How do celebrity scholars influence knowledge?

A Reputation-Based Simulation

Scroll to continue

The Problem

In scientific communities, some scholars have disproportionate influence on research directions.

Scientists must choose which hypotheses to investigate

Celebrity scholars attract more attention and followers

Does this help or hinder convergence on truth?

The Model Setup

100
Scientists
1000
Hypotheses
1
True Hypothesis
5
Celebrities

Each scientist works on one hypothesis at a time. Only one hypothesis is actually true.

Evidence Generation

True Hypothesis
80% success rate

Working on truth usually yields positive results

False Hypothesis
20% success rate

False paths occasionally show misleading success

The Reputation System

Hypotheses build reputations based on accumulated evidence.

Reputation = Successes ÷ Total Tests

Each hypothesis tracks its success rate over time

Celebrity Weight: 5×

Celebrity test results count 5× more toward reputation

Rule-Out Threshold: <15%

Hypotheses with very poor track records get abandoned

When Scientists Switch Hypotheses...

80%
Exploit

Follow the highest-reputation hypothesis; prefer more tests as tie-breaker

20%
Explore

Try an untested or low-test hypothesis to discover new options

The Key Question

Does celebrity influence accelerate or impede convergence on truth?

Regular Group
All test results contribute equally to reputation
vs
Celebrity Group
Celebrity results have 5× weight on reputation
Scroll to explore the simulation

Interactive Simulation

Reputation-based model with evidence accumulation

The Setup

Scientists
Hypotheses
1 True Hypothesis
Celebrities

Evidence Generation

True H
80%
False H
20%

Reputation Settings

Celebrity Weight 5×
Rule-Out Threshold 15%
Min Tests to Rule Out 10
Regular Celebrity On Truth On False
Regular Celebrity

Live Statistics

0% On True Hypothesis
Active Hypotheses 50
Hypotheses Tested 0
Ruled Out 0
Time Step 0
Stable 90% (5 rounds) —

Top Hypotheses

Current Mode

Celebrity Group: Celebrity test results have 5× weight on hypothesis reputations.

Simulation Results

1,000 runs per scenario · Time to reach stable 90% (5 consecutive rounds)

Regular Group
93
average steps
median: 80
Celebrity Group
94
average steps
median: 79
~1% slower
Key Insight: With reputation-based decisions and 5× celebrity weight, celebrity influence has virtually no effect on convergence speed (~1%). The reputation system effectively neutralises celebrity bias over time.
Parameters: 100 scientists · 1,000 hypotheses · 5 celebrities · 5× celebrity weight

Model Development Journey

How different modelling choices affected the celebrity influence gap

1

Memoryless Model

Scientists follow celebrities with fixed probability, no evidence accumulation

62% slower with celebrities
2

Reputation System

Hypotheses accumulate evidence; scientists follow highest reputation

14% slower with celebrities
3

Stable Convergence

Require 90% for 5 consecutive rounds, not just first hit

5% slower with celebrities
4

Evidence-Based Tie-Breaking

When reputations are equal, prefer hypotheses with more tests

~1% slower with celebrities

What We Learned

Memory Matters

A memoryless system where scientists blindly follow celebrities shows 62% slower convergence. Adding evidence accumulation dramatically reduces this bias.

Reputation Corrects

When scientists track hypothesis success rates over time, celebrity mistakes get corrected. The 5× celebrity weight becomes less distorting.

Evidence Quantity Helps

Preferring well-tested hypotheses (as tie-breakers) further reduces noise. A hypothesis with 75% reputation from 50 tests beats 100% from 1 test.

Final Result

With proper epistemic practices, celebrity influence shrinks from 62% to ~1%. The system self-corrects over time.

Takeaway: Celebrity influence is not inherently harmful—it's the epistemic infrastructure that determines whether it distorts or gets corrected.