- Researchers warn that generative AI may be homogenizing writing, thinking, creativity, and cultural values as more people use large language models.
- One study likens this phenomenon to “McDonaldization,” where society becomes more uniform due to the efficiency and predictability of technology.
- An experiment comparing 22 large language models with 102 people showed that AI produces unique ideas, but responses between models are more similar than those between humans.
- Analysis of short stories created by humans and 5 AI models found that AI works cluster around similar motifs, whereas human works are more diverse.
- A 2024 study discovered that AI helps users write more creative and engaging stories, except for the top tier of writers, but simultaneously makes the stories more alike.
- A 2025 study and a meta-analysis published in April concluded that AI particularly reduces diversity in the ideation stage, especially for complex or highly constrained tasks.
- Analysis of over 400,000 scientific papers showed that after ChatGPT’s launch in late 2022, the number of papers per author increased, but content and linguistic style became more similar.
- Research on local news, arXiv preprints, and Reddit noted a decrease in writing style diversity after ChatGPT appeared; AI grammar correction also erases traits of personality, moral values, and demographics.
- In a study with participants in India and the US, auto-suggestion features made expressions between the two groups more similar, with Indians using language closer to Americans.
- Some studies indicate that AI not only influences what users write but can also impact their political views and beliefs, with effects lasting weeks even when users are warned about AI bias.
- Researchers call the phenomenon where former AI users continue to provide similar ideas two months after stopping use “creative scarring.”
- Technical causes are attributed to AI prioritizing common patterns in training data, optimizing based on human feedback, and the risk of declining diversity when learning from AI-generated data.
- Proposed solutions include diversifying training data, applying Diversity-Aware Reinforcement Learning (DARLING), designing prompts that encourage multiple solutions, and providing more cultural context or the user’s unique writing style.
- Many scientists believe the current impact is small but will become significant when AI is used at scale, while others believe humans still tend to seek differentiation and form new subcultures to resist homogenization.
📌 Nature journal synthesizes multiple studies showing that generative AI can reduce diversity in language, ideas, creativity, and even influence how humans form opinions. Evidence from 22 AI models, 102 participants, and over 400,000 scientific papers all point to an increasing trend of homogenization. However, researchers suggest that the impact is currently limited and can be mitigated by improving models, diversifying data, and encouraging users to maintain their unique identity when using AI.
