Origins

In July of 2024, we won the UC Berkeley AI hackathon with Skyline: a city-simulation to model citizens of Los Angeles and use density patterns throughout their day to optimize transit.

We're now The Human Simulation Lab: a collective of students from the University of Waterloo building scalable optimizations to infrastructure through human modelling. Our website is live at humansimulation.ai, where we're going to continuously post about new experiments, research, and insights.

Skyline

Our first experiment, Skyline, was a city-scale simulation of Los Angeles using census data to model the behavior of hundreds of thousands of virtual citizens. These AI "people" mimicked real-life behaviors with daily routines, influencing each other's choices to create a realistic picture of city movement patterns.

To optimize transit, we used Reinforcement Learning and Markov Decision Processes, with a proximal policy optimizer to balance exploration and exploitation. Our model aimed to minimize skew between stops, reaching hotspots and efficiently routing through high-density areas.

Our demo video from the hackathon:

The results were an estimated 18% decrease in drivers on the road, correlating to a reduction of 13 Million Tons of CO2 annually. Our system could dynamically adjust as the city's population patterns evolved, ensuring long-term effectiveness of the optimizations and demonstrating the potential of AI-powered simulations in urban planning and environmental impact reduction.

What's Next

More case studies are published or in the works, including:

• Building a self-sustaining startup entirely with AI Agents, from research, to business development, to coding
• Transforming crowdsourced complaints in Toronto into actionable proposals with visualized urban planning changes

Reach out to any of us if you're interested in learning more. We're open to micro-grants and any new ideas!