As the world races to harness the power of artificial intelligence (AI), some organizations have been reaping the benefits of machine learning (ML) and related technologies for over a decade.

Take Manish Jethwa, CTO of Ordnance Survey (OS), the UK's national mapping service. He’s on a mission to merge AI advancements—especially in generative AI—with their rich datasets to enhance the way geospatial data is accessed and utilized.

The Power of Foundation Models for AI

In a recent talk with ZDNET, Jethwa revealed how OS uses Foundation Models (FMs) to create specialized applications that make querying geospatial data more efficient. These FMs not only serve as a technological backbone but also allow data to be adapted and reused for various applications, thus maximizing their utility.

1. Develop a Strong Use Case

According to Jethwa, successful AI projects begin with clear and actionable use cases. OS focuses on extracting environmental features while navigating copyright considerations, utilizing its extensive experience in high-precision data collection to inform AI initiatives. He describes their approach as building foundation models from scratch, allowing for flexibility and repeatability across multiple data analyses.

2. Establish Purposeful Methods

Mindful training methods are crucial for optimizing resource allocation when building foundation models. Jethwa emphasizes that focused and deliberate training can significantly reduce operational costs. OS feeds training data in manageable chunks and gradually increases labelled datasets to ensure quality and efficiency.

3. Enhance with Large Language Models (LLMs)

Even with in-house foundation models, OS does not shy away from leveraging established LLMs. Jethwa noted their collaboration with commercial tools like Microsoft’s Azure for machine learning enhancements, focusing on fine-tuning LLMs based on specific organizational documentation.

4. Consider Commercialization

With the development of advanced foundation models, OS is considering the potential for commercialization—but it’s not simple. Crown copyright laws dictate how these public sector assets can be used or shared. Balancing public benefit with safeguarding these valuable resources poses a unique challenge for Jethwa and his team.

5. Focus on Future Innovations

Jethwa envisions an exciting future for AI at OS. He anticipates an interactive approach where users can engage with data directly by specifying their interests in a dynamic interface—like selecting schools on a map and querying for specific types in real time. He believes that true success lies in creating AI that aggregates data effectively while providing precise and accurate information.

By establishing strong foundational models, OS paves the way for a future where AI not only enhances data analysis but also informs critical decisions in the public domain, all while offering a model that can inspire others to follow suit.