TL;DR: From Static Maps to Decision Engines
The geospatial industry is hitting a major inflection point. We are moving away from Mapping 2.0—the pursuit of the perfect, high-fidelity “Ground Truth” reflection of the past—and entering the era of Decision 1.0.
While industry veterans like Ed Parsons and Sean Gorman have recently highlighted how “maps are beginning to dream” through probabilistic simulation and generative reconstruction, this shift represents more than just a technical upgrade. It is a transition from measurement to simulation.
- The Surveyor’s Monopoly is Ending: Sparse, cheap data can now be “inferred” into high-accuracy 3D models, commoditizing traditional data collection.
- Precision vs. Timing: In a world that doesn’t sit still, a “reasonably accurate” simulation today is more valuable for taking action than a 100% accurate map that arrives tomorrow.
- The Rise of World Models: Unlike LLMs that predict the next word, World Models are digital sandboxes that predict the next moment in physical reality. They allow us to run “what-if” scenarios—moving from “Where is my asset?” to “What happens to my strategy if this variable breaks?”
The goal is no longer just to polish a mirror of reality, but to build a Reasoning Engine that drives Decision Velocity—the only remaining competitive advantage in an AI-augmented world.
I mentioned to my friend Asaf that I had thought I was alone in the geospatial world discussing world models. This week two geospatial industry veterans – Ed Parsons and Sean Gorman – both of whom I greatly respect, have weighed in on the topic.
It is nice to no longer be alone.
But more than that it is fantastic that geospatial veterans like Ed and Sean are now converging on an area I have been discussing for some time. An area I believe will be a huge disrupter to the established geospatial industry.
The Convergence of the Giants
It is a significant moment. Ed, formerly of Google, has provided a profound philosophical look at how the “Map is beginning to Dream“, moving us from a century of static measurement to an era of probabilistic simulation. He has, in my view, correctly identified that the “Ground Truth” we once treated as an absolute is shifting beneath our feet.
Meanwhile, Sean has grounded this vision in technical reality. By highlighting tools like JEPA (see Yann LeCun and AMI). He has suggested that we no longer need to see every brick to understand a building. The experiments he discusses demonstrate that we can now infer or “generatively reconstruct” the world with high accuracy using sparse, cheap data.
Together, they have effectively signaled the end of the “Surveyor’s Monopoly.”
Moving the Conversation Forward
I believe that Sean has validated that Ed’s vision is technically feasible. But ..
They are both still focused on Mapping 2.0 – the passive layer.
My focus is on Decision 1.0 – the active layer.
This is the paradigm shift, and truly a huge mental jump for those of us with long histories embedded in the old geospatial orthodoxy. Ed suggests this in his piece – geospatial is a horizontal tool whereas geography is how we understand the world. I have long argued that the former is a product, whereas the latter is the cognitive framework. And, as we will discuss, world models align beautifully with that framework.
For decades, the mapping world has treated “Ground Truth” as the holy grail. Ed touches on this, and on the surface, it sounds right: a map is only useful if it’s accurate. But let’s be brutally honest, “Ground Truth” can be seen as just a fancy term for a perfect picture of the past.
Now don’t get me wrong, if you are building a bridge, you need that precision. But the GIS industry has forced this high-bar accuracy onto everything, turning it into … a convenient myth.
The reality? The world doesn’t sit still for a photo.
While the ‘old guard’ – the long established geospatial order – have been busy polishing a perfect reflection of what was, they missed the shift. We don’t need a flawless record of yesterday; we need a high-speed simulation of now and a “what-if” for tomorrow. This sets the stage nicely for the core thrust of this article, but before we proceed let’s dive into that incredibly important word simulation.
The Three Levels of Simulation
So .. what does that word simulation mean?
In the geospatial world, ‘simulation’ is used in three very different ways, summarized in the chart below:
Level 1 is simply a mirror of reality. It is the current GIS shift – moving from a focus on 2D maps to building 3D digital twins. This remains largely an historic data exercise showing how the world ‘used to look’. It is worth noting there is a push currently to build 3D maps on the fly from real or near real-time data; some are calling these fast maps. Here and Google have focus in this area. This closes the historic data temporal gap.
Level 2 is centred on how we construct digital twins. In this case AI builds a mental model of the building’s three-dimensional logic. So what cannot be seen is inferred (guessed). These 3D structures can then populate that 3D scene. Ed and Sean are excited about this level since it commoditizes the data – if I can “infer” the building, I don’t need a $100k LiDAR scan. This is the “commoditization of sight”. But my view of this – commoditizing the data in this way is just a race to the bottom for the data companies.
But a 3D digital twin, however it is reconstructed, is still just a static asset.
Level 3 is my focus and is truly the paradigm shift. In my view this is where the next phase ‘rubber really hits the road’ sits. This isn’t about looking at a pretty 3D picture; it’s about breaking things in a sandbox to see what happens.
