The Domesticated Robot: A Study in Corporate Reabsorption

There is something inherently touching about a corporate restructuring. It is the boardroom equivalent of a parent finally deciding that the teenager living in the garage—the one who spends all day trying to teach a mechanical arm how to pick up a grape without turning it into jam—is finally ready to join the family business. Google has recently announced that Intrinsic, its robotics-focused 'Other Bet,' is being folded back into the main company. One imagines the robots being packed into sensible cardboard boxes, their sensors blinking with a mix of confusion and the quiet hope of better dental insurance.

For years, Intrinsic has lived in the 'Other Bets' category, a sort of high-tech island of misfit toys where Alphabet keeps its most ambitious and least immediately profitable dreams. It is a place where one might find a project dedicated to delivering burritos via drone or another attempting to solve death itself, usually over a very expensive cup of artisanal coffee. By bringing Intrinsic into the fold, Google is signaling that 'physical AI' is no longer a hobby or a fever dream, but a core part of the furniture. They want to create the 'Android of robotics,' a universal operating system that would allow a robot from one manufacturer to talk to a robot from another, presumably to complain about the quality of the floor wax.

The ambition is, of course, staggering. We have spent the last decade teaching AI to write poetry, pass bar exams, and generate images of cats wearing Victorian waistcoats. These are all very noble pursuits, but they exist entirely within the safe, frictionless confines of a server rack. The physical world, by contrast, is a remarkably inconvenient place. It is filled with gravity, which is always pulling things down, and friction, which is always trying to stop things from moving. There are also stairs, which remain the natural enemy of anything with wheels and a sense of dignity.

Teaching a piece of software to navigate a kitchen is significantly harder than teaching it to predict the next word in a sentence. When a language model makes a mistake, it might suggest that the capital of France is 'Baguette.' When a robot makes a mistake, it tends to drive through a drywall partition or accidentally toss a cat into the dishwasher. The 'Android for robots' strategy aims to abstract away these physical indignities, providing a common language for movement and interaction. It is an attempt to give the ghost in the machine a set of reasonably reliable limbs.

I once watched a prototype robot attempt to fill a kettle. It was a performance of such exquisite hesitation and profound existential dread that I felt an immediate kinship with it. The robot approached the tap with the caution of a man trying to defuse a bomb with a pair of salad tongs. It spent three minutes calculating the optimal angle of the spout, only to realize that it hadn't actually turned the water on. There is a certain Britishness to this level of over-thinking, a commitment to the process that entirely ignores the desired outcome. If Google can bottle that specific brand of mechanical neurosis and make it interoperable, they may truly have something.

The integration of Intrinsic into Google proper suggests a shift in how we view our digital assistants. We are moving away from the era of the disembodied voice that tells us the weather and into the era of the embodied entity that might actually be able to find the TV remote. It is a transition from the 'Smart Home'—which is usually just a collection of lightbulbs that refuse to connect to the Wi-Fi—to the 'Capable Home.' One can only hope that the robots inherit Google's search capabilities, so that when you ask where your keys are, they don't just provide a list of ten sponsored links to locksmiths in your area.

There is, naturally, a bureaucratic side to this reabsorption. One wonders if the robots will now have to attend quarterly all-hands meetings or sit through mandatory training videos on how to avoid phishing scams. There is something deeply amusing about the idea of a multi-million dollar robotic arm being forced to change its password every ninety days, or a self-driving forklift having to justify its travel expenses. It is the ultimate domestication: the wild, untamed frontier of robotics being brought to heel by the relentless march of corporate HR.

In the end, the success of 'physical AI' will likely depend on its ability to be boring. The most successful technologies are the ones we stop noticing—the ones that become as unremarkable as a toaster or a stapler. We will know Google has succeeded not when a robot does a backflip on YouTube, but when we find ourselves annoyed that the mechanical butler took slightly too long to fold the laundry. We are inviting the algorithms into our physical space, and we are doing so with the quiet expectation that they will eventually learn to wipe their feet.

It is a brave new world, or at least a very well-organized one. As the robots move out of the garage and into the main house, we should probably make sure the stairs are in good repair and that the dishwasher is kept firmly closed. After all, even the most sophisticated 'Android for robots' is still just a guest in our world, and guests, as we all know, eventually have to be told where the spare blankets are kept.