PLOTKA — WAW // ZONE. Not a game, not quite a simulation.

/// I can provide pictures if you want; feel free to edit or remove the text as you see fit.

AN EXPERIMENT IN SIMULATING SOCIAL BEHAVIOUR WITH A LOCAL AI MODEL

Can a small language model simulate the behaviour of thousands of people making independent decisions in changing conditions? Could this eventually be used in strategy games, war games, social simulations or systems analysing complex scenarios?

These questions led to PLOTKA, an experimental simulation in which information spreads through a fictional Warsaw on the brink of an epidemic. Each resident can hear a rumour, believe it, ignore it, check it or pass it on. Every time it is passed on, its meaning may change.

It is a kind of digital telephone game — or rather the Polish “tomato” word game — in which AI affects both the circulation of information and the behaviour of the whole community.

THE CITY AND ITS RESIDENTS

The current simulation covers 70–5,000 residents across Warsaw’s 18 districts. I plan to keep increasing the population.

Each of the 5,000 characters has a social role: a taxi driver, pensioner, nurse, dealer in stolen goods, influencer and so on. Each starts with one of 12 personality types, initially generating a rumour with a set of individual traits.

The user enters any rumour and watches it take on a life of its own. Who will pass it on? Who will believe it? How will its content change? Will it cause panic, make people stockpile supplies or go unnoticed?

Each resident can retain their own version of the information. This makes it possible to trace its path through the city, successive modifications and social consequences.

AI THAT MAKES DECISIONS INSTEAD OF WRITING DIALOGUE

At the heart of the experiment is basal-1.5-mini, running locally on an ordinary graphics card.

The model does not generate free-form dialogue for characters. It acts as a classifier, assessing possible behaviours from a situation described in Polish.

Depending on the simulation stage, it determines:
- The nature of the information: whether it is a rumour, what it concerns, its tone and threat level.
- The sender’s behaviour: whether they will pass it on and how they will present it.
- The recipient’s reaction: whether they believe it, decide to check it or ignore it.
- The resident’s action: whether their beliefs lead them to stockpile, stay home, flee or do nothing.

Decisions are not entirely deterministic. For example, the model might estimate a 37% probability of passing on a rumour. The simulation’s random mechanism determines the final outcome. Even similar residents in similar circumstances can therefore behave differently.

BASAL INTERPRETS CONTEXT AND ESTIMATES BEHAVIOUR PROBABILITIES

The program handles the world’s mechanics: residents meeting, phone calls, social media, movement between districts and the epidemic’s progression.

World parameters can be changed, including access to water, electricity and transport, the type of threat, and the conditions under which information spreads. This allows experiments with different scenarios and observations of how changing one factor affects the whole city.

AN EXPERIMENT WITHOUT A PREDETERMINED OUTCOME

PLOTKA was not created as a finished game or commercial product. It is primarily a way to explore how language models can work with conventional simulation systems.

I am interested in the process as much as the final result: designing mechanisms, testing architectures, observing unexpected behaviour and finding the limits of small AI models.

Perhaps someone will eventually use this kind of model in strategy games, conflict simulations, social behaviour modelling or tools supporting game production. For now, the goal is not a finished product but discovering approaches that may help build future solutions.