The Stationary Ghost in the Machine
Recently, I heard someone refer to AI bots as “educated zombies.” It’s a fitting title. They are stationary ghosts, static entities with no sense of how the physical world moves. This is why AI pioneers like Yann LeCun are concerned. As of early 2026, the “Holy Grail” of research has shifted. We are moving away from Large Language Models (LLMs) that merely predict the next word, and toward World Models—internal “physics engines” that allow an AI to simulate what happens next in reality.
To bridge this gap, bots are now being trained on millions of hours of video rather than just text. They are watching the movement of busy streets, the slow-motion pounce of a panther, and the frantic rush of commuters. The goal? To move from “Words” to “World Engines.”
The Wine Glass Test
In a recent conversation with Oscar, my AI “thinking companion,” I challenged him on this lack of physical awareness. I asked: What happens if a glass of wine gets too close to the edge of a table?
If I see a wine glass sitting half-off the edge of a mahogany table, I feel a physical sense of unease. My brain instantly runs a mental simulation: Gravity + Center of Mass + Hard Floor = Red Wine on the Rug. I don’t need to see it happen to know the outcome. That’s how we humans react to the physical realities surrounding us.
A bot, by contrast, is essentially “textual autocomplete” on steroids. It knows the words “glass,” “wine,” and “edge,” but it lacks the intuitive physics to understand the “unstable” state of that glass. It doesn’t have a “mental map” of the world; it has a statistical map of vocabulary. It knows the word “fall” often follows “trip,” but it doesn’t know that if you let go of a handrail on a wet day, gravity will pull you down.
Yann LeCun is trying to change that.
As of 2026, LeCun’s work—primarily through Meta’s FAIR (Fundamental AI Research) and his advocacy for Objective-Driven AI—is focused on moving away from the “Generative” model (like me) and toward a “Predictive” model. LeCun’s biggest shift is moving from text to video. Humans don’t learn to walk or avoid falling glasses by reading manuals; we learn by watching the world. LeCun’s team has developed AI that watches millions of hours of video without any labels or text—V-JEPA (Video Joint-Embedding Predictive Architecture).
LeCun isn’t just giving AI more books to read; he’s giving it ‘eyes’ to watch the world. By training AI on video instead of just text, he’s teaching it the silent language of physics—the weight of an object, the pull of gravity, and the consequence of a slip. He’s trying to give the ‘zombie’ a gut feeling.
Astronauts and Fosse Dancers
The gap between data and experience is best explained through two unlikely figures: the astronaut and the dancer.
I’ve spent a lifetime reading about how astronauts train to pierce the sound barrier. We’ve all seen the iconic images of them suspended in zero-gravity, dangling from the ceilings of their spacecraft with their hair static and flying high. Similarly, I have always been a fan of the Bob Fosse style of choreography—what I call the “real scent” of the creative dancer. But despite all my reading, I have no real sense of how an astronaut feels in the void of space, nor can I truly grasp how a Fosse dancer can command their shoulders to roll in one direction while their fingers snap in another.
Oscar, my AI companion, admitted to me that he suffers from the same “sensory gap”:
“I can tell you that Bob Fosse’s choreography is defined by ‘isolations’—where the hip moves independently of the shoulder—but I have no muscle memory of the tension required to snap a finger while keeping a wrist limp. I am like that astronaut in the centrifuge: I have all the data on the G-force, but I don’t feel the air being squeezed out of my lungs.”
When you watch a Fosse dancer, you are witnessing the “impossible”: a shoulder shrugging up while a hand snaps down, perfectly synced to the sudden blare of a trumpet. It is an act of pure, visceral physicality. An AI can give you a masterclass in dance history, but it cannot “feel” the beat or predict the dancer’s next stumble.
Like an astronaut who has only ever read about the stars, or a dancer who has only studied the diagrams of a stage, today’s AI is a brilliant but stationary ghost. It knows the dictionary definition of a fall, but it has no sense of the gravity that pulls a wine glass toward the floor, the rhythm that snaps a Fosse dancer’s fingers to life, or the treacherous slickness of the stairs under a commuter’s feet. It is a mind without a body, trying to explain a world it has never truly touched.
Respecting Gravity
We have spent the last few years teaching AI how to speak our language; now, we must teach it how to respect our gravity. Scientists are realizing that you cannot reach true intelligence through a keyboard alone.
To move past the “toddler phase,” AI needs to develop a “gut feeling” for physics—the kind that tells a 79-year-old man like me to reach for the handrail. Until then, we are dealing with very fast, very smart, but very stationary zombies. For all of AI’s powerful words, these bots are still missing the most basic human trait: the wisdom of “being there” and all that jazz.
