• Login
  • Register

Work for a Member company and need a Member Portal account? Register here with your company email address.

Project

Large Population Models

Copyright

Camera Culture - Media Lab

Ayush Chopra MIT Media Lab

Think about how a delayed shipment, a cancelled flight, or a glitchy app can disrupt your week. Now scale that up: a virus like COVID-19 cost the world trillions, a ship blocked in the Suez Canal snarled global supply chains, and a CrowdStrike update grounded aircraft and paralyzed businesses worldwide. How do we spot these cascading disruptions before they become crises?

These systemic failures emerge from millions of seemingly unrelated individual decisions that interact in ways no single entity can predict. Behind each data point is a person making choices—a farmer deciding when to ship eggs, a logistics manager rerouting deliveries, or an IT specialist scheduling updates. Each decision seems reasonable in isolation, but together they create unforeseen consequences that can ripple through our interconnected world.

Current AI research has made remarkable progress creating "digital humans" that mimic individual reasoning. Yet it falls short of understanding how millions of decisions collectively drive systemic risks. This is where Large Population Models (LPMs) deliver breakthrough capability—building 'digital societies' that simulate entire populations with their complex interactions and emergent behaviors.

Imagine a digital microscope revealing an entire city - 8.4 million New Yorkers living their daily lives in a computational world. Here, Maya, a Brooklyn restaurant owner, switches to a pancake-focused menu when egg prices spike. This seemingly small decision triggers cascading effects: customers modify dining habits, suppliers revise deliveries, and local price dynamics shift. This living laboratory of human behavior transforms simulations into dynamic systems that provide actionable insights to prevent crises before they emerge.

Copyright

Media Lab Copyright

Research: Three Fundamental Breakthroughs

LPMs address the core challenges of modeling complex systems with millions of interacting agents. To show how, consider Maya facing H5N1 bird flu—a disease spreading from poultry to humans, threatening both food security and public health. This disease emerges from a series of interconnected events. Wild birds migrate and carry the virus to farms. Farmers must then decide whether to cull their flocks to limit the spread or risk further losses. For Maya, Egg shortages raise her costs, and visits to the farmers’ market expose her to health risks. Here’s how LPMs help.

1. Modeling Millions of Individuals: We need to understand millions of people—farmers, supply chain workers, consumers—and their interactions across multiple networks. Traditional simulations forced an impossible tradeoff: either model realistic behaviors for a few hundred individuals OR track simplified movements for millions—but never both at once. It's like trying to simultaneously film an entire stadium while capturing each person's facial expressions.

Our breakthrough: LPMs can simulate 8.4 million individuals, each with their own behaviors, on a single GPU - 600 times faster than previously possible - without sacrificing the rich detail of each person's unique situation. First, we efficiently process billions of interactions simultaneously across customer, supply chain and community networks - in minutes instead of hours. Second, we learn behavioral patterns across individuals, allowing us to accurately capture unique decisions for millions of people while separately modeling only a few thousands - recreating a digital New York for $500. This helps see how Maya's decisions ripple out across the city - shaping consumer behavior and public health.

2.Learning from Multi-scale Data: Managing H5N1 requires combining varied data sources: migration patterns of wild birds, production losses on farms, logistics records, and consumer purchasing trends. Traditionally, these sources remained fragmented. Researchers built surrogate models—simplified approximations—to align simulations with real-world data, but these approximations sacrificed the detailed mechanics of how disease moves from birds to farms, or how farm losses disrupt supply chains. This led to imprecise predictions and incomplete solutions.

Our breakthrough: LPMs eliminate the need for simplified approximations by making our simulations differentiable—transforming months of computation into minutes. This allows simulations to learn directly from diverse real-world data sources—hospital records, mobility patterns, economic indicators—providing 2-20x better precision and  3000x faster calibration over traditional surrogate models. When Maya's restaurant sees fewer customers, our model rapidly determines whether this resulted from rising infections, new restrictions, or consumer confidence changes—and projects how specific interventions might help her business while improving public health.

3. Closing the Simulation vs. Reality Gap: Traditional simulations treat agents like Maya purely as synthetic entities that mimic real people. This creates a fundamental disconnect—the digital Maya can never truly reflect how the real Maya adapts to changing conditions, and insights from the simulation can't easily reach the real Maya when she needs them. By the time data is collected, cleaned, and fed into models, the real world has already moved on.  During COVID-19, contact tracing apps collected data on exposures and sent it to central servers for processing, only notifying users days later. These apps could only provide post-hoc notifications ("You were exposed 3 days ago"), when the damage was already done. This created a critical gap between simulation insights and real-world action, severely limiting adoption and effectiveness. A similar centralized approach for H5N1 would repeat this flaw, hindering efforts to protect health and food systems.

Our breakthrough:  LPMs can decentralize the simulation by transforming Maya’s personal device into active simulation nodes. Through advanced cryptographic techniques, we can use Maya’s real-time behavior and interactions to execute simulations locally without compromising her privacy. The simulations run on networks of phones, not central servers, reflecting people’s actions as they happen. This creates a powerful two-way connection: Maya's actual restaurant decisions help update our digital New York in real-time, while insights from millions of simulated scenarios provide her with personalized recommendations. With H5N1, this means a phone can warn, “Avoid the market tomorrow; risk is up,” using live data from farmers and shoppers. It’s instant, keeps data private, and helps prevent health and food problems before they grow, unlike COVID-19’s delays.

LPMs realize this vision by making fundamental advances in agent-based modeling, decentralized computation and machine learning. Our research has resulted in several publications at top-tier AI conferences and journals, and received multiple best-paper awards. Our work has received research awards from industry (e.g. JP Morgan, Adobe) and government (e.g. NSF).

Copyright

Media Lab Camera Culture

AgentTorch: Tools for Digital Societies

AgentTorch, our open-source platform, makes building and running massive LPMs accessible. It integrates GPU acceleration,  differentiable environments, large language model capabilities, and privacy-preserving protocols in a unified platform—allowing researchers to build, calibrate, and deploy sophisticated population models without specialized expertise. Think PyTorch, but for large-scale agent-based simulations. Find below a quick demo and a code-snippet. The AgentTorch platform is open-source at github.com/AgentTorch/AgentTorch

Copyright

Media Lab Camera Culture

Real-world Impact

AgentTorch LPMs are already making impact globally. They've been used to help immunize millions of people by optimizing vaccine distribution strategies, and to track billions of dollars in global supply chains, improving efficiency and reducing waste - across governments and enterprises.  

As your read this, AgentTorch LPMs  are helping the New Zealand crown stop a measles outbreak, facilitating peer-2-peer energy grids in small Indian towns and enabling global enterprises to reimagine their supply chains for a sustainable future.  Our long-term goal is to "re-invent the census": built entirely in simulation, captured actively and used to safeguard nations worldwide.

From pandemics to climate adaptation to urban planning, LPMs turn the chaos of millions of decisions into clear, actionable solutions—reshaping how we tackle our toughest challenges.

Copyright

Camera Culture Media Lab

Curious about LPMs: Learn More

We would love to collaborate with you in advancing fundamental research and deploying LPMs within your enterprise. For thoughts and questions, please reach out to Ayush Chopra at [ayushc] [at] [mit.edu]. We look forward to hearing from you!