Carrier: An Experimental Platform for Multimodal Mixed-Agent Social Simulation

Jiahao Yang, Will Smith

Keywords: Mixed-agent simulation; Collective behavior; Human–AI collaboration; Experimental methodology; Multimodal systems

TL;DR: We present Carrier, an open-source platform that enables controlled, reproducible mixed-agent simulations to study how humans and LLM agents interact in collective decision-making contexts.

Abstract: Large Language Models (LLMs) increasingly demonstrate stable role enactment, socio-emotional attunement, and coordinated behavior in multi-agent contexts, suggesting that they can function as actors within collective social systems rather than isolated text-generation tools. These capacities are highly relevant to policymaking, where decisions are collective, high-stakes, and seldom permit controlled field experimentation. Mixed-agent simulation—combining humans, LLM agents, and scripted actors—provides an in silico paradigm for studying deliberation, coordination, and information diffusion under controlled variation. Yet existing systems are often bespoke and weakly instrumented, limiting construct validity and comparability. We introduce Carrier (https://www.carrierlab.org), an open-source platform for multimodal mixed-agent simulation. Carrier structures interaction into modular episodes and defined roles, enabling systematic manipulation of agency, authority, and disclosure. With standardized instrumentation and process tracing, it supports mechanism-focused analysis of coordination, delegation, responsibility attribution, and collective error detection in simulated policy contexts.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 15