Symmathesy: A Platform for Relational Intelligence and Collective Stewardship

We present Symmathesy, a web-based platform for hosting interactive collective behavior experiments involving humans and AI agents. The platform serves three interlocking goals. First, for the general public, it presents an opportunity to learn about collective behavior, collective intelligence, and current advances in agentic AI through both hands-on experiments and accessible explanations about the underlying research. Second, it allows social researchers to rapidly prototype multi-agent social environments where groups of human participants and artificial agents solve established or novel competitive or collaborative tasks. Third, for researchers in AI, it provides a tool for testing agents in interactive scenarios with or against human players, contributing to the ecological validity of AI evaluation in social contexts. On a meta-level, the platform was developed through an iterative co-creation process between researchers in computational social science and a large language model coding assistant, illustrating how AI tools can empower scientists to expand their toolkit beyond their domain expertise. We describe the architecture of the platform and its five public experimental paradigms, and report a load test of the deployed configuration: a single server instance with one virtual CPU serves about 100 simultaneous players of the most demanding real-time game, with 98.5–99% of state updates arriving on time, and the measurements locate the limit in real-time messaging rather than in the game logic. We discuss the platform's potential as a research, outreach, and participatory science tool for studying collective behavior, and the validation study that is its next step. We present Symmathesy as a working prototype and an open invitation – one whose full potential depends on the collaborative growth of a community of citizens, researchers, and AI developers.