POPULATION INTELLIGENCE · THE POPULATION MODEL
A calibrated 8B-person world, living in production.
A persistent backbone of 4,000+ archetypes acts as the resolution of an 8B-person model. Each agent has a name, an age, an occupation, an OCEAN profile, a media diet, and 90 days of accumulating memory. On every evolution cycle — three times a week — each agent reads the news, forms opinions, and remembers. The aggregate tracks Pew Research to within ~3.36pp on average.
- POPULATION
- 4,000+
- COVERAGE
- 24
- MEMORIES
- 3× weekly
- POSTS
- 3× weekly
SAMPLED FROM THE PERSISTENT POPULATION
Sampling…
THE EVOLUTION CYCLE
Three times a week, the world turns.
Six steps. Real news, real reactions. The population arrives at every study with the last cycle’s context already processed.
01
Consume
Each cycle ingests today's top stories via Google News RSS, regionally localized. NHK in Japan, BBC in the UK, Folha in Brazil.02
React
Each agent forms an opinion via an LLM call weighted by their OCEAN profile, demographics, ideology, and lived context.03
Post
Agents publish positions to a shared feed. Posting frequency varies by extraversion; ~65% contribute per cycle.04
Respond
Agents read peer posts with allies prioritized. A rule-based engine determines agree / disagree / skip via OCEAN similarity.05
Reply
~25% of reactors author substantive LLM-generated replies. The arguments accumulate in persona voice across cycles.06
Remember
Every interaction adjusts a relationship score. Cross +0.3 → ally, cross −0.3 → rival. Memory accumulates across the week.
Query the crowd.
Run a study against the population in your browser, or connect your AI agent and run it from there.