Synonymix: Unified Group Personas for Generative Simulations
This work addresses the need for group-level simulations in AI and social science, offering a novel approach for sensemaking and synthetic persona generation, though it is incremental in extending existing simulation scales.
The paper tackled the problem of enabling meso-level simulation in generative agent simulations by proposing Synonymix, a pipeline that constructs a unified group representation from individual personas, resulting in preserved behavioral signals with statistical significance (p<0.001, r=0.59) and privacy guarantees (max source contribution <13%).
Generative agent simulations operate at two scales: individual personas for character interaction, and population models for collective behavior analysis and intervention testing. We propose a third scale: meso-level simulation - interaction with group-level representations that retain grounding in rich individual experience. To enable this, we present Synonymix, a pipeline that constructs a "unigraph" from multiple life story personas via graph-based abstraction and merging, producing a queryable collective representation that can be explored for sensemaking or sampled for synthetic persona generation. Evaluating synthetic agents on General Social Survey items, we demonstrate behavioral signal preservation beyond demographic baselines (p<0.001, r=0.59) with demonstrable privacy guarantee (max source contribution <13%). We invite discussion on interaction modalities enabled by meso-level simulations, and whether "high-fidelity" personas can ever capture the texture of lived experience.