Columbia Study Examines Large Language Models and Patterns in Human Decision-Making

🤝Verified by 2 independent AI models•100% agreement

Columbia Study Examines Large Language Models and Patterns in Human Decision-Making

2 sources2 min readTechnology3 blindspots

LeftCenterRight

Bias ratings by MBFC

✅ 6 verified✅ Verified🤝 100% model agreement

Framing Analysis0 balanced · 0 biased · 0 diverge from source label

CBS Newsdull, bland mulch

Washington Examinerexistential threats, usurp human control

A Columbia Business School study led by Sandra Matz analyzed over 110,000 real-world decisions from 1,000 participants using data from the Facebook myPersonality project. It compared unaided human choices to outputs from both generic and personalized AI agents. The research, published via CBS News on June 30, 2026, found that LLMs tend toward statistically average predictions.

CBS News🏢Washington Examiner👤

Synthesized by AI from 2 sources

View all 2 sources

[CBS News 🏢] AI could make people dull, one scientist fears. Here's why.[Washington Examiner 👤] The existential threats to artificial intelligence

How outlets covered this

Left

“AI could make people dull, one scientist fears. Here's why.”

Read at CBS News →

Center

No center sources covered this

Right

“The existential threats to artificial intelligence”

Read at Washington Examiner →

How different viewpoints frame this

Progressive

The study highlights how AI favoring statistical averages could erode the diversity of choices that fuel social movements and challenges to power.

“Concentrated tech development reinforces conformity over collective agency and equity.”

Conservative

AI systems compress variance and replace independent judgment with lowest-common-denominator predictability that echoes central planning.

“Undercuts individual responsibility and entrepreneurial risk-taking.”

Libertarian

Individuals voluntarily delegate reasoning to predictive models; any homogenization reflects revealed preference rather than coercion.

“Remedies must remain non-coercive and centered on personal responsibility.”

Devil's Advocate

All perspectives accept the causal claim without examining dataset limitations, historical tool precedents, or cases where reduced variance is beneficial.

“Statistical likelihood is treated as destiny rather than a tunable artifact.”

Source Credibility

CBS NewsLeft-Center✓ High

Washington ExaminerRight✓ Mostly Factual

Ratings by MBFC

Ask about this story

epistemic.news/story/columbia-study-examines-large-language-models-and-patterns-in-human-decision-mak