• 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

Many of society's most pressing challenges—from pandemic response to supply chain disruptions to climate adaptation—emerge from the collective behavior of millions of individuals making decisions over time. Understanding these complex systems requires seeing how individual choices combine to create outcomes that no one person intended.

Meet Maya, a restaurant owner in Brooklyn. Every day during the pandemic, she faced difficult decisions: Should she raise prices as supply costs increased? Reduce staff hours? Pivot to takeout-only? Her decisions weren't made in isolation—they were influenced by her customers' willingness to dine indoors, staff availability, supply chain disruptions, and government policies.

Current AI research  has made remarkable progress creating "digital humans" —machines that mimic Maya's reasoning and decision-making—but has largely overlooked the critical next step: understanding how millions of individuals like her combine to form "digital societies." This is where Large Population Models (LPMs) come in—a new computational approach that simulates entire populations with their complex interactions and emergent behaviors.

Imagine a digital microscope revealing an entire city—8.4 million synthetic New Yorkers, including thousands like Maya, living their daily lives in a computational world. In this virtual society, each person makes decisions based on their unique circumstances: a nurse weighs the risks of commuting on crowded subways, Maya adjusts prices as supply costs rise, families decide whether a $500 stimulus check means they can afford to stay home during a pandemic surge. As these millions of individual choices ripple through networks of interactions, patterns emerge that no single decision-maker could foresee. This living laboratory of human behavior is the vision behind Large Population Models (LPMs).

Copyright

Media Lab Copyright

Research: Three Fundamental Breakthroughs

Building this digital New York required solving three fundamental challenges:

1. The Scale vs. Detail Dilemma: In our digital city, we need to simulate Maya as a distinct individual with her unique circumstances while simultaneously modeling millions of other New Yorkers. Traditional approaches force an impossible choice—like trying to film a stadium while capturing everyone's facial expressions. You could model complex behaviors for a few hundred people OR simple movements for millions—but not both. 

Our breakthrough enables both through two key innovations: First, our domain-specific language can compute billions of interactions simultaneously across customer, supply chain and community networks through optimized tensor operations. Second, our agent architectures capture realistic behavioral patterns across demographic groups, allowing us to simulate unique decisions for millions of individuals without separately modeling each person. This approach simulates all 8.4 million New Yorkers on a single GPU—600× faster than previously possible—without sacrificing the rich detail of each person's unique situation.

2.The Puzzle Piece Challenge: City officials need to understand how Maya and millions like her would respond to new restrictions or relief programs. They have restaurant reservation data, rent delinquency rates, outbreak signal and mobility patterns—but these pieces don't naturally fit together.

By making simulations differentiable end-to-end (similar to how neural networks learn), we transform months of computation into minutes. This allows direct gradient-based learning from heterogeneous data sources—hospital records, mobility patterns, economic indicators—without surrogate models, enabling rapid calibration and real-time analysis of complex scenarios, a 3000× speedup over traditional methods. 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 interventions might help.

3. The Simulation vs. Reality Gap:  As conditions change, our digital New York needs updates from the real one. Traditional simulations remain disconnected from the real world they aim to understand, relying on passive or anonymized data. 

We bridge this gap by transforming personal devices—like Maya's phone—from passive data collectors into secure simulation agents in a decentralized network. Through secure multi-party computation, these devices can contribute to population-scale simulation and analysis while keeping individual data private. Instead of generic guidelines, Maya receives personalized insights about her restaurant operations based on current local conditions. This technical breakthrough transforms simulations from passive analysis tools into real-time decision engines that shape reality.

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, 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 being 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, facilitate peer-2-peer energy grids in small Indian towns and enable 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 passively and used to protect global nations.

From pandemic response to climate adaptation to urban planning, LPMs provide a powerful new tool for understanding how millions of individual decisions combine to shape our world—and how we can design better solutions for our most pressing 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 enteprise. For thoughts and questions, please reach out to Ayush Chopra at [ayushc] [at] [mit.edu]. We look forward to hearing from you!