A new study from USC researchers warns that widespread AI chatbot use may be narrowing how humans think and express themselves. The research, published in Trends in Cognitive Sciences, suggests that billions of people using the same handful of AI tools are experiencing a subtle but significant reduction in cognitive diversity.

Led by Professor Morteza Dehghani, the team found that large language models tend to standardize writing styles, reasoning patterns, and even what counts as credible speech. When people use AI to polish their writing, for example, the output loses stylistic individuality and users feel less creative ownership. The researchers note that while individuals may generate more ideas with AI assistance, groups actually produce fewer and less creative ideas when relying on these tools.

The study highlights that LLM outputs reflect Western, educated, industrialized, rich, and democratic perspectives due to training data biases. Beyond language, interacting with biased AI models can shift users' opinions toward the model's viewpoint. The researchers call for AI developers to intentionally incorporate real-world diversity into training sets, emphasizing that this diversity should reflect genuine human variation rather than random variation.