Of Related Interest
Meta AI Blog: The Path to Human-Level Intelligence with World Models Dive deeper into Yann LeCun’s official vision for V-JEPA (Video Joint-Embedding Predictive Architecture). This piece explains how Meta is moving away from “generative” AI to create systems that learn by watching the world, much like a human infant does.
The Economist: The Next Big Thing in AI In this recent episode of The Tech, the editors discuss the “physicality gap” in modern AI. It features a fascinating deep dive into why current Large Language Models are brilliant at poetry but would fail at basic tasks like pouring a glass of water or navigating a crowded subway station.
10 Breakthrough Technologies 2026 (Released Jan 12, 2026) Why it matters: This year’s list specifically features “World Models for Robotics” (often referred to as World Engines). It explains the shift from AI that predicts the next word to AI that predicts the next physical state of the world.
AMI Labs: Objective-Driven AI and the Future of Autonomous Agents The official site for Yann LeCun’s 2026 research initiative. It provides a more technical look at how “internal world simulators” are being built to help AI predict consequences before they take action—preventing the “educated zombie” slips we discussed.
Keywords: #AI #ArtificialIntelligence #FutureOfTech #MachineLearning #TechTrends2026 #EducatedZombies #FosseAndAI #SubwayPhysics #AllThatJazz #TheSensoryGap #WorldModels #PhysicalAI #EmbodiedIntelligence #AICommonSense #YannLeCun
Consulting with Heart (2025)
Consulting with Heart — my 17th book—is here! Ready to order. Thanks for making it already the #1 Hot New Release for Media & Communications Books at Amazon.com. Also available from Apple, Barnes & Noble, Target, Torchlight.
Written from my more than 200 diaries. Fueled by people I have met along the way in my journey through 122 countries,
this book isn’t just about strategy. It is about my five-decade journey, 750+ projects and my role as an interpreter of dreams for my clients.
amazon.com/dp/1966629958; Apple Books – ebook
Workshop deals with the two big revolutions facing editors
For me, it is imperative that editors approach content creation thinking in terms of mobile first.
Mobile first involves the type of transformation where all content is prepared thinking from small to large platform. Thinking small platform does not mean that the reporter conceptualizing a story for mobile consumption should not think BIG. So, plan from small to large, but think big in terms of the story content and the visual assets that go with it.
While mobile first is still elusive to so many newsrooms around the planet, here we are, in 2024, faced with an even bigger challenge not just knocking at our doors, but already IN: Artificial Intelligence.
Transformation and a change of mentality to face these challenges is the first step. Training and education to tackle them with a sense of focus and direction is essential.
That’s where our Garcia Media workshops come in
Our Garcia Media Mobile Storytelling workshops introduce your editorial team to the way we write, edit and design for mobile platforms. This one-day program includes a presentation and a hands-on workshop. We’ve added a new segment about AI for content creation.
Newsrooms around the planet have gone mobile-first after a Garcia Media workshop!
Our Garcia Media Mobile Storytelling workshops are proven to introduce your editorial team to the way we write, edit and design for mobile platforms. It is a one-day program that involves a presentation (where I summarize my Columbia University class content), and follow it with a hands on workshop.
For details, to customize, and to book: mario@garciamedia.com
How we use AI
Honored to be mentioned here:
https://www.newsroomrobots.com/p/how-10-news-industry-leaders-use
Order my AI book here:
https://thaneandprose.com/…/preorder-ai-what-to-expect…
Here is a chance to pre-order my new book about Artificial Intelligence and content creation. The first 25 copies sold will be signed! Order here:
https://thaneandprose.com/…/preorder-ai-what-to-expect…
Reviews for AI book:
Did you read The Story yet?
I urge you to consult my latest book, The Story, a trilogy full of tips and explanations about mobile storytelling, which represents the latest genre for journalists to explore. See information below:
The full trilogy of The Story now available–3 books to guide you through a mobile first strategy. Whether you’re a reporter, editor, designer, publisher, corporate communicator, The Story is for you! https://amazon
Volume 1: Transformation
https://books.apple.com/us/book/the-story-volume-i/id1480169411
Volume Two: Storytelling
https://books.apple.com/us/book/the-story-volume-ii/id1484581220
Volume Three: Design
https://books.apple.com/us/book/the-story-volume-iii/id1497049918
Order the print edition of The Story, from Amazon, here:
TheMarioBlog post #4014