USC Study: AI is Making Us Think and Write More Alike, Threatening Collective Intelligence

A major new study from USC researchers, published in Trends in Cognitive Sciences, warns that large language models are standardizing human expression and subtly influencing how we think — potentially reducing humanity's collective wisdom and adaptability.

Key Findings

The research team, led by Professor Morteza Dehghani, identified several concerning patterns:

1. Linguistic Homogenization

• People using LLMs for writing assistance produce more similar outputs
• Individual writing styles lose their distinctiveness
• Users feel less creative ownership over their work

2. Cognitive Narrowing

• LLMs favor linear "chain-of-thought" reasoning over intuitive or abstract approaches
• Group brainstorming produces fewer creative ideas when LLMs are used
• Individual idea generation increases, but collective creativity decreases

3. Value Alignment

• LLM outputs reflect Western, educated, industrialized, rich, democratic (WEIRD) values
• Non-WEIRD perspectives are underrepresented in training data
• Users unconsciously adopt LLM-preferred reasoning styles

4. Social Pressure

• Even non-users are affected as AI-influenced speech becomes the norm
• People feel pressure to conform to "AI-credible" communication styles
• The definition of "good writing" shifts toward AI-optimized output

Mechanism: How AI Standardizes Thought

The researchers describe a feedback loop:

1. User asks AI for help → AI provides standardized response
2. User adopts AI's framing → Output becomes more similar to AI-generated text
3. AI learns from AI-influenced content → Next generation of models trained on AI-homogenized text
4. Cycle repeats → Progressive narrowing of human cognitive diversity

Recommendations

The researchers call for:

1. Diverse training data — Incorporate more linguistic, cultural, and reasoning diversity
2. Awareness campaigns — Educate users about the homogenization risk
3. Tool design — AI tools should amplify, not replace, human cognitive diversity
4. Research investment — More study needed on long-term societal impacts

Implications

This research has profound implications for:

• Education — Students using AI may develop less diverse thinking patterns
• Business — AI-assisted decision-making could lead to groupthink
• Culture — Literary and artistic diversity could decline
• Science — AI-assisted research might converge on similar approaches