Short Abstract

The workshop showcases previous examples - Prague and Haifa and creates new ones from Krakow’s Sentinel-2 data. Participants turn buildings and rivers into AI personas that speak in a civic theater, exploring how data and grammar shape public reasoning. https://github.com/anonette/satelite_personas

Description

An older version of the demo https://drive.google.com/file/d/1T7ZTEQthoUtkUDIjaSmp6txqiOK82qMf/view?usp=sharing

This demo invites participants to experience environmental data not as numbers on a dashboard but as voices in a civic theater. Using Sentinel-2 satellite images of Prague, Haifa and Krakow (developed in a workshop), the system generates AI personas (LLM-based agents) that embody spectral conditions such as drought, vegetation vitality, or urban sprawl. Instead of producing static graphs, these personas speak, argue, and contradict one another. The city becomes a stage where data is dramatized as conflict, opening a space for playful yet critical exploration of urban ecologies and planetary futures. At the venue, visitors will interact with a web interface. They can select satellite data, generate characters, and act as “directors” who arrange the cast, insert surprise agents, or re-order debates. Dialogues are displayed on screen and can be heard through AI-generated voices, creating the sense of a live performance. The experience is not about solutions or optimization but about witnessing multiple perspectives at once. It is humorous and unsettling, turning remote sensing into speculative civic theater. The demo explores alternative modalities of crisis management in which neighborhoods and environmental elements speak directly to one another, decentralizing authority in disaster scenarios.

Questions the Demo Raises

How does AI persona design shift when agents are generated from spectral or environmental data rather than human traits or demographics?

Can AI agents function as civic performers rather than assistants or optimizers, and what forms of engagement does this afford?

How does staging AI interactions as theater differ from rule-based simulation models in terms of reproducibility, interpretation, and public participation?

What does it mean to treat AI dialogues as performances to be witnessed, rather than predictions to be validated?

Can simulation be reframed as a civic rehearsal space, where publics experiment with conflict and uncertainty rather than resolve it?