World Models, Spatial Computing, and What ChatGPT Can't Do

[Reminder! Sunday, Feb 15th, 2:00 to 4:00, Meow Wolf & Future of 3D Immersive Arts, at Boxing Bear Brewing, Bridges on Tramway Taproom. Look for Meow Wolf shirt...]

"ChatGPT, while proficient in language, struggles with spatial reasoning and lacks a "world model" (an internal understanding of physics, geometry, and cause-and-effect), leading to hallucinations. World models are emerging to address this by training on video/3D data, powering spatial computing in robotics, AR, and navigation.

What's Wrong with ChatGPT (Limitations):

Lack of Physical Understanding: ChatGPT does not have a body and struggles to understand "affordances" (e.g., how objects fit together or interact), making it poor at spatial reasoning and navigation tasks.

Hallucination in Spatial Tasks: It struggles with tasks requiring spatial logic, such as navigating directions or interpreting simple, novel diagrams.

Poor Math Skills: Despite language proficiency, it often fails at complex mathematical calculations and reasoning.

Text-Centric Limitation: Trained on text, it cannot understand the visual, physical, and 3D data required for real-world interaction.

World Models and Spatial Computing (The Next Step):

What are World Models? They are AI models that simulate the physical world—understanding physics,, and spatial dynamics, not just predicting the next word.

The Data Shift: Instead of text, these models train on video, 3D point clouds, and sensor data to predict how scenes change, which is vital for robotics and autonomous vehicles.

Spatial Computing Integration: World models are foundational to spatial computing (e.g., AR/VR), allowing AI to understand and operate within 3D environments, blending digital content with the physical world." (AI Overview: Gemini AI)

On April 1st, I am teaching the class:

ChatGPT for Dummies & Deep Thinkers (class full, but join waitlist!)

You may be curious about what is perhaps the most well-known AI tool at the moment, ChatGPT. Whether you want to use it to draft limericks, create amusing images, read tarot cards, summarize boring reports, act as your personal editor, or just pass the time, this class explores how today’s generative AI works, and why everyone’s talking about it. This talk balances fun experiments with serious discussion: no tech background required, just curiosity and a sense of humor. Perfect for skeptics, dabblers, and deep thinkers alike!

In August, I am teaching Beyond ChatGPT: Exploring the Creative AI Universe (no link yet)

ChatGPT is just the beginning. Let's explore the wider world of generative AI, tools that create images, music, voices, video, and even computer code. We’ll compare how different AIs “create,” what they do well (and poorly), and what this means for artists, writers, programmers, and everyday curious minds. Expect demonstrations, discussion, and thoughtful reflection. No technical background required, just curiosity, creativity, and a willingness to be amazed (and occasionally puzzled).

The first class focuses primarily on what ChatGPT can do. But I'll touch on what it doesn't do well (or, practically, at all). I'll expand that coverage during the second class. A key limitation will be ChatGPT's lack of spatial intelligence.

Read the following and come prepared to discuss (or listen to others who came prepared to discuss!)

From Words to Worlds: Spatial Intelligence is AI’s Next Frontier (Dr. Fei-Fei Li)
The Worlds I See: Curiosity, Detours, and Discovery at the Dawn of AI (Review) (Chuck Webster)
TIME’s Person of the Year, the "Architects of AI,” Meets My Mirehaven Backyard! (Chuck Webster)

Note: This is not a traditional presentation (slides, presentation, etc.), although there may be handouts. The format is that of a conversation over beverages of your choice.