edparsons.com

The promise of frontier AI models is often limited not by compute power or architectural ingenuity, but by the quality, legality, and sheer accessibility of the data used to train them.

My goto phase for the last few years remains..

The source code for AI is Data !

We’ve spent years in the geospatial and data policy world wrestling with silos and proprietary APIs. Now, the AI industry is finding its own answer to this fragmentation: the Model Context Protocol (MCP).

While much of the recent press focuses on MCP’s role in powering sophisticated AI agents and reducing LLM hallucinations at runtime, the real long-term impact lies further up the chain—in standardising how models acquire the data they need to be built and continuously refined.

The AI sausage factory..

Historically, training large AI models meant engaging in a bespoke, often messy, pipeline for every new data source. Licensing agreements, ingestion methods, metadata parsing, and normalisation all required custom engineering. Yes I know GIS people have also been dealing with this for decades too !

So the core problem, which MCP solves, is the lack of a universal data contract for AI. Data lives in thousands of disparate silos—internal databases, cloud file systems, proprietary APIs, and data commons. Every integration is a one-off effort

The Model Context Protocol, open-sourced by Anthropic and rapidly adopted across the industry (including by Google and OpenAI), solves this by establishing an open standard for bidirectional communication between an AI model (the client) and external systems (the MCP server)

Think of MCP as the USB-C interface for AI. Instead of creating a custom cable (API connector) for every data source, developers simply need to ensure the data source exposes an MCP server, guaranteeing a standardised interface for discovery, authentication, and structured data delivery

Better training and more ?

While MCP is primarily aimed at training models using Retrieval-Augmented Generation (RAG)—giving models real-time context—its standardisation effects may have profound benefits beyond the lifecycle of creating and maintaining high-quality training datasets, and improve geodata management practices in general.

For example;

• Simplified Data Curation and Continuous Learning.
  MCP defines specifications for data ingestion and contextual metadata tagging. For data providers, this means the effort required to make a dataset consumable by AI drops dramatically. This standardisation allows model developers to ingest data from diverse sources (Postgres, GitHub, Google Drive) through a unified mechanism. 
• Standardized access enforces a clearer chain of custody.
  When a model uses an MCP server to access a dataset, the connection is structured and auditable. This aids in creating transparent data cards for trained models, making it easier to trace which specific version of a dataset, exposed via a specific MCP server, contributed to a model’s training, boosting trust and compliance
• Access to Niche and Enterprise Data
  For specialised models , particularly in domains like geospatial or finance, the most valuable data often sits locked behind firewalls or in niche, non-public systems. By providing a secure, standardized way for enterprises to expose their internal data repositories via an MCP server, the protocol opens up a powerful channel for targeted, permissioned data access necessary for domain-specific model training and custom enterprise LLMs. 
  
  And yes this is where I think the real money will be made in AI !

A Case Study from my old life..

The real-world potential of MCP’s standardisation is well demonstrated by facilitating access to complex public datasets. The recent announcement by Google of  Data Commons Model Context Protocol (MCP) Server is a great example.

Data Commons aggregates vast, interconnected public statistical information—data on health, economics, demographics, and more—which is often scattered across thousands of disparate silos and organized by complex, technical jargon.Historically, using this data in an AI application required developers to learn and integrate with a complex, proprietary API.

The immediate benefit is that The server allows any MCP-enabled AI agent to consume Data Commons natively using standardized MCP requests, accelerates the creation of data-rich, agentic applications – the current flavour of the month!

Still work to do…

It is still early days for MCP, and we are some way away from universal adoption, and there are still challenges that will need to be addressed..

Introducing a standardized, two-way connection to sensitive data systems, where an autonomous AI agent (the MCP client) can request and act upon data, inherently introduces new security vectors. If not secured with robust authentication and granular authorisation (something MCP aims to support, but which is server-dependent), the protocol could become a standardized way for compromised AI agents to access enterprise databases.

