In ancient Greece, leaders travelled to Delphi seeking wisdom about the future. Today, European cities are building their own oracles—but these run on machine learning, not mysticism. The question they're asking is deceptively simple: Who does this mobility system actually work for?
The GOLIA project, a Horizon Europe initiative bringing together 21 partners across Florence, Pilsen, and Antwerp, is attempting something that sounds obvious but rarely happens in practice: designing sustainable mobility interventions that don't leave vulnerable populations behind. The centrepiece is the Social Optimum Mobility Index, or SOMI—a decision-support tool that shifts the focus from purely economic metrics to a more human question: does this policy actually improve wellbeing for the people who need it most?
Here's the thing. Most mobility planning optimises for speed, cost, or congestion. SOMI asks different questions. As Vasileios Giannoudis, Research Associate at the Hellenic Institute of Transport, explains in the Eurocities report:
Consider a policy that slows down traffic. To a commuter, that might look like a loss of availability. But to a parent, that is a victory for safety. The software knows the physics, but it doesn't know the values.
Vasileios Giannoudis
This is the implementation challenge that keeps me up at night. The model is the easy part. Teaching a system who cares about what—that's where projects succeed or fail.
What GOLIA Is Actually Building
The project isn't just another smart city dashboard. According to UITP's project documentation, GOLIA is developing a suite of interconnected tools:
SOMI itself measures mobility decisions against indicators including social support, health, equity, accessibility, and environmental impact. Unlike static indices, it's designed as a dynamic, scenario-based tool. Cities can test different policy choices—from street redesigns to regulatory changes—and explore how these decisions might affect different user groups over time.
The GO-X modules (GO-DATA, GO-INFO, GO-KNOW, GO-WISDOM) are structured around the Data-Information-Knowledge-Wisdom pyramid, designed to integrate local and cross-sectoral data sources. This matters because mobility data is typically siloed across departments—transport, health, urban planning, tourism—and never talks to each other.
The ENGAGEMOVE toolkit provides co-creation methods to involve citizens and stakeholders in mobility decision-making. This isn't consultation theatre. It's structured engagement designed to capture what different personas actually value.
The three partner cities each bring distinct challenges. Pilsen faces strong commuting patterns with high car dependency despite good public transport infrastructure. It's difficult to change habits, says Jaroslava Kypetová, Head of Grants and Integrated Territorial Investments at the city. Florence, as Elena Aversa from the city's Fundraising and EU Projects Office notes, as a tourist city, faces a high flow of mobility. Antwerp is using the index to upgrade its Smart Ways to Antwerp digital tool for vulnerable groups. The app will work for them, but first we need to know what they're looking for, explains Michiel Penne, the tool's coordinator.
Three follower cities—Marseille-Provence, Glasgow, and Riga—will engage in peer learning to test replication strategies.
The Counterpoint: Why This Could Go Wrong
Before we celebrate the oracle, let's talk about what could break. Because in my experience, the gap between promising EU project and deployed system that actually helps people is where most initiatives die.
The bias problem is real. Research published in the Journal of the American Planning Association identifies critical ethical concerns in AI-driven urban planning: Bias in AI systems can lead to unequal outcomes, disproportionately affecting marginalized communities. If the training data doesn't adequately represent people with disabilities, migrants, or low-income residents, the system will optimise for the people it knows—typically the majority who already have decent mobility options.
The black box problem persists. The same research notes that transparency issues arise from the black box nature of AI, complicating understanding and trust in AI-driven decisions. If planners can't explain why SOMI recommends one intervention over another, they can't defend those decisions to sceptical stakeholders or affected communities.
Data fragmentation is the silent killer. As Matteo Salani, Senior Researcher at SUPSI-ISIN, acknowledges in the Eurocities article: The fragmentation of mobility data and siloed decision-making, which prevents holistic urban planning, is a key challenge. GOLIA's GO-X modules are designed to address this, but integrating data across departments requires political will, not just technical architecture.
Vulnerable groups are hard to reach. A UN report on AI and disability inclusion notes that while technology has great potential to transform lives, more than 2.5 billion people who need assistive products are denied access, particularly in lower-income contexts. The people GOLIA most wants to help may be the hardest to include in the design process.
What Makes SOMI Different (If It Works)
The conceptual shift here is significant. Traditional mobility indices measure vehicles—speed, congestion, throughput. SOMI attempts to measure people's transport needs. That's a fundamentally different optimisation target.
The project documentation describes SOMI as putting together data not only from mobility, but also from implications on health, urbanism, tourism. As Marisa Meta, Project Coordinator from Fit Consulting, puts it: SOMI will make sure all users have equitable access to mobility systems, as mobility planning tends to ignore vulnerable groups such as persons with disabilities or migrant communities.
The scenario-based approach is also worth noting. Andrea Baldassari, Researcher from SUPSI, explains that after you implemented it, you can monitor the situation, and check if the actual SOMI has increased as predicted. This creates a feedback loop—the system learns whether its predictions were accurate and adjusts accordingly.
This is exactly the kind of observability I advocate for. Before accuracy: observability. If you can't measure whether your intervention actually improved outcomes for the target population, you're flying blind.
The Implementation Checklist
For cities watching GOLIA and considering similar approaches, here's what I'd want to see before deployment:
Define good enough for each vulnerable group. What does success look like for wheelchair users? For elderly residents? For migrants with limited language skills? If you can't articulate specific, measurable outcomes for each persona, you're not ready.
Build the rollback plan before the launch plan. What happens if SOMI recommends an intervention that makes things worse for a specific group? How quickly can you detect the problem and reverse course?
Map the data dependencies. Which departments need to share data? What are the privacy implications? Who owns the integration when it breaks? These aren't technical questions—they're governance questions.
Test with the hardest cases first. Don't validate on the easy scenarios. Test whether the system can actually identify and prioritise the needs of a Turkish woman in Antwerp, a person with a disability in Pilsen, a retired teacher in Florence. If it can't handle the edge cases, it's not ready for production.
Document what the system doesn't know. Every model has blind spots. Make them explicit. If SOMI can't account for certain types of disability or certain cultural barriers to mobility, say so clearly.
The Bigger Picture
GOLIA sits within a broader European push toward what UITP calls Mobility as a Right—the idea that access to transport is a fundamental entitlement, not just an economic function. The project will contribute to a Mobility Smart Governance Handbook with policy recommendations aligned to the European Green Deal and UN Sustainable Development Goals.
The UN-Habitat Global Assessment of Responsible AI in Cities found that AI research in urban governance is concentrated in Europe, North America, and Asia, with Africa and Central America underrepresented. It also found that AI integration with public administration remains mostly theoretical with limited practical applications. GOLIA has the opportunity to change that—but only if it moves from concept to deployed, monitored, iteratively improved system.
The project runs until 2028. That's enough time to test, fail, learn, and iterate—if the team treats failure as data rather than shame.
The Question That Matters
Here's what I'll be watching: Can GOLIA actually demonstrate that SOMI-informed interventions produce better outcomes for vulnerable groups than traditional planning approaches? Not in a pilot. Not in a controlled study. In real cities, with real constraints, over time.
The oracle at Delphi was famously ambiguous. Modern AI systems can be too—confident in their predictions but opaque in their reasoning. The test for GOLIA isn't whether it can generate sophisticated indices. It's whether those indices translate into mobility systems that actually work for the people who've been overlooked.
If you can't explain the decision to the user, you can't ship it. And if you can't measure whether it helped, you haven't finished the job.