Ep73 “The Dangers of Group Think on Decision Making” with Adi Sunderam

| Podcasts | February 19, 2026 | 2.91 Thousand views | 33:40

TL;DR

This episode explores how flawed decision-making stems not from mathematical errors in Bayesian updating, but from people restricting their 'model set'—the range of explanations they're willing to consider a priori. When groups socially exclude certain hypotheses, members invent increasingly convoluted interpretations to maintain their worldview rather than updating beliefs.

🧠 The Model Set Problem 2 insights

People exclude true explanations rather than updating incorrectly

Instead of failing at Bayesian math, decision-makers exclude the true model from their consideration set, forcing them to assign higher probability to increasingly unlikely alternatives as confirming data arrives.

Dogmatic priors make beliefs immune to evidence

When individuals commit to the position that no data could ever change their view, they exhibit a 'dogmatic prior' that renders rational debate and belief revision impossible.

👥 Group Think and Social Enforcement 2 insights

Communities enforce forbidden hypotheses through social penalties

Groups impose severe costs for considering certain explanations, creating echo chambers where members collaboratively generate convoluted alternative theories rather than facing excluded truths.

Intelligence enables sophisticated rationalization

Smart people are particularly adept at constructing creative explanations that fit data within restricted model sets, making them especially susceptible to this form of collective intellectual waste.

📊 Evidence From Markets and Politics 3 insights

Stock Twits shows persistent belief clustering

Analysis reveals Tesla bulls initially reacting to negative earnings announcements but quickly coalescing on post-hoc explanations like 'Elon Musk is a genius' to preserve their prior worldview.

Identical economic data gets filtered through opposing models

During the 2021-2022 inflation surge, doves and hawks interpreted the same CPI reports through conflicting frameworks, with doves continuously shifting explanations rather than updating beliefs.

Political polls show temporary volatility but long-term stickiness

Scandals cause short-term poll movements that revert to baselines within weeks, consistent with people using flexible interpretations to absorb contradictory data without changing core views.

Bottom Line

When confronting entrenched beliefs, ask 'What specific piece of evidence would change your mind?' before presenting facts to expose dogmatic priors and force consideration of previously excluded models.

More from My First Million

View all
Ep78 “What’s Wrong With Taxing Billionaires More?” with Joshua Rauh
35:42
My First Million My First Million

Ep78 “What’s Wrong With Taxing Billionaires More?” with Joshua Rauh

Finance professors from Stanford and Wharton argue that California's proposed 5% billionaire wealth tax would likely lose money for the state due to taxpayer flight, violates the implicit contracts that drive innovation, and ignores that the wealthy already pay the vast majority of income taxes while generating massive economic value.

6 days ago · 9 points
Our AI Future: From Abundance to Apocalypse
45:11
My First Million My First Million

Our AI Future: From Abundance to Apocalypse

Stanford economist Chad Jones explores AI's economic potential through two divergent scenarios: explosive growth driven by recursive self-improvement and full automation, versus continued 2% annual growth constrained by historical patterns and persistent human bottlenecks in production chains.

14 days ago · 7 points
Tekedra Mawakana, Co-CEO of Waymo: Building a Safer Way Home
49:24
My First Million My First Million

Tekedra Mawakana, Co-CEO of Waymo: Building a Safer Way Home

Waymo Co-CEO Tekedra Mawakana discusses her journey from Mississippi to leading autonomous vehicle technology, emphasizing that transformative innovation requires betting on terrifying opportunities while building safety-first cultures that aim for a 13x improvement over human drivers rather than settling for 'safe enough'.

20 days ago · 9 points