While the protocol has seen rapid initial adoption, the ecosystem is still in its infancy (literally a year old !). There is the risk that a major player could splinter the standard by introducing a non-interoperable alternative, leading to the exact data fragmentation the protocol was designed to solve. Hey that’s never happened right ?

For simple use cases, the MCP architecture (Host, Client, Server, Transport Layer) adds another layer of abstraction between the model and the data. This complexity can introduce overhead in terms of latency, debugging, and infrastructure management compared to direct, custom API calls.

What not How !

The Model Context Protocol is a defining step towards treating AI data access as an infrastructural concern rather than a bespoke engineering project. By standardizing the interface, MCP shifts the focus from how to connect data to what data to connect.

MCP may also serve as a wake up call to Geospatial Data users to rethink how we publish and consume data in a world changed by the requirements to train AI.

Imagine a world where you never have to worry about buying the right train ticket again. Ok I know this does not sound like much of a problem, but then how well do you know the UK’s now mostly nationalised Railway system…

No more fumbling for change, wrestling with a complex app, or trying to figure out which fare is the cheapest for your journey.

This isn’t a distant dream—it’s a potential application of geospatial technology, and it’s being trialled in the UK right now.

Geospatial technology uses data related to a specific location, combining it with real-time information like GPS. This is the same tech that powers your favourite navigation apps, from Google Maps to Waze. However, its potential extends far beyond simply getting you from point A to point B.

The UK government’s recent trial of location-based ticketing is a perfect example of this. Backed by nearly £1 million in government funding, the trial is happening in the Midlands and North, on East Midlands Railway and Northern trains.

The system uses a location-based app that tracks a passenger’s journey using GPS. As you travel, the app intelligently calculates the best fare for you, automatically charging your account at the end of the day.

This approach offers a significant upgrade from traditional paper tickets and even modern mobile tickets that use QR codes. It eliminates the need to pre-book, making travel more flexible and spontaneous. For ticket inspections or passing through barriers, the app generates a unique barcode. This is a game-changer for the daily commuter and the occasional traveller alike, as it ensures you always pay the optimal price without the hassle of guesswork.

While this technology has already been tested in Switzerland, Denmark, and Scotland, its implementation in England marks a significant step forward. It showcases how geospatial technology can be leveraged to modernise our transport systems, improve the passenger experience, and even encourage more people to use public transport.

As the world becomes increasingly connected, we can expect to see more innovative applications of geospatial technology, from smart urban planning to logistics and supply chain management. The rail trial in the UK is just the beginning of a new era of seamless, intelligent travel.

Oh and by the way is not called GeoTrainTicket 😉

It’s the end of summer so I give you a light hearted experiment in the application of Generative AI.

My quest to recreate the cinematic masterpiece The Battle of Britain, but with a twist, has hit a slight snag. The twist, you ask? Cats. Yes, I wanted to harness the power of GenAI to create a feline-filled tribute to the classic war film.

Think Spitfires with whiskers, air raid sirens replaced by meows, and a stoic Winston Churchill… also a cat.

The results, as you might have guessed, were not a complete success. Just take a look at the video.

My first discovery was that generative AI, bless its digital heart, has a very… interesting interpretation of “important elements.” I tried to get specific. I wanted to see brave pilots walking out to their planes, but with a squadron of cats alongside them. What I got back was a lot of things. Pilots. Planes. Cats. But putting them all together in a cohesive, logical scene was a different story.

As you can see, the cats are simply… there. They’re just following the men, doing their own thing (well they are cats I suppose). The AI seems to get the gist of the individual components but struggles to grasp the relationship between them. It’s like the cats and the pilots exist in the same frame but not the same reality (deep eh?).

The real comedy gold came from the images that were almost right. My favorite moment has to be the black cat, trotting along, seemingly completely uninterested in the magnificent Spitfire in front of it. Of course for the Aviation Buffs amongst you, that Spitfire is just weird isn’t it !

