OGD4All: A Framework for Accessible Interaction with Geospatial Open Government Data Based on Large Language Models

LLMs can make government data more accessible to ordinary citizens: OGD4All demonstrates that large language models can translate complex geospatial datasets into understandable answers, potentially enabling non-technical citizens to access public information about their cities without specialized training or technical skills.

The system is designed with transparency and auditability in mind: The framework uses "sandboxed execution" and produces "verifiable multimodal outputs," meaning users can see how answers were generated and verify the reasoning behind them—addressing a key concern about AI systems operating as "black boxes."

Hallucination risks can be substantially reduced through proper design: The system achieved 94% recall while "reliably rejecting questions unsupported by available data," suggesting that well-designed AI systems can be trained to admit when they don't know something rather than making up plausible-sounding false answers.

High accuracy is achievable on real-world government data: Testing across 430 actual City-of-Zurich datasets and 11 different LLMs, the system reached 98% analytical correctness, indicating that responsible AI applications can perform reliably on practical governance tasks.

Expert validation matters for responsible AI deployment: The paper emphasizes that statistical testing and "expert feedback" were essential to demonstrating reliability and "social relevance," suggesting that responsible AI requires human oversight alongside technical safeguards.