LLM Social Simulation Explained for Forecasting

Quick answer

LLM social simulation is a way to inspect plausible group reaction, not a way to claim certainty.

The value comes from structure: multiple actors, shared context, evolving state, visible assumptions, and a review loop that shows where confidence drops.

Definition

LLM social simulation models how different actors may interpret and react to the same event.

Instead of asking one model for a prediction, the simulation gives language-model agents roles, context, constraints, and a reason to respond. The system then watches how one reaction can change the next.

Simulation vs prompt

LLM social simulation is different from a single prediction prompt.

Structure

Single prompt: One model answers from one prompt.

Social simulation: Multiple bounded actors react inside shared context.

State

Single prompt: The answer is usually static.

Social simulation: Reactions can change the next round of behavior.

Review surface

Single prompt: The operator reviews prose quality.

Social simulation: The operator reviews actors, pressure, assumptions, and evidence gaps.

Why it matters

Many outcomes are social before they are numerical.

A launch becomes risky because the first audience frames it differently than the team expected.

A policy change becomes unpopular because perceived intent matters more than the official wording.

A rumor becomes dominant because a trusted or antagonistic actor amplifies it early.

A product message fractures because each stakeholder group reads the same claim through a different incentive.

MiroFish model

MiroFish turns social simulation into a reviewable workflow.

Layer 1

Represent actors

MiroFish identifies stakeholders, audiences, institutions, critics, amplifiers, and other actors that can shape the scenario.

Layer 2

Preserve context

The source packet becomes shared context: claims, constraints, evidence, motives, and points of tension that agents can react to.

Layer 3

Simulate reaction rounds

Agents respond over multiple rounds so one reaction can influence the next instead of collapsing the forecast into one static answer.

Layer 4

Report assumptions

The forecast report highlights likely paths, weak assumptions, confidence limits, and what evidence would change the result.

The operator should review pressure, not just prose.

Are the important stakeholder groups present in the graph?
Does the simulation explain why a reaction spreads or stalls?
Are confidence limits visible instead of hidden behind polished language?
Does the output identify evidence that would change the forecast?

Prompt template for LLM social simulation

Treat the uploaded brief as a live event. Simulate how different online actors interpret it, which frame spreads first, and what evidence would change the forecast.

FAQ

Questions about LLM social simulation

What is LLM social simulation?+

LLM social simulation uses language models as bounded actors inside a shared scenario so teams can inspect how reactions, narratives, and incentives may evolve over time.

Is LLM social simulation the same as asking ChatGPT to predict the future?+

No. A useful simulation represents multiple actors, context, relationships, and evolving state instead of returning one generic prediction paragraph.

What makes a social simulation useful?+

It is useful when it exposes missing stakeholders, escalation points, incentive conflicts, weak assumptions, and evidence that would change the forecast.

How does MiroFish use LLM social simulation?+

MiroFish turns a source packet into a graph of actors and pressures, runs multi-round agent reactions, and produces a forecast report that operators can review.

LLM Social Simulation Explained