It’s like the AI got an assignment, did its homework, but then forgot to read the instructions on the back. It knows the keywords—”cat,” “Spitfire,” “battle”—but the context is all a bit… fuzzy. It feels like the models are trying their best, pulling together a collage of what they’ve been trained on, but they’re still missing that spark of creative understanding that connects the dots in a meaningful way.

My experiment wasn’t a total bust, though. I discovered a secret: GenAI absolutely thrives on the generic. A simple prompt like “A London street scene” works much better, although it suffers from the Hollywood problem of over reliance on London Buses and Taxi’s to create a sense of place.

So, while my epic film saga, The Battle of Britain (Starring Cats), remains unproduced for now, I’ve learned a valuable lesson. GenAI is fantastic for creating beautiful, generic scenes, but when you try to get too specific, too creative, or too… cat-tastic, it might just show you the limitations of its training.

The geospatial industry has long been on a quest for a universally accepted system to identify real-world entities like buildings, roads, and businesses. Such a system promises seamless interoperability, but its pursuit has been fraught with challenges, often clashing with the commercial realities and vested interests of powerful data owners. Let’s explore this journey through two powerful metaphors: the “One Ring” and the “Holy Grail.”

The Temptation of the “One Ring”: Proprietary Control

Imagine a single, powerful “One Ring” – an entity ID system controlled by one dominant vendor. This system would indeed be powerful, offering a seemingly straightforward way to identify and link data within that vendor’s ecosystem. However, like the One Ring, its singular control could lead to significant issues: vendor lock-in, a “master of all” scenario, and a lack of true openness for the wider geospatial community.

We’ve seen this play out historically. In the UK, the OS Mastermap TOID (Topographic Identifier) was an early attempt at location identifiers. While the concept was valuable, its primary drawback was its proprietary nature and the significant cost associated with licensing the underlying Mastermap database. Users found it difficult to leverage TOIDs effectively without incurring substantial licensing fees, making it a “pointless creation” for those without a Mastermap licence. Such systems, while providing internal consistency, inherently limit widespread adoption and true interoperability outside their controlled environments.

Building on this, the “Holy Grail” metaphor represents the ultimate prize: a single, seamless, universal entity ID system that allows effortless interoperability across all geospatial datasets, regardless of their origin. Yet, its attainment often proves elusive, primarily due to the vested interests of data owners.

The Elusiveness of the “Holy Grail”: Universal Interoperability vs. Vested Interests

Consider WOEID (Where On Earth ID’s), launched by Yahoo! in 2008. This initiative aimed to provide an API for place IDs, promising a “neat way of providing a little more structure around the geotag clouds currently multiplying”. It offered a more open alternative compared to the restrictive OS Mastermap TOID. Despite its initial promise, WOEID did not achieve the widespread, lasting success of a truly global, open standard. While not explicitly stated as “vested interests” in the source, the implicit challenge for any such system is gaining universal buy-in and consistent adoption from all potential data contributors and consumers, many of whom have their own established systems or commercial motivations.

Fast forward to today, and we observe the impressive efficacy of powerful, single-vendor solutions such as Google’s Place IDs. These textual identifiers uniquely identify a place within the Google Places database and on Google Maps. They are widely accepted across a multitude of Google Maps Platform APIs, including the Places API (New and Legacy), Geocoding API, Routes API, Maps Embed API, and Roads API. Developers can readily find a Place ID using the Place ID finder or through various search requests…

However, the very strength of Google Place IDs within their ecosystem highlights the “Holy Grail” dilemma. As a proprietary system, Place IDs inherently tie users to Google’s platform. They are unique identifiers for places in Google’s database. While they can be stored, Google recommends refreshing them if they are more than 12 months old due to potential changes in their database. An obsolete Place ID, perhaps because a business closes or moves, will return a NOT_FOUND status code. This dependency on a single vendor’s database underscores the challenge of proprietary systems achieving universal interoperability across all geospatial datasets, particularly those from other providers. Using Google Place IDs primarily facilitates integration within Google’s sphere, potentially leading to vendor lock-in and continued data integration costs when combining with data from non-Google sources. The “vested interest” here is the company’s control over its own vast and frequently updated dataset, which, while beneficial to its users, does not inherently solve the broader industry’s need for open, cross-platform identification.

