Misemployed Android

Unitree, one of China’s biggest humanoid-robot makers, is listing in Shanghai today after its IPO was more than 8,000 times oversubscribed. Yesterday, Reuters reported that the industry is facing a less glamorous test: whether robots can actually do useful work.¹

»Perhaps the real opportunity in robotics is not to make machines fit the world we have, but to let them change the world we build«

The humanoid robot is one of the great technological fantasies of our time. Douglas Adams even imagined the Electric Monk: a machine whose job was to believe things for people who were too busy or too tired to do so themselves.² Much of today’s spectacle is similarly anthropocentric. Robots lift boxes, fight, backflip, and run. Unitree has previously unveiled a humanoid it says can reach a top speed of 12.66 metres per second – faster than Usain Bolt’s top speed during his 100-metre world record.³

But why is that impressive?

A robot that can run like a human is reproducing a solution evolution has already found. A robot that could clean a building continuously for ten years without human intervention would be doing something much stranger. Yet we instinctively regard the first as a technological achievement and the second as a question of engineering.

Robots can of course become more like us if we want them to. But the interesting question is why we keep asking machines to inhabit a world designed around the human body, rather than asking what they might become if we allowed them to be something else – and what architecture might become as a result.

A vacuum cleaner doesn’t need legs. A window-cleaning robot doesn’t need a torso. A machine designed to work in a kitchen might not need to stand on the floor at all.⁴ If we know what the machine is for, we can design its body around the task. This sounds obvious. Yet we’ve spent decades trying to make robots fit into our world. Perhaps we should do the opposite.

In AI research there’s a useful distinction between updating your model of the world and acting on the world so that it better conforms to your expectations. This is one way of thinking about Karl Friston’s idea of »active inference«.⁵ So what if we taught environments to adapt to robots?

The skyscraper offers a precedent. When the lift was invented, architects did not try to make people better at climbing stairs. They changed the building. The vertical city became possible because a new machine changed what a building could be.⁶

Robotics could do the same. Instead of making a humanoid robot capable of navigating every human space, we could make spaces capable of accommodating machines on their own terms. A cleaning machine needs somewhere to recharge, not somewhere to sleep. A window-cleaning system might live in the building’s façade. A robotic kitchen could occupy a fraction of the space of a human one. Machines could move through narrow cavities, behind walls, above ceilings or into other spaces humans cannot conveniently use.

This suggests a new kind of ergonomics: not designing machines around human spaces, but designing spaces around machine bodies. The building itself becomes an interface between different kinds of bodies. Humans inhabit the rooms. Machines inhabit the poche.⁷

Anthropologist Edward T. Hall showed how deeply social relationships can become embedded in spatial arrangements: the dimensions and position of a room can reflect who is expected to occupy it and what they are expected to do there.⁸ The old maid’s room is a particularly uncomfortable example. Its small dimensions were not arbitrary. They belonged to a particular organisation of domestic labour.

Now imagine a different kind of occupant. Not a servant, but a machine. Not a small human room, but a robotic space designed around what a robot actually needs. This is not simply about making buildings smaller. It is about changing the spatial problem rather than reproducing the human solution. Humans carry things because we have arms. Buildings have dumbwaiters. Humans climb stairs. Buildings have lifts. Humans clean windows. Buildings could clean themselves.

There are counterexamples. Folding a shirt is a difficult robotic problem precisely because fabric is soft, deformable, and unpredictable. A human hand is extraordinarily good at dealing with it. Sometimes the human body really is an excellent template. But that should be a conclusion reached by looking at the task, not the starting assumption.

The question is what shape a machine needs to perform its work, and what environment allows it to perform that work most effectively. We tend to assume that the building is fixed and the robot is the variable. But buildings have never been fixed. They have changed with lifts, plumbing, electricity, cars, heating, ventilation, and countless other technologies. Perhaps the real opportunity in robotics is not to make machines fit the world we have, but to let them change the world we build.

I love robots. The only problem I have with humanoid robots is the humanoid part.

References