The robots in my building are multiplying. It started with one roughly the size of a doghouse that cleans the floors, and not very well — a commercial-grade Roomba that talks to you if you get in its way. Somehow, I’m always in its way.
My landlord was clearly excited about the new, technical marvel of an addition to the building, which takes up half the size of a New York City block. There are plenty of floors to clean and human hours of labor to save. Then my landlord told me the robot, which had been confined to the lobby, could now wirelessly connect to the elevator and control it. The robot now rides up and down all day, exiting the elevator to clean each floor’s hallway. The landlord, pleased with this new complexity, got two more, bigger robots to complete the fleet. In the spring, he told me with a straight face, there would be drones to clean the windows. I fully expect to see them as soon as Daylight Savings Time kicks in.
If you believe the press releases, we’re about to start seeing more robots everywhere — and not just doghouse-sized Roombas. Humanoid robots are on track to be a $200 billion industry by 2035 “under the most optimistic scenarios,” according to a new report from Barclays Research. The cost of the hardware needed to give robots powerful arms and legs has plummeted in the last decade, and the AI boom is giving investors hope that powerful brains will soon follow. That’s why you’re now hearing about consumer-grade humanoids like the 1X Neo and the Figure 03, which are designed to be robot butlers.
The full picture of what humanoids can do is more complicated, however. As James Vincent explained in Harper’s Magazine last month, the promises robotics startups are making often don’t line up with the reality of the technology. I’ve been learning this firsthand as I work on a feature of my own about embodied AI, which recently took me inside a number of labs at MIT. (Stay tuned for that in the coming weeks.)
One of the robots I saw there was the 4-foot-tall Unitree G1, which can dance and do backflips. It’s like a mini Atlas, the humanoid robot built by Boston Dynamics that you’ve probably seen on YouTube, but made in China for a fraction of the price. Will Knight recently profiled Unitree for Wired and argued that China, not the United States, is poised to lead the robot revolution on the back of its cheap hardware and ability to iterate on new designs. Still, a dancing robot is not necessarily an intelligent one.
The geopolitical pieces of the puzzle
If you haven’t heard of a “thing biography,” you’ve definitely come across one of the books. Mauve: How One Man Invented a Colour That Changed the World by Simon Garfield is sometimes credited as the accidental original example of the genre. Cod: A Biography of the Fish That Changed the World is the book that turned me onto it, when it became a bestseller nearly 30 years ago. You can now read thing biographies, also known as microhistories, about bananas, wood, rope — really any thing has a fascinating history that you may find sitting on a shelf at an airport bookshop. (Slate’s Decoder Ring podcast has a great episode explaining the phenomenon.)
What makes these books especially fun is that they’re not at all about the things themselves. They’re about us. The history of cod is really about what the fish tells us about exploration and human ingenuity. One of my favorites from the genre is The World in a Grain: The Story of Sand and How It Transformed Civilization. It is nearly 300 pages about sand, which is in fact what everything important, from concrete to microchips, is made of. And we’re running out of it.
AI is inherently physical, because it needs hardware to exist. And I’m not just talking about the actuators, motors, and sensors that make machines move. The high-powered Nvidia chips that promise to provide the processing power needed to provide dumb backflipping robots with a brain that can turn them into general-purpose appliances? They’re made of sand. It’s really good sand, of course — sand that’s been purified and processed in some of the most advanced manufacturing facilities humankind has ever built. But as the conversation around advanced hardware powered by even more advanced software is changing our relationship with technology, I find it grounding to know that we’re dealing with familiar ingredients.
If you think that sitting around reading books about sand is too escapist, let me offer a compromise. For a dose of reality, you should check out Chip War: The Fight for the World’s Most Critical Technology by Chris Miller. It’s also about sand, but it’s specifically about the history of semiconductors in the United States and the arms race it eventually kicked off with China. As the Trump administration inches closer to attempting to seize Greenland, many are left to worry that China’s Xi Jinping will invade Taiwan and take control of its advanced chipmaking facilities. If China cuts off Taiwan, which produces 90 percent of the advanced chips needed for AI applications, the digital economy would grind to a halt, according to my Vox colleague Joshua Keating. China wouldn’t just lead the robot revolution. It would own it.
Trial and failure
The robots in my building, I’m guessing, weigh about 120 pounds apiece. It’s an informed guess, because I’ve had to pick them up to move them out of my way. If you move too quickly or intimidate them too much — not that I’ve done this on purpose — they freeze. As a safety feature, this is great. But the other day, I was getting on the elevator, freaked out a robot, and the elevator wouldn’t move. I took the stairs.
In a sense, though, these failures are essential. Every couple of weeks, I see a technician come and work on the robots. They might be replacing a part, updating its software, or just giving them a pep talk. It’s a reminder that inching toward a future in which embodied AI, probably robots, helps us unlock humanity’s greatest potential is a process, and probably a long one.
Many people credit Elon Musk with starting the race to build a general-purpose humanoid, when he announced Tesla’s effort to do so back in 2021. Musk has shown off various prototypes of the Tesla humanoid, Optimus, in the years since then. Many of them are just puppets, operated by employees behind the scenes. This week, Musk admitted that manufacturing the humanoids would be “agonizingly slow” before it hopefully got faster. I truly wonder, what’s the rush?
A version of this story was also published in the User Friendly newsletter. Sign up here so you don’t miss the next one!