The dilemma of the Holy Grail – a singular, universally perfect solution – often runs aground on the reality of diverse data ownership and commercial interests. Each major data provider has invested significantly in its own data collection, curation, and identification systems. To simply adopt one provider’s system as the universal standard would require others to concede their proprietary advantage, a highly unlikely scenario.

The Overture Maps Foundation and GERS: Forging a Shared Path

Recognising these challenges, the Overture Maps Foundation aims to offer a different path – not to possess the Holy Grail, but to make its essence universally accessible and shareable through open collaboration. Their Global Entity Reference System (GERS) is positioned as the first truly open, global, and entity-based ID system. Its open nature means it does not lock users into a single vendor’s ecosystem. Instead, it is supported by a diverse community of nearly 40 organisations, including major players like Amazon, Meta, Microsoft, TomTom, and Esri, who have committed to and are dependent upon Overture data.

GERS seeks to overcome the “data conflation tax” – the significant time and resources organisations spend on data preparation and integration. By providing persistent, unique identifiers for geospatial entities, GERS aims to transform weeks of complex geospatial conflation into simple column joins within minutes. This is achieved through its four key components:

1. Overture Reference Map: Monthly validated datasets where each entity carries a unique GERS ID, which is open, free, and accessible
2. Data Changelogs: Detailed records of changes between releases, allowing for efficient updates.
3. The GERS Registry: A comprehensive database of every GERS ID ever created, offering validation, history, and lookup capabilities.
4. Bridge Files: Pre-built mappings between GERS IDs and identifiers from other popular datasets like OpenStreetMap, Meta’s data, and Esri Community Maps, enabling instant integration.

Crucially, GERS IDs themselves utilise the UUID (Universally Unique Identifier) format, ensuring they are globally unique, system-agnostic, and compatible with existing tools and libraries. This means no collisions and universal compatibility across modern databases and programming languages. In essence, while the Holy Grail of a single, centrally controlled, and universally imposed ID system might remain elusive due to the inherent vested interests of numerous powerful data owners, GERS presents a compelling alternative. 

It’s not about one entity holding the Grail, but about forging a shared path towards its benefits through open collaboration. By building a system that is openly accessible, jointly maintained, and designed for interoperability rather than exclusive control, GERS offers the closest practical solution to uniting disparate geospatial data without demanding that any single data owner surrender their fundamental interests. 

Time will tell as to the success of this Grail Quest. I’m keeping my fingers crossed, but I also know that if this doesn’t work out, future Knights of Geospatial will probably pass this way once again!

This famous phrase by Chairman Mao can be subverted to focus on the final step of a journey, which is particularly relevant to solving one of the hardest problems in geospatial technology: the final 10-metre problem.

How can we provide high-accuracy pedestrian navigation or wayfinding to ensure users reach their actual final destination, such as the front door of a building, the office inside a large building, or the check-in desk at an airport?

For many, the solution to the final 10-metre problem would be a great convenience. However, for some, it’s vital because it represents a step change in accessibility, creating a more equitable experience for those with visual impairments and other disabilities.

I’m thrilled to share some exciting news about a new advisory role I’ve taken on. I’m now serving as a Technical Adviser to the Board at Waymap, an innovative London-based  firm that’s making significant waves in the personal navigation space. Waymap is one of a handful of fascinating small companies I’m supporting, and their mission particularly resonates with my passion for geospatial technology and accessibility.

