Silicon Valley is buzzing with new terminology, but there is one concept that every decision-maker truly needs to understand: the World Model.
Why? Because it marks the shift from 'AI that writes a nice email' to 'AI that understands how physical reality works'.
It is the reason why Pim de Witte (opent in nieuw venster), the 30-year-old Dutch founder of Medal.tv (opent in nieuw venster), reportedly turned down a $500 million acquisition offer. And it is the fundamental reason why I have long argued that we should by no means concede the AI race to OpenAI just yet.
Let us strip away the hype and examine both the technology and the business case.
The Librarian vs. The Pilot
To understand the difference between what we currently have (LLMs such as ChatGPT) and where we are heading, I like to use the following analogy:
- Current AI (LLM) is a librarian. He has read every book on aviation. He can explain perfectly how an aeroplane works. But put him in the cockpit, and he crashes. He knows the theory (text), but not the practice.
- The World Model is a pilot. He has spent thousands of hours in a flight simulator. He 'feels' what happens when you pull the control stick. He understands cause and effect in three-dimensional space.
Why Video is the New Gold
If you want to teach an AI how the world works (opent in nieuw venster) (gravity, object permanence), text alone is insufficient. You need video. A world model learns the 'laws' of our reality through observation.
This brings us to Pim de Witte. His platform, Medal.tv, is the largest repository of gaming clips. Games are, in essence, simulations. For training a World Model, this represents 'clean' data of inestimable value. Pim recognises that he owns the gymnasium where the next generation of AI models will be trained.
Why Google is the 'Sleeping Giant'
This insight — that video and simulation are the keys to intelligence — is precisely why I firmly believe in Google's leading position over the long term.
Everyone is watching OpenAI, but what is their data? Primarily text from the internet. Now look at Google. They hold an uncontested monopoly on the raw materials for world models:
- YouTube: Billions of hours of video enabling deep analysis of human interaction and physics.
- Waymo: Their self-driving vehicles have been gathering data on physical roads, unexpected situations, and spatial understanding for years.
- Android & Chrome: An endless stream of user context.
OpenAI must generate video (Sora) or purchase data (potentially from Medal.tv) to learn to understand the world. Google already owns the global library of video and movement. They are slow, but their foundation is superior.
The Pragmatic Perspective: What Does This Mean for You?
The implications for your business are closer than you might think.
- Robotics & Logistics: Current robots execute scripts. World Models enable robots that 'see' and anticipate. If a box falls, the robot understands why.
- Your Data Strategy: Many companies focus on digitising documents. But as we move towards world models, the value lies in operational data. Camera footage from your production line. Telemetry from your vehicle fleet.
Conclusion
The dominance of the chatbot is temporary. The future belongs to systems that can not only talk about the world, but also act within it. The parties with the most 'real' data — whether that is a 'startup' like Medal.tv or a giant like Google — hold the winning hand.