Part of what makes us human is the unique way we think and solve problems. But using large language models like ChatGPT could destroy this uniqueness and lead people to think and communicate in the same way, according to a team of scientists and psychologists who have put together a new opinion paper.
“The richness of how different people write, argue and think is one of the most important resources for human understanding,” Zhivar Sourati, a computer scientist at the University of Southern California and first author of the paper, told CNET.
If these distinctions are made by the same LLMs, Zhivar argues, the result is fixed discourses and thoughts for all users.
“Diversity of language, perspective, and thought isn’t just a cultural beauty; it’s functional. It’s what drives creativity, innovation, and collaborative problem-solving,” Sourati said.
The paper, published Wednesday in the journal Trends in Cognitive Sciences, examines how hundreds of millions of people around the world use a few chatbots and what that means for our personalities.
Thinking inside the box
Pew Research found that one-third of all Americans used ChatGPT last year, doubling the figure by 2023. And chatbot use is more common among young people: Two-thirds say they use chatbots, and almost a third use them every day.
Businesses are also getting into all the artificial intelligence. Stanford found that 78% of organizations reported using AI by 2024, up from 55% by 2023.
So we use AI a lot. But the danger is that we can lose diversity in the ways we think. The team points out that the writing produced by the LLM is slightly different than what people come up with on their own.
Sourati and the paper’s authors argue that LLMs pose a new threat to diversity thinking.
“Previous technologies have shaped understanding: the Internet has accelerated the spread of dominant cultural practices, GPS has eliminated spatial thinking. But those tools primarily facilitate preservation and retrieval,” said Sourati. “LLMs generate thinking and expression on their own, on your behalf.”
LLMs can give users “a finished way of thinking about something,” Sourati said. “And when a few similar systems do that to hundreds of millions of people at the same time, the homogenizing power is unmatched by anything previous technology has produced.”
According to the authors of this paper, part of the reason that LLMs may push the assumption is that the data is used to train them. Sourati says LLMs are trained to focus on statistical principles in their training data, which can over-represent dominant languages and ideas, leading to results that are skewed to a small slice of human experience.
Why diversity thinking is important
There is a good reason why the authors caution against this practice. Homogeneous thinking reduces pluralism, the idea that many ideas are good for society as a whole.
“This importance of diversity is based on the long-standing principle that sound judgment requires exposure to diverse perspectives,” the authors wrote in the paper. “Unchecked, this homogenization risks undermining the attitudes that drive collective intelligence and adaptability.”
So we use different ways of thinking to find more solutions to the problem. If we lose the ability to think and communicate differently, it can affect how we adapt to new situations.
“The concern is not just that LLMs shape the way people write or speak, but that they redefine what is important as honest speech, positive opinion, or good thinking,” Sourati said in the announcement.
The authors also say that this trend affects even people who do not use chatbots.
“If a lot of people around me think and talk a certain way, and I do things differently, I can feel pressure to conform to them, because it seems to be a reliable or socially acceptable way to express my views,” said Sourati.
Owen Muir, an interventional psychiatrist, agrees with the paper’s views, saying that “a ‘common language’ is embedded in human communication, even when machines are not in the room.”
