Social Sandbox: The Reaction Simulator

Social Sandbox is a multi-agent simulator of digital and real-world reactions. When a user inputs an event, a synthetic society of hundreds of agents, modeled on authentic platform demographics, react. First movers post, reactions breed reactions, the story crosses from one platform to another, mutating in tone, the news picks it up as a second ignition, and reputations rise or collapse. Underneath is a mechanistic model of society. We wanted to move past the LLM prompting into a real overview of how society interacts online. We built this as we saw how easy it is to build bots which interact almost flawlessly with real people on social media, but nothing that repurposes that technology towards helping humans. Every agent holds reputational capital that is specific to a field or platform (drawn from Bourdieu's theory of cultural capital) and a signed trust score per audience segment. The LLMs only voice the agents who actually react, turning each agent's computed emotion, stance, and arousal into a real post in that platform's grammar. Reaction spread follows established models as the design is grounded in named social theory. Each agent has their own personality, their own risk tolerance, tech literacy, and beliefs. The buyers are individual creators, PR and comms teams, brand-risk and crisis units, political campaigns, corporate marketing, and agencies. "Test-drive your announcement before you post it" is a pitch every one of them understands, and synthetic-audience testing is a fast-growing category. While PR firms and marketing agencies may prefer the dashboard-style visualization, we also target the wider creator economy. We use a 3D rendered visualization of a world where creators can interact with the bots directly, questioning them on their opinions and allowing for a deeper understanding of different perspectives outside of their own echo chamber. The design gamifies the experience to ensure creators remain an engaging and memorable experience.