Nvidia CEO Jensen Huang believes the next major breakthrough in artificial intelligence (AI) will take the technology beyond computer screens and into the physical world. He calls it Physical AI, a technology that could allow machines to understand their surroundings, make decisions, and perform tasks without requiring instructions for every movement.
How is it different from the AI we already use? ChatGPT can respond to queries, write emails, and analyze information.
Physical AI, however, might be able to pick up a cup, understand it is fragile, and place it down on a table. The technology fuses AI with cameras, sensors, and robotic systems to enable machines to respond to real-world scenarios.
It is not just for humanoid robots. It could be used for self-driving cars to predict when a pedestrian is about to walk across the street, or for automated robots in a warehouse to navigate around humans and pick up items without the need for predetermined routes.
Other industries investigating its use include hospitals and factories. A major challenge, however, is teaching machines the workings of the real world.
A chatbot can provide a wrong answer, but a robot making a similar mistake could cause damage to equipment or injury to a person. To correct this, Nvidia is working on simulation technology that lets robots practice in a virtual environment before operating in real life.
For instance, its Cosmos AI models are capable of creating realistic scenarios to train machines on the movement and interaction of objects. Huang said that while there are technical, safety, and cost limitations before robots can be readily adopted, they are now going through their own 'ChatGPT phase.' For the average person, the distinction may become as straightforward as asking an AI to perform a task, rather than asking how to accomplish it.