A merged Franco-German city with the Eiffel Tower and Berlin Dome and half-timbered houses below a clear sky

Franco-German Exchange on AI in Urban Development

10/15/2025

The ISCN hosted a first Franco-German exchange on AI usage specifically in urban development. As a side event on the Smart Country Convention 2025 in Berlin, it brought together many experts and practitioners for knowledge exchange, networking and conceptual alignment.

Main content

artificial intelligence data management & data strategies organisational structures

Carrying on the “spirit of Toulon” that German chancellor Friedrich Merz invoked after the 25th Franco-German Council of Ministers at the end of august: That was part of the mission for this side-event of a Franco-German exchange on AI in urban development at the Smart Country Convention 2025 in Berlin. Renate Mitterhuber, Head of the Division Smart Cities and Regions from the German Federal Ministry for Housing, Urban Development and Building (BMWSB), invited the attendees and the collaborators in both countries in her opening remarks to see the potential of AI for common good-oriented urban development in two ways: For one, as a means to realize notable efficiency gains in established processes. This is often the dominant perspective where Europe sees itself challenged to catch up to other big players and mobilizes many resources. At the same time Ms Mitterhuber also highlighted the chance for a creative application of AI-driven technologies that may enable outright new policies, measures or ways to live in the city and which can benefit also from a reflective stance in Europe. 

The agenda was divided in two parts, with the first focussing on sharing use cases, and getting to know as well as mapping projects in both countries. And the second part putting an emphasis on Data ecosystems for urban AI. The programme can be downloaded here for reference, recordings of each presentation are available below.
 

Agenda of the Franco-German exchange on AI in urban development

Part 1: Sharing Use Cases, mapping projects

A coordinate with four quadrants clustering local AI projects in France across the dimensions public or administrational use and supporting existing policies or innovating and creating new services
Observatoire Data Publica (2025), IA & Territoires: Après la découverte, le temps des premiers choix

In order to set the scene, Marlene Damerau presented URBAN.KI as one project in Germany connecting many municipalities in their use for AI in urban development and which is federally funded by BMWSB. Enoh Tabak from the International Smart Cities Network (ISCN) swiftly gave some impressions on AI initiatives in France that are applicable to the German context. The graph below from Observatoire Data Publica (2025), IA & Territoires: Après la découverte, le temps des premiers choix shows for example that currently most municipal AI projects in France are focussing on applications of improving internal and established processes (bottom-left quadrant) and in Germany the situation seems to be similar. The top-right quadrant, where municipalities, civil society and private sector can jointly and more openly create new processes is so far not in focus, enriching it thus – in reference to Ms Mitterhuber’s welcome address – one of the sparks the event aimed to set.

In the next keynote Christian Kuhlmann from Westphalia university gave some additional theoretical backgrounds on the genesis of AI technologies and their current potentials for use in urban development. He then presented as examples projects of spatial analysis and landcover detection in several municipalities in North Rhine-Westphalia. The respective Foundation-Transformer-Model in the project “sursentia” shall produce deep insights from aerial images without priors on the given land coverings. 

Marie Bernard from Nantes Métropole described the streamlined governance process for structuring and clearing the pipeline of urban AI development projects. Notably, Nantes decided in a participatory process to completely ban the use of biometric data in AI applications. Yet this stance slightly straighter than broader EU legislation the proliferation of AI projects is vibrant.

Turning to the stakeholder group of civil society, Julian Stubbe presented the Civic Coding project which forms an innovation network for common good-oriented AI projects on behalf of four German federal ministries. He outlined the partial “AI-readiness” of civil society organizations in Germany as per a study they conducted, and interactively interviewed Carolin Johannsen from the project-turned-startup Reimagine Spaces that aims to use AI to streamline many layers of bottom-up urban planning processes.

Part 2: Data Ecosystems for Urban AI

The second half began with a remote video input by Natalia Carfi from Open Data Charter who mentioned some of their projects linking Open Data efforts with emerging AI trends. She mentioned for example the need for much more preparatory human validation work for establishing data fundaments from the Global South to train AI models. Moreover, through their TrustLaw project they compared AI regulatory frameworks in six countries identifying areas for improving regulatory design.

How ecosystems of AI models and data can already be leveraged across France and Germany was shown by Jonas Bostelmann from the State Office for Geoinformation and Surveying of Lower Saxony (LGLN). At their development hub they used training data from France to refine their model for spatial applications in Germany. In a way he presented precisely the example that all participants hoped to see more from: Mutual collaboration and reuse of resources, models and software that emerge in both countries, to not develop the same in parallel but improve solutions for mutual benefit.

Finally Björn Schwarze from Addix GmbH and the CAPTN project in Kiel illustrated how public-private partnerships can improve model development and solutions through data sharing. He showed how his company offers Wifi Connection data for analyzing passenger volume in busses, something hitherto extrapolated from a subsample of busses with cost-intensive sensors. In the CAPTN project data sharing is also used to work on steering and operating autonomous ferries.

Key points from working sessions

The two brief working sessions in between discussed firstly challenges and bottlenecks that municipalities still experience when wanting to adopt more AI use in the conceptualization of projects and processes, and secondly what existing or hypothetical datasets would be of high interest as contributions by various stakeholders so that urban AI use ushers into an age of both efficiency and effectiveness gains, but also creative new ways of urban development.

Some inputs are stated in the lists below:

 

Challenges and bottlenecks for AI adaptation by municipalities in both countries
  • Often lack of strategies specifically for AI
  • Lack of datasets and (labeled) training data – the latter of which can be partly done during the process if AI learns along human labelers
  • Lack of standardization and – in decentral structures – synchronicity of data structures and models
  • Concerns of data protection and (mis)understandings of legitimate interest
  • Many tools on the market lack transparency and appropriate handling of personal data
  • Often existing solutions’ outputs are too generic, e.g. regarding urban planning drafts, specification needs more local data
  • Lacking access to computing power for municipalities, with the acquisition of GPU capacities (e.g. in units of hours) not fitting with traditional procurement processes; largely a lack of centralized computing services => are collaborations by municipalities with public computing clusters sufficiently explored?
  • Institutional setup: Support from higher-level government could be more streamlined and directed at pools of municipalities
  • Budget and time constraints warranting clearer ROI benchmarks for some cases
  • Individual commitments to drive projects
  • Potential collisions of decarbonization efforts and energy-intense AI use
  • Capacity building for staff to improve expertise, as well as methodological and social skills for innovative AI projects
  • Clearly outlined use cases and/or designated possibilities to experiment, through technical sandboxes for example
Training set “wishlist”
  • Material throughputs in construction or productions, lifecycle data of buildings
  • Demographic fluctuations and predictions for transport utilities
  • Electrical devices and batteries in the city
  • (Anonymized) granular movement data
  • Aerial image data – RGB; near-infrared data (in parts available)
  • Drone-produced data (e.g. for accessibility assessments)
  • Granular noise levels
  • Health needs
  • Structured data on community project outcomes and participation processes
  • Data on usage and occupancy of (semi-/private) parking lots, parks or other resources
  • Granular carbon emission across value-chains
  • Proved synthetic data

But: Also bearing in mind that good foundation models can reduce the need for training sets

 

Next events on (municipal) AI use in France, Germany and internationally

The remainder of the year holds two more indicative events for (municipal) AI use in France, Germany and internationally and to which the threads and contacts developed in the side-event will contribute: