Opening Infrastructure Maps to AI Demands Governance and Sovereignty Before Speed
By Professor Ashley Braganza
A hydrant located on an electron map is old technology; asking AI about it is not. Utilities, farmers, city planners and energy companies have recorded their infrastructure assets such as pipes, valves and grid connections in mapping software for decades. The software is mature. The change is AI that lets a manager put a plain question to that map and receive a ranked answer in a chat window. That change widens access to critical infrastructure data at a moment when trust between nations is weakening, and it turns data governance from a technical chore into a question of civic resilience.
My guest on The AI Adoption Podcast is Dino Ravnić, Co-founder and CEO of GIS Cloud. Dino has spent more than 15 years bringing location data closer to utilities, government and energy organisations. Geographic information systems, or GIS, store and analyse data by where things are, and present it on a map. Dino's case is that the cloud opened GIS to non-specialists and that AI completes the job, provided a layer of control sits between the AI and the data.
AI, not the map, is the change boards should watch
GIS began as a scientific tool. Specialists queried databases, printed maps and interpreted them for colleagues. Dino describes those specialists as a bottleneck in many organisations: they cannot meet the demand placed on them, and the role is hard to scale. Cloud software moved part of the work onto phones and browsers, so a field worker inspecting an asset records the inspection on site. AI goes further. A manager can ask which assets should be repaired first, and the AI can draw on other sources, in the case of hydrants, the locations of schools, hospitals and homes for elderly people who are hard to evacuate in a fire, then run the query through the GIS and return a prioritised list.
The value for leaders lies in speed and in visibility over physical assets. Dino gives the example of a solar plant built using a GIS, in which each panel is a small polygon on the map. This allows decision makers to follow progress from frames to panels to wiring in real time. In the field, AI can compare a photo of a asset with what the inspector reported and flag discrepancies for a person to check. Dino is clear that a person signs off, and that the organisation must know who is responsible.
The result is a live system used by field crews, planners and budget holders, rather than a map consulted by a few specialists. In Dino's account, a GIS is dependable only when those groups use it daily; stale or inaccessible data undermines it.
Feeding whole datasets to a model is the wrong design
The obvious design is to hand a large language model, the type of AI behind ChatGPT and Claude, the entire dataset. Dino rejects it. Tokens, the units in which AI providers charge for the text a model reads and writes, would "burn your budgets". AI can hallucinate, producing answers that sound plausible and are false. The alternative is a deterministic system, one that returns the same answer to the same query each time, sitting between the data and the AI. The AI asks the question; the GIS executes the query; the result comes back to the user's chat.
This complicates the democratisation story. Dino describes AI as "the final step where all the barriers in using GIS will be demolished" and as "the final integrator of all that data coming in", from sensors, field crews and AI agents. Yet the same argument insists that AI must not be trusted with the full dataset, and that prompting is a skill: "AI can be dangerous if you are not skilled at it." Moreover, whether AI technologies are capable of integrating sensor feeds, field records and enterprise systems to this extent, and of ranking repairs reliably, is questionable.
The tension is productive. Barriers to asking questions fall; barriers around what the AI can see and change rise. On this evidence, democratisation means wider access to answers, not wider access to the underlying data. Leaders who conflate the two are likely to open more than they intend.
The data that speeds repairs can also guide an attack
A map showing a crew where water mains and grid connections run would show the same thing to anyone wishing to disrupt a city. When I raised this, Dino called it "one of the biggest challenges": organisations want the efficiency that cloud services and AI offer, and opening data up carries danger.
Dino's answer turns on trust. The vendor's origin and the data's location matter more than before. European customers often want data kept within the EU, or held in their own networks and data centres. On models, Dino's assessment is that the United States has the state of the art, Europe lags a little, models also come from China, and open weight models, which an organisation can run itself, need more setup and tuning. Mistral is the European option Dino names. In Dino's view, regions that care about sovereignty and face a problem of trust "will have to build their own AI".
Mapping has handled a version of this before. Military sites and some critical infrastructure are blurred on public maps or need special permission to view. Dino suggests comparable measures for AI: opening data gradually, barring some parts from AI access, and anonymising or scrambling sensitive records.
My reading is that sovereignty over infrastructure data is likely to become a board responsibility, comparable to financial control. The choice of model, where it is hosted and which layers of data it touches is a decision about civic resilience and not a procurement detail.
Permission to share and permission to let AI act must be separate
GIS Cloud's design offers a working model. A data owner can share a map or dataset with a colleague and, separately, grant or withhold permission for that colleague to use the data with AI, for reading or for editing. Dino's principle is plain: "I as a data owner, I'm responsible for my data and that's the bottom line. But once you start to share the data, then you lose control."
For leaders, visibility comes before scale. Dino calls for transparency over what AI is doing with the data, which data it is accessing and which data is off limits. Shadow AI sits outside any such layer. Dino observes that staff can export data and drag it into ChatGPT or Claude, sometimes without the organisation being aware, so a permission system that ignores this route governs only the front door. Zero data retention policies, under which an AI vendor keeps no trace of the data sent with a question, belong in the terms agreed with providers. GIS Cloud describes itself as agnostic on the model, so customers can choose which model runs on their data.
This is a point I put to boards and senior managers I work with: before setting a timetable for AI across asset data, establish who can grant AI access to that data, and who would know if the access were misused. Dino places the GIS manager at the centre of this, a specialist in control of adoption rather than a bottleneck to it.
The hardest test lies beyond the organisation. Dino extends the argument to citizens: publicly owned data on kindergartens, schools, hospitals, hydrants and outages already exists, often as open data, yet sits in technical portals that few people use. AI could let a family moving house ask for the nearest school and receive an answer drawn from city GIS data. Dino sees the greatest value of AI and GIS together here, and it widens the same exposure. The challenge for city leaders is to decide which layers of their map citizens and AI may see, before an adversary decides for them.
Listen to the full conversation
Comments