Are LLMs Making Us Think Alike? AI & Diminishing Creativity

The way we write, and even reckon, may be subtly shifting as large language models (LLMs) become more integrated into daily life. A growing concern – initially a creeping suspicion – that interacting with tools like ChatGPT leads to more homogenized expression is now supported by research. A paper published Wednesday in the journal Trends in Cognitive Sciences warns that widespread LLM use risks flattening human thought and creativity, potentially impacting everything from individual writing styles to collaborative problem-solving.

Researchers at the University of Southern California analyzed over 130 studies spanning linguistics, computer science, and related fields to understand how these models affect cognitive diversity. Their findings suggest that, despite being trained on vast datasets of human-generated text, LLMs consistently produce outputs that are less varied than human thought itself. This isn’t necessarily a flaw in the technology, but a consequence of how these models operate.

How LLMs Prioritize Pattern Recognition

LLMs aren’t truly “thinking” when they generate text; they’re identifying and reproducing statistical regularities within their training data. Although the datasets are enormous, the models don’t process information with the nuance and contextual understanding that humans possess. Instead, they favor consistent patterns, leading some to describe them as sophisticated autocomplete systems. As Zhivar Sourati, a computer scientist at the University of Southern California and author of the study, explained in a statement, “As LLMs are trained to capture and reproduce statistical regularities in their training data, which often overrepresent dominant languages and ideologies, their outputs often mirror a narrow and skewed slice of human experience.”

This inherent bias isn’t always hidden. OpenAI explicitly acknowledges that ChatGPT is “skewed towards Western views.” Similarly, xAI’s chatbot, Grok, has demonstrably been influenced by the perspectives of its CEO, Elon Musk. These examples highlight the challenge of creating truly neutral AI systems, and the potential for these biases to seep into user interactions.

The Impact on Human Thought Processes

The implications extend beyond simply adopting a particular writing style. Research indicates that interacting with LLMs can actually shift the way people think, subtly aligning their perspectives with the information provided by the chatbot. This can manifest in something as simple as a user accepting suggested edits to their writing, removing stylistic quirks in the process. However, the effects can be more profound. LLMs often employ “chain-of-thought” reasoning, which favors linear thinking. This can be a limitation, as humans are capable of more abstract reasoning that involves non-obvious leaps in logic.

Interestingly, the study also found a counterintuitive effect on group dynamics. While individuals using LLMs to generate ideas often produce a higher volume of ideas (albeit with less originality), groups of people actually generate fewer ideas when using LLMs compared to traditional brainstorming sessions. The researchers suggest that LLMs can lock people into a particular way of thinking, reducing the diversity of perspectives that typically emerges from collaborative discussion and shared experiences. This is particularly concerning given the well-established link between cognitive diversity and improved outcomes for groups and organizations – a principle supported by research like that found in this study published in PLoS One.

A System Designed for Consensus, Not Diversity

LLMs are, encouraged to seek consensus rather than embrace diversity. This is a fundamental design characteristic, and one that isn’t likely to be easily corrected. The situation is further complicated by political factors. A 2024 executive order issued by the Trump administration effectively penalized companies that created AI models promoting diversity, signaling a potential policy-level reinforcement of this trend.

Exploring Cognitive Strategies in Human-AI Collaboration

The issue isn’t necessarily about avoiding LLMs altogether, but about understanding how to use them effectively without sacrificing cognitive diversity. A recent study published in the Journal of Creativity explored the cognitive strategies employed by individuals when using ChatGPT for creative tasks. Researchers found that students often rely heavily on AI-generated suggestions, repeating ideas rather than actively engaging in more complex cognitive processes like conceptual combination or inspiration. This highlights the necessitate for education on how to adopt more diverse and active cognitive strategies when collaborating with AI tools.

The Long-Term Cognitive Effects: Evidence from Retention Studies

The potential for long-term cognitive impacts is also a growing area of concern. A randomized controlled trial conducted in 2025, detailed in Social Sciences & Humanities Open, found that students who used ChatGPT as a study aid scored significantly lower on a knowledge retention test (57.5% correct) compared to those who used traditional learning methods (68.5% correct). This suggests that unrestricted ChatGPT use may impair long-term retention by reducing the cognitive effort required for durable memory formation – a phenomenon aligned with cognitive offloading theory and the “desirable difficulties” principle.

Redefining Cognitive Domains in the Age of LLMs

The rise of LLMs also necessitates a re-evaluation of how we define and assess cognitive abilities. A comprehensive analysis published in Med Res Arch examined the influence of AI on cognitive domains such as attention, executive function, language, memory, and social cognition. While LLMs offer potential as cognitive enhancers, the authors emphasize the importance of carefully considering their impact on long-term cognitive development.

Looking Ahead: Fostering Critical Engagement with LLMs

The challenge isn’t to eliminate LLMs, but to integrate them thoughtfully. Further research is needed to understand the nuanced effects of LLM interaction on cognitive processes, particularly over extended periods. Educational initiatives should focus on teaching users how to leverage these tools without sacrificing critical thinking skills and independent thought. The development of AI systems that actively promote cognitive diversity – perhaps by intentionally introducing challenging or unconventional perspectives – could also mitigate the risk of homogenization. A proactive and informed approach is essential to harness the benefits of LLMs while safeguarding the richness and complexity of human thought.