Computational social science has traditionally relied on three kinds of modeling: descriptive, interpretative, and predictive. According to a new arXiv paper, however, none of these fully combines in-world human intervention with outcome forecasting.

The paper proposes a "Computational Multi-Agents Society Experiments" framework built on generative agents. This framework aims to fill that gap by letting researchers intervene within a simulated society and then forecast the consequences of those interventions.

The abstract is brief and does not provide implementation details or experimental results. Still, the proposal points toward a more integrated approach to social modeling, one that could make simulations more responsive to human actions.