Imagine you are working on an urban design project to improve social quality in a particular neighbourhood, you will want to know what people think about your proposal. Collecting data from every resident most of the time has a lot of limitations, such as cost, data privacy concerns, and people’s willingness to be involved. Also your project can be localized in a different city or country and getting in contact with the population could be even harder. That’s when the use of synthetic population comes in. Synthetic population can be used for analyzing, predicting outcomes and patterns, in short as a tool to validate ideas before execution
Design Goals
Problem Statement
How can we use AI to generate participatory simulation during the design process
Objective
Create an AI agent based n8n workflow to generate synthetic populations as tools for feedback on urban design.
Pipeline Part I
To initiate the pipeline, we first generated a synthetic population, testing our approach on Barcelona. We began by preprocessing demographic data to create a proportional sample of 50 individuals representing the city. This sample served as the foundation for our static data. Additionally, insights from interviews conducted in another course informed the development of our synthetic population. These community members formed a focus group to provide feedback on the design phase, which takes place in the second part of the project.
Our final data source involved scraping Reddit posts from the Barcelona subreddit, applying filters related to sound, noise, and Gràcia—where the urban design project is being implemented. The next step was utilizing our AI agent, powered by Google Gemini, to generate detailed personas that aligned with the age, gender, and ethnicity distributions derived from our demographic data.
Pipeline Part II
In the second phase, we employed a second AI agent to interact with our SP. The process began with the AI agent asking each of the 50 members to select a color they would like to see represented in the project. Next, the system called an n8n model to generate an urban design art intervention, which the SP then evaluated and provided feedback on. Finally, the AI agent leveraged ComfyUI to refine and regenerate the urban art installation based on the SP’s input, ensuring the design aligned with their preferences.
Outcomes
In one of our iterations, the first member of the SP generated was a Catalan man named Jordi—one of the most common Catalan names. He disliked the initial design concept, as it did not align with the cultural elements he valued. Instead, he proposed an alternative intervention that better reflected his perspective.
Video Tutorial
Key Terms
Agent-Based Model: A computational model used to simulate interactions of individuals within an environment.
Synthetic population: An artificially created model that samples statistically a real population allowing the execution of agent-based microsimulation.
Statistical sampling: A method based on probability and statistics used to select a representative subset of individuals or households from a larger population.