Throughout my long career, first at Ordnance Survey and then at Google, I’ve been deeply intrigued by how we can use geography to organise and make the world’s information more accessible. Waymap is at the forefront of a new wave of personal navigation, with a crucial focus on precision for pedestrians, especially in challenging indoor environments and the crucial “last ten metres”.

valuable but difficult problems

What truly excites me about Waymap is their approach to solving “valuable but difficult problems” by breaking new ground to make personal navigation truly accessible and precise for all, thereby delivering significant social benefits.

For more details, visit the Waymap website. Some interesting aspects of their solution include:

• Waymap, founded by Dr. Tom Pey, a sight-loss survivor, is committed to accessibility at its core. Their mission is to genuinely work for everyone, transforming the lives of blind and disabled users and making the world easier to navigate for all.
• Waymap’s high-precision location determination allows for pedestrian navigation indoors and outdoors. This is a significant advantage, as Waymap uses a revolutionary algorithm that utilises only the motion sensors on your phone and your steps. This means no Wi-Fi, GPS, or external signal is required, enabling Waymap to work in various environments, including deep underground.
• Unlike Bluetooth beacons and costly installations, Waymap doesn’t require any physical infrastructure. This dramatically reduces deployment and upkeep costs for venues, as they only need a map of the building.
• Waymap offers a personalised experience by learning your walking style and understanding your individual requirements. It guides you hands-free with audio instructions.

Beyond its profound impact on accessibility, Waymap’s precision should improve the experience for everyone, including tourists, shoppers, and hospital patients. By deploying Waymap’s solution, organisations position themselves as leaders in the social aspect of ESG (Environmental, Social, and Governance).

I am thrilled to be working with the Waymap team as we push the boundaries of what’s possible in navigation.

When Snap Map first launched within the Snapchat app, it was met with a mix of curiosity and privacy concerns. Yet, this integrated feature has quietly become a significant success story, particularly among its core demographic – Millennials. While perhaps not always the focus of mainstream tech commentary, Snap Map has proven to be far more than just a simple location-sharing tool; it’s a dynamic social and discovery platform that resonates strongly with this generation.

Snap Map’s genius lies in how it caters to Millennials, a generation that grew up alongside the internet and the rise of social media. It transforms the static concept of a map into a vibrant, real-time representation of their social world. Users can see cartoon-like “Actionmojis” of their friends’ Bitmojis placed on a map, indicating their general location and what they might be doing (based on factors like speed or time of day).

This real-time social layer is incredibly powerful for Millennials. It facilitates spontaneous meetups (“Oh, you’re nearby? Let’s grab coffee!”), provides a sense of connection to friends scattered across a city or even the globe (“Just checking in on what everyone’s up to”), and subtly fuels that ever-present fear of missing out (FOMO) in a visually engaging way. It’s not just about knowing where someone is, but feeling connected to their immediate reality and potential activities, aligning with the Millennial desire for authentic, in-the-moment experiences.

Beyond the social aspect, Snap Map has evolved into a powerful discovery engine. Heatmaps appear on the map, indicating areas where a lot of Snapchat activity is happening – be it a major concert, a local festival, or just a popular hangout spot. Millennials can tap on these areas or specific locations to view public Snaps submitted by people at that place, offering a ground-level view of events and environments as they unfold. This transforms the map into a living, breathing news feed of local happenings, curated by the community itself, perfect for a generation keen on exploring and experiencing their surroundings.

For Millennials, Snap Map integrates seamlessly into their existing Snapchat habits. Checking the map becomes as natural as viewing stories or sending snaps. It leverages the platform’s visual language and social graph to create a unique, engaging experience that keeps users within the app longer and reinforces their connection to their friends and their local environment.

A Different Kind of Local Search: Competing with Giants?

While Google Maps remains the undisputed king of navigation and comprehensive business listings, Snap Map presents an interesting, albeit different, form of local search and discovery, particularly for its core Millennial audience. Traditional platforms excel at providing structured information: business hours, reviews, directions, and contact details. Snap Map, on the other hand, competes by offering something more experiential and immediate.

