AI Is Learning to Understand the Physical World, Not Just Words and Images
 |
| Image: Digiopedia / Illustration |
Generative AI has made enormous progress with text, images and code.
But the real world is considerably more complicated.
A physical environment contains depth, movement, objects, people and unpredictable interactions. An AI system that operates there needs a different kind of understanding.
That is why spatial intelligence is becoming an increasingly important area of AI research.
AMD's agreement to acquire World Labs for $8.2 billion is one of the clearest recent signs of how seriously the industry is taking the field. Reuters+1
Language is only one form of intelligence
A language model can describe a room.
That does not necessarily mean it understands the room in a way a robot can use.
A physical AI system needs to understand relationships.
Where is the table?
How far away is it?
Which objects are on it?
Can the robot move around it?
What happens if another object is placed nearby?
These are spatial problems.
World models could become important
Researchers are developing systems that attempt to construct internal representations of environments.
Instead of simply recognizing individual objects, these models can represent how objects relate to each other and how environments behave.
World Labs has focused on this area.
AMD said the acquisition will strengthen its ability to develop hardware, software and systems around emerging AI models and applications. AMD Newsroom
That is important because future AI workloads may look very different from today's chatbots.
Robotics is the obvious destination
Robotics needs this technology badly.
A robot operating in a factory can work in a carefully controlled environment.
A household robot cannot.
Every home is different. Objects move. Lighting changes. People walk around. Furniture gets rearranged.
A better understanding of physical environments could make robots more adaptable.
The same technology could also be useful in autonomous vehicles, simulation, industrial design and virtual environments.
Why chip companies care
AI models determine what computers need to calculate.
If future models increasingly simulate environments and physical interactions, the hardware supporting those models will also change.
AMD's acquisition therefore represents more than an investment in another AI startup.
It is an attempt to connect AI research with the hardware required to run it. Reuters
The next major AI breakthrough may not be a model that writes better paragraphs.
It could be one that understands where things are, how they move and what is likely to happen next.
