Rev Lebaredian, in Washington speaking at the Special Competitive Studies Project’s Robotics Summit, stopped by the Washington AI Network to join host Tammy Haddad for a wide ranging interview on NVIDIA’s role in building robot brains, the industry’s race toward physical AI, competitiveness in the AI industry, and his journey from creating computer graphics for Hollywood films to leading AI-driven robotics technology at NVIDIA.
“With recent advancements in AI and particularly generative AI, transformers, LLMs, we now have a line of sight to building good robot brains that are general purpose,” said Lebaredian while explaining the process for building robot brains. “You can just have one human instruct the robot on what to do, and that robot will be able to see through all of its sensors and understand what it’s seeing, and create a plan to go execute on the task that you’ve given it.”
When asked about U.S. Competitiveness in AI and robotics, Lebaredian explained NVIDIA is “literally the only AI company that works with every other AI company… We’ve been very, very careful and deliberate about what we go work on. We go work on the things that nobody else is working on… so we can go in there and be the first.”
Lebaredian went on to explain how NVIDIA was able to put computing capability into gaming and consumer chips: “It’s Jensen, our CEO. He had the foresight, looking at the fundamental dynamics of computing and the markets, that basically every time we’ve seen more computing capabilities, more compute available for the same amount of money, same amount of energy, new applications were discovered. We find important things to do with this computing… But with Moore’s law ending, that was no longer going to be true… So we started designing a new kind of computer, what we call, accelerated computers or accelerated computing, taking the processors and the type of computing we did for this one purpose, computer graphics and rendering, and generalizing it so you can use it for many other kinds of problems like physics simulations and things that require really, really big computers. We put that capability into our install base, which was largely gaming computers, and waited many, many years until somebody figured out what to do with it.”
Lebaredian also discussed the recent wave of humanoid robots: “If we’re going to build a general purpose robot that can do many different kinds of tasks and do it with us, assist us, assist actual humans, they’re going to have to operate in spaces we’ve already built. And we’ve built those spaces for humans, factories, warehouses, hospitals, office spaces, nurseries, kitchens. They’re all designed around humans… The other advantage with humanoid robots is as we build the robot brains, we want to give them examples of the right way to behave inside the world around us, and we have lots of examples of humans doing things in these spaces. We don’t have examples of seven armed, wheeled robots, creatures running around for them to learn from.”
“In order to create AIs, you need massive amounts of data,” explained Lebaredian, while discussing robot brains. “In order to create this data, the only path we see to create enough of the data and do it in a safe and efficient way is to simulate these worlds and generate the data. In order to simulate those worlds, we have to actually create some virtual worlds that match the physical world as closely as possible.”
When asked about Elon Musk’s Tesla’s humanoid robots: “That’s the first time we’ve seen that many humanoid robots intermingling with that many humans in this unscripted environment where anything could happen… the fact that we saw that means that the technology has actually gotten to the point where we’re comfortable having these robots intermingle with the humans.”
What is NVIDIA’s role in the creation of robots? “As NVIDIA, we actually don’t feel that we’re competing with anyone directly in this space, because we’re not building everything about all of these robots. We don’t build the mechanical robots. We don’t build even the AI brains for the specialized applications. That’s not our position in all of this. What NVIDIA does and NVIDIA does well is we build computers. We build the computing systems, and that’s a special kind of computer that we call an accelerated computer. That’s a combination of hardware, chips, systems design, as well as software that reaches all the way into the application layer of the stack. That’s where our focus is… We need others, other partners to go build the physical robots, to go build the robot brains, to build the systems, to manage fleets of these robots and deploy them and service them, and specialize them for all the different industries and different domains.”
On Sam Altman’s prediction of superintelligence in 1000 days: “I don’t know what that means exactly when people use these terms like super intelligence or AGI, the artificial general intelligence. It’s really hard to even define what intelligence means…Intelligence is generally the things that is a main component of things that we build, systems that we build, that do interesting things. And we can benchmark some of those things, those skills, against humans. In many respects, we already have super intelligence in specific areas.”
Listen to the full episode on Audioboom, Apple Podcasts, or wherever you get your podcast