Why AI governance starts with data management

Civil servants today increasingly turn to artificial intelligence (AI) tools for answers that they once sought from government websites. 

Despite better efficiency and personalised responses, the AI assistant can still give incorrect information.

Open Knowledge Foundation (OKFN)’s CEO, Renata Ávila, believes that AI readiness depends on whether civil servants have the data, infrastructure and accountability to give citizens reliable information.

OKFN is a global non-profit organisation dedicated to promoting open data, open content and open-by-design digital infrastructure. Their current initiatives focus on ethical digital governance, helping public officials to build AI literacy. 

“Everybody seems to have forgotten about data,” she says. “If you don't have good data practice, or if you have disconnected data sets across different institutions, you can’t deliver the quality information.”

For governments, the AI challenge may not be adopting a more advanced tool, but making sure the information across the system is accessible and reliable.  

Data management is the key to building trustworthy AI  

Information is harder to manage today, unlike the past where government portals were managed by dedicated teams. 

“Something that was quite tangible has become a much more complex relationship between governments, platforms and algorithms,” Ávila tells GovInsider. 

“The bad quality of data usually comes from the most under-resourced and overworked civil servants,” she adds. 

When the information is inaccessible or unreliable, citizens may lose confidence in official channels and look elsewhere for answers, including AI that can produce convincing but inaccurate information.

Instead of correcting the misinformation after it spreads, governments should make authoritative information easier to find and verify from the start.  

This is what OKFN has been working on, by connecting AI with government open-data portals. 

Known as Model Context Protocol (MCP) and currently implemented in Brazil and Uruguay, the work explores how AI could retrieve answers directly from government data, citizens can receive answers based on information that can be checked and verified.

For instance, a citizen receiving a risk alert on the phone wants to know what to do next. 

Instead of piecing together scattered updates from social media, they could ask for real-time information grounded in verified government data. 

The source would be traceable to the relevant authority, creating a clear chain of responsibility if the information is wrong.

AI readiness starts with managing the data right

Despite the right data source, AI can still misinterpret what it means. For example, an AI could introduce climate-related explanations that are not supported by the underlying energy data.

This is why AI readiness is less about the tool but more about how people manage the data.

To address this, OKFN has developed the Open Data Editor (ODE), a no-code tool that allows non-technical civil servants to identify and fix errors in datasets.

“For someone who has no code experience, it enables them to upload datasets,” Ávila says. “It guides you through the errors that your datasets have, and it enables you to fix it.”

This saves civil servants’ time and effort while providing a cleaner dataset for more accurate AI responses. 

“Knowledge and data are part of the critical infrastructure,” she says, “the same way we consider a bridge, electricity or water supply as critical infrastructure.”

Governments can start by co-developing a prototype with the citizens, instead of waiting for a perfect system.  

“Prototyping accelerates the learning curve of the public servants and enables a space for local techies, universities, and small companies to participate and increase citizens’ trust,” she shares. 

The ODE approach has been adopted by Open Knowledge Nepal to automate dataset audits and standardise metadata. 

This reduced error-resolution time and fixed 95 per cent of errors while empowering the local government staff to initiate training sessions for non-technical users. 

Building public trust by making information accessible and reliable

Beyond pursuing AI innovations, Ávila believes that what matters more is creating inventories of government datasets, knowing who maintains them, understanding access and permissions among the civil servants and external technology providers.

Ultimately, public institutions’ success in AI adoption should not be measured only by how advanced the technology is, but also by whether citizens using the service get the information they are looking for. 

“A citizen wants to know whether a road is safe, whether medication is available, how much fuel costs.

“If the government can provide that information quickly, accurately and with a clear source, citizens have a reason to return to official channels.”

That is a more meaningful measure of AI impact, she says.