience.”

Why Cognitive Diversity Matters

The concern isn’t simply about stylistic uniformity. The authors emphasize the importance of “pluralism” – the idea that a diversity of perspectives is crucial for a healthy and adaptable society. As they write in the paper, “sound judgment requires exposure to varied thought.” A reduction in cognitive diversity could hinder collective intelligence and our ability to respond effectively to novel challenges.

Different approaches to thinking allow us to generate a wider range of solutions to complex problems. If individuals increasingly rely on LLMs to formulate their thoughts and express their ideas, the potential for innovative and unconventional thinking could be diminished. Sourati adds that the impact extends beyond those who directly use chatbots. “If a lot of people around me are thinking and speaking in a certain way, and I do things differently, I would feel a pressure to align with them, because it would seem like a more credible or socially acceptable way of expressing my ideas.”

How LLMs Shape Thought Processes

The mechanisms by which LLMs might influence human thought are subtle but potentially far-reaching. The authors suggest that LLMs don’t just shape *how* people write or speak, but also subtly redefine what is considered “credible speech,” “correct perspective,” or even “good reasoning.” This can create a feedback loop, where individuals internalize the patterns and biases embedded within the LLM’s output.

Sourati’s research, as detailed on his personal website zhpinkman.github.io, focuses on the reasoning capabilities of language models and how to improve their alignment with human cognition. He draws on insights from cognitive science, including analogical reasoning and the distinction between System 1 (intuitive) and System 2 (analytical) thinking, to better understand how LLMs can be made more human-like in their reasoning processes. He has collaborated with researchers at institutions including the Vrije Universiteit Amsterdam and Oracle on related projects.

Beyond the Technology: Sociolinguistic Implications

The paper also touches on the broader sociolinguistic dynamics at play. The increasing reliance on LLMs could alter the way humans communicate and interact with each other. If a significant portion of the population is using the same tools to generate text, it could lead to a decline in linguistic diversity and a homogenization of communication styles. This, in turn, could have implications for creativity, critical thinking, and the ability to engage in nuanced and productive dialogue.

Training Data and Bias

A key factor contributing to this potential homogenization is the composition of the training data used to build LLMs. These datasets often reflect existing societal biases and power structures, which can be inadvertently amplified by the models. The researchers emphasize the need for greater transparency and accountability in the development and deployment of LLMs, as well as ongoing efforts to mitigate bias in training data.

Zhivar Sourati’s work, as highlighted on his LinkedIn profile, also includes exploration of human-AI collaboration and the interaction between linguistic behavior and emerging technologies. He has collaborated with researchers from institutions like the University of Waterloo and the University of California, San Diego, on these topics.

What comes next involves continued research into the cognitive and social impacts of LLMs. The authors call for further investigation into the long-term effects of widespread chatbot use, as well as the development of strategies to promote cognitive diversity and critical thinking in an age of increasingly sophisticated AI. The field will also need to grapple with the ethical implications of these technologies and ensure that they are used in a way that benefits society as a whole.