Instead of searching for a restaurant’s phone number, a Millennial user might check Snap Map’s heatmaps to see what’s buzzing in a particular neighborhood, or tap on a location to see real-time Snaps from inside a bar or event. This provides an authentic, user-generated glimpse into the current vibe, crowd, and activities happening at a place – information that static listings often can’t convey. It’s less about finding information about a place and more about experiencing what’s happening there right now.

For a generation that values experiences and peer validation, seeing friends or other users actively sharing moments from a location can be a powerful motivator to visit. Snap Map’s local discovery is inherently social and visually driven, offering a complementary, and sometimes even preferred, way for Millennials to decide where to go and what to do, especially for social outings and events. While it won’t replace the need for navigation or detailed business info, Snap Map carves out its niche by offering a dynamic, community-powered lens on the local world, posing a subtle, experience-focused challenge to traditional local search paradigms.

While it may fly under the radar for those outside the typical Snapchat demographic, Snap Map’s ability to blend real-time location data with social interaction and content discovery has made it a sticky, successful feature that significantly contributes to the overall appeal and utility of the Snapchat platform for millions of Millennials worldwide. It’s a prime example of how location services, when designed with a specific user base and their social behaviours in mind, can become incredibly compelling.

For people of my generation the term Splat will always be associated with the Saturday Morning Children’s TV show ‘Tiswas” and was to the result of someone getting a “custard pie” in their face…

Now the hot topic in 3D Geospatial rendering are Gaussian Splats..

In Geospatial 3D, Gaussian splats are used as for rendering and representing surfaces, particularly in point-based rendering techniques. Unlike traditional polygon-based methods, which use vertices and edges to define a surface, Gaussian splats represent the surface as a collection of points or particles, each with an associated Gaussian function.

This function is typically a bell-shaped curve, which spreads out in space, creating a smooth, continuous representation of the surface.

The advantage of Gaussian splats is that they can represent complex, organic shapes and soft surfaces without the need for explicit mesh structures, as a result this is an alternative to most of the 3D techniques used in Geospatial Technology so far..

Moreover, Gaussian splats are valuable in managing large datasets, such as those found when working with large point clouds. Since each splat is a smooth approximation of the surface at that point, it allows for efficient rendering, even with millions of points.

Overall, Gaussian splats provide a flexible and efficient approach to 3D rendering, offering smooth approximations of surfaces with minimal computational overhead.

To try out both the capture and rendering abilities of Gaussian Splats I used the recently developed iPhone App Scaniverse from Niantic (yes the Pokemon GO people!) on a riverside walk to capture a large sculpture at Runnymede on the banks of the River Thames called “The Jurors”. The scan using my iPhone took about 5 minutes and the results are very impressive.

This is still very early days for the technology, but it’s nice to see some innovation in the 3D visualisation space, Niantic have a vision to map the world using Scaniverse, that’s quite a challenge but then so was Google Earth !

It was a great pleasure to spend half a hour talking with my friend Jerry Brotton on his excellent History of Maps podcast, “What’s Your Map“. This excellent podcast series cover the history of cartography by getting people to select a single map to place it’s significance in the development of mapping.

The podcast is supported by the wonderful Oculi Mundi a digital heritage website which is the home of The Sunderland Collection of world maps, celestial maps, atlases, globes and books of knowledge.

My map is he ‘Christian Knight Map’, produced by Jodocus Hondius in 1597, and the first map to use Mercator’s Projection after the death of its inventor, Gerard Mercator.

Now today the Mercator projection is the subject of much criticism, mostly as a result of it’s misuse and no doubt the impact of a episode of the much loved “West Wing” TV show which featured a group of cartographers lobbying the President to ban it’s use !

I make the point in the podcast that beyond any issues of “Social Equality” the projection certainly had many advantages in the early days of web mapping when the size of Greenland did not seem to matter…

Well…