Synthetic Populations
In one sentence: 10,000 cognitive agents with realistic personalities, emotions, habits, and social bonds — for testing products, campaigns, and policies before launch, at 1000$\times lower cost than real focus groups.
Theory sources: BM (G-profiles as personality, WM competition, cognitive biases), SM (population dynamics, social bonds, stigmergy), NM (trust hysteresis, information cascading), AGI_F (lifecycle, ontogeny)
Implementation: Same bmc-core engine. Environment = product/campaign stimuli. Oracle = engagement/conversion metric.
The Problem
Companies spend **$85B+ per year** on market research. The gold standard --- focus groups --- costs 50–100K per study, takes weeks, and samples 8–12 people. The results are biased by group dynamics (conformity, social desirability) and impossible to reproduce.
LLM-based synthetic users (GPT-4 with a persona prompt) are cheaper but fundamentally flawed: they have no internal state, no memory between sessions, no habits, no real emotional dynamics. Ask the same question twice, get a different answer.
The BMC Solution
BMC agents are cognitive decision-makers, not text generators:
| Capability | Real Focus Group | LLM Synthetic Agent | BMC Agent |
|---|---|---|---|
| Cost | $50–100K per study | $0.10/min/agent | $0.001/min/agent |
| Speed | Weeks–months | Minutes | Minutes |
| Sample size | 8–12 people | 100–1,000 | 10,000+ |
| Memory | Yes (human) | No (context window) | Yes ($\kappa -levels) |
| Habits | Yes | No | Yes ($\kappa=2 memes) |
| Emotions | Yes (uncontrolled) | Simulated text | Yes (8 Panksepp drives) |
| Cognitive biases | Yes (uncontrolled) | Partial (prompt) | Yes (6 mechanisms: H, I, W, G, A, R) |
| Reproducible | No | No (stochastic) | Yes (seeded) |
How It Works
- Calibrate population: Define G-profiles matching target demographics (age $\to WM capacity, income $\to FEAR sensitivity, brand history $\to $\kappa -levels)
- Introduce stimulus: Product concept, ad campaign, or policy enters S-layer of each agent
- Observe cognitive processing: I-filter acceptance/rejection, emotional response (SEEKING = interest, FEAR = price anxiety, DISGUST = rejection), WM competition with existing brand loyalty ($\kappa=2 )
- Measure outcomes: Adoption rate, segment-level reactions, churn drivers, price sensitivity curves
- Test interventions: “What if price +15%?” — re-run with changed parameter, compare distributions
Use Cases
Product Testing
Run 10,000 agents through a product launch scenario. Segment by G-profile:
- Early adopters (SEEKING$\uparrow , FEAR$\downarrow ): fast acceptance via SIT-gap (“novelty!”)
- Conservatives (FEAR$\uparrow , SEEKING$\downarrow ): rejection via I-filter (high novelty penalty)
- Brand loyalists (CARE$\uparrow , $\kappa=2 ): high switching cost, emotional barrier
- Price-sensitive (FEAR$\uparrow , loss aversion): FEAR $\times price delta $> SEEKING $\times value delta $\to rejection
Campaign Optimization
Test messaging variants against cognitive segments. BMC reveals why a message works or fails: did it trigger SEEKING (curiosity) or FEAR (urgency)? Did the I-filter reject it as incompatible with existing beliefs? Which $\kappa=2 habits does it need to overcome?
Churn Prediction
Not statistical correlation ($P_{churn} = \sigma(\mathbf{w} \cdot \mathbf{x}) ), but a mechanistic model: competitor meme enters WM, competes with brand meme ($\kappa=2 , CARE-bond). Trust hysteresis means a single bad experience can break what took months to build.
Policy Simulation
10,000 citizens with calibrated G-profiles react to a policy change. Model panic (FEAR cascades), resistance (I-filter rejection of new rules), migration (SEEKING activation when local conditions deteriorate), protest (RAGE when perceived fairness $< threshold).
Market
| Segment | Size (2024) | CAGR |
|---|---|---|
| Market research | $85B+ | 5% |
| Synthetic data | $1.5B+ | 35% |
| Customer analytics | $20B+ | 12% |
| GovTech / policy simulation | $30B+ | 15% |
What We’re Looking For
- Market research firm interested in cognitive-grade synthetic populations (vs statistical synthetic data)
- Enterprise partner with churn/conversion data for calibration and validation
- Government / think tank for policy simulation pilot
For the cognitive model of personality that underlies agent calibration, see Personality. For the theory of population dynamics, see Swarm Memetics.