A tool visualising future urban policy scenarios through generative AI and foresight methodology to encourage more experimental policymaking.
By Jorge, Dhrishya and Sophie
We formed our group based on our common interests of speculative future visualisation, better policymaking and civic engagement and the potential of emerging generative AI technologies. Personally, I used this microchallenge as an opportunity to work on my research project seeking to bring more innovative foresight tools in policymaking.
We wanted to introduce innovation into policymaking and civic engagement by creating a technology that would enable users to visualise and experience the future impact of policy interventions, creating avenues for innovative civic feedback mechanisms and better policymaking. The technology would be supported by AI, and would use a specific foresight methodology to envision possible future scenarios for a given picture, based on planned policy interventions.
1. OpenAI did not produce the expected visualisations of future scenarios. All images rendered by the custom bot were similar as the bot could not understand that modifications should be done to the original picture. To overcome this challenge, we used a stable diffusion AI model which gave much better images.
2. Coding the Crystall Ball automated tool was more challenging than expected. We ran into the problem of having too many tokens requested for the type of API access we had, so the future scenarios could not be generated. To be able to finalise the tool, we had to pay to upgrade our API access, which we decided not to do.
The technology needs to be more mature for the experience to become optimal. The product will evolve as the AI models used evolve. We think there is a lot of potential for this project as this can bring in more experimental approaches to policymaking and decision-making overall, as this could also be used in other environments. By making the feedback mechanism more immersive, this tool can also help citizens be more aware and involved in policy as it makes policy jargon more accessible. Overall, this project served as a practical learning experience in managing new digital tools, fostering innovation in public policy, and making strategic decisions under constraints. It also demonstrated the potential of AI in shaping future urban landscapes and policy decisions.
As the state-of-the-art technology evolves, we want to create different ways to showcase this future scenario to enable different feedback processes: through virtual reality, through video, through image portals in the city, through physical objects present in the virtual scenarios etc. We also want to create experiences where different types of stakeholders could provide feedback, and automatize the feedback collection. We want to reach out to stakeholders of relevance in the policy sector and in the civic community to test our product.
Drag & Drop Website Builder
Process
Iteration 1 - Crystal Ball Custom GPT
We first wanted to go through the entire process with a custom bot built on ChatGPT 4 to test the workflow and see if we could get to a final output. We trained a custom bot with Dator and Smart's future archetypes foresight methodology and gave the EU policy documents as input (Renovation Wave, EU Circular Economy Plan, Smart and Sustainable Mobility Strategy, New EU Bauhaus). The bot was trained to perform a foresight analysis based on the EU policy documents We had to reconfigure the bot several times to get to the optimal output. As we weren't satisfied with OpenAI's future image output, we decided to use another AI model using stable diffusion, which was better at doing image to image outputs.
Iteration 2 - Crystal Ball Tool
We wanted to make the entire experience automatic for users who do not have ChatGPT 4 so we decided to created an interface that would integrate the entire process. We used Modmatrix's source code and adapted it to create our own Crystal Ball interface on Replit. We used OpenAI's API to describe the input image and undertake the foresight analysis based on the EU policy documents but we couldn't use the stable diffusion model to create the future image as it wasn't open source. So we found another image to image model on replicate that could be integrated in the interface with an API. As the ultimate aim was to showcase this future scenario through an immersive experience with glitch, we worked with 360 images.