This is about what-if scenarios.
Think about a reconstructed 3D scene. That is a pretty picture of today or yesterday’s reality. But suppose we could take that scene and run a series of “what-if” scenarios against it? We move from a visualization to a decision-making foundation.
Think about asking this question: “Where is my warehouse and what does it look like?” versus “What happens to my bottom line if that warehouse floods at 3:00 AM?”
That is Decision Velocity.
Mapping 2.0 v Decision 1.0
Before we touch on my advance into decision velocity, let us close out mapping 2.0. I disagree with a key part of Ed’s perspective, I believe:
Precision is a luxury; timing is a necessity.
This will be proven over time but picture this: A hurricane is approaching New Orleans. One emergency team is waiting for 100% map accuracy – today’s reality and a lagging indicator that arrives too late. The other team is comfortable with 85% certainty – see Sean’s mention of ‘reasonably accurate’ threshold – they don’t rely on an accurate map, they reroute the supply chain before the storm even hits – a leading indicator.
So as one team is focused on ‘high accuracy’ to document the disaster; the other is taking action to avoid it.
In my view, we need to stop polishing the mirror and start building the engine. Using world models to drive Decision Velocity means you are no longer just drawing a more expensive picture of the past, you are taking action with the help of a second brain in the room.
A Simple Guide to the World of World Models
If our current LLM-centric AI world – like ChatGPT – is a librarian that has read every book ever written, a World Model is an explorer that has walked the earth and learned how things actually move, break, and interact.
What are they?
In the simplest of terms, a World Model is a digital sandbox inside an AI’s mind. Instead of just predicting the next word in a sentence, it predicts the next moment in physical reality. It is a simulation engine that allows an AI to dream about what might happen next before it actually takes an action.
How do they work?
The secret sauce of a world model is a three-part loop that mimics how we learn as humans:
- The Vision: It squashes a chaotic scene into a simple mental map of the most important objects.
- The Memory: It uses that map to imagine the future. It asks: If I push this button, what happens to the world?.
- The Controller: It decides on the best move based on those imagined futures, and shares that with a human – the AI serving as that ‘second brain in the room’.
One Goal, Different Paths
Not all World Models are built for the same reason. The industry is currently split into three major camps:
- The Dreamers: Models like OpenAI’s Sora or Google’s Genie focus on Generative reality. They prove they understand the world by being able to recreate it in stunning, photorealistic detail.
- The Architects: This is where World Labs and Niantic Spatial, Inc. – see this fascinating talk by Brian McClendon, CTO at Niantic – sit. Unlike a fleeting dream, World Labs’ Marble lifts images into persistent 3D volumes that you can actually navigate without the scene morphing or dissolving. These models create physics-ready environments that act as a permanent bridge between a digital hallucination and the real world.
- The Thinkers: Models based on the JEPA architecture (mentioned earlier) argue that an AI doesn’t need to dream in high-definition pixels; it just needs to understand the concepts. It cares that the car is moving left, not the exact texture of the asphalt.
Why This Matters for You
This is not just a technical upgrade; its a fundamental shift in how we solve problems. While the Dreamers are revolutionizing entertainment and gaming, and the Architects are building persistent, physics-ready foundations for digital twins, the Thinkers are becoming the infrastructure for a new era of robotics and automated strategy.
However, the real magic happens when we move from just observing these models to using them to accelerate our choices. To see how these internal simulators are being used to collapse the time between seeing a problem and solving it, check out my deep dive on Decision Velocity: The Only Competitive Advantage Left.
We are moving past the era of what happened into an era of what if. You can explore the technical bridge between these dreaming maps and the engines that drive real-world action in my follow-up piece: Beyond Next-Token Prediction: How World Models and Reasoning Engines Drive Decision Velocity.
Closing Thoughts
I’ll admit to being elated that others in our wonderful world of geospatial are beginning to explore world models and how they will inevitably disrupt the established geospatial world.
In this article I hope to have taken readers further – beyond the map. Without meaning to upset my long term friends and colleagues in the industry; I do wonder whether maps, as we have long known them, have becoming simply a legacy visualization for humans.
My focus has moved away from the geospatial process and the bottleneck ridden world of actionable insight to outcomes – decision velocity and AI augmented intelligence. This is an exciting and rapidly expanding space.
Let me close by saying this. In the end, we must decide if we want to be the archivists of what was, or the architects of what’s next. The map is no longer the territory, it is just a reference point for the machine that is already moving past it.
Matt Sheehan
Matt Sheehan is a senior executive and geospatial strategist with over 25 years of industry experience. He specializes in the advance of AI from pattern matching to causation, focusing on increasing decision velocity and reducing decision latency for complex organizations. Matt bridges the gap between traditional geospatial intelligence and the emerging frontier of agentic, reasoning-based AI systems.