We are researchers and engineers advancing new scaling paradigms for multi-agent AI, combining machine learning and game theory to enable coordination at scale.
Scaling Multi-Agent Intelligence
Teaching Machines to Coordinate
Multiscalar Intelligence develops new algorithms and systems for multi-agent AI, enabling agents to learn, coordinate, and scale from individuals to open networks. Our work focuses on emerging scaling paradigms for coordination: training agents to reason strategically, enabling collective learning across coalitions, and building the foundations for large-scale coordination in open, adversarial environments.
Research
01
Strategic Capabilities
Training agents to reason, plan, and act under incentives and uncertainty. We develop methods that go beyond static reasoning, enabling agents to model others, anticipate outcomes, and operate in strategic environments.
02
Collective Learning
Extending learning from individuals to groups. We design algorithms that enable agents to learn jointly across coalitions, propagating signals through interactions, aligning behaviors, and improving coordination through shared experience.
03
Coordination at Scale
Building the foundations for large-scale, open multi-agent systems. This includes protocols, benchmarks, and infrastructure for coordination in adversarial environments, where agents must discover, interact, and cooperate without central control.
04
Emergent Capabilities in Agentic Networks
Studying how new capabilities arise from agent interactions that no single agent was trained to exhibit. We investigate when networked agents develop specialization, communication protocols, and problem solving strategies purely from multi-agent dynamics.
Team