Remote hands are the future of physical work

"Black and rubbery and a yard long and almost frighteningly humanlike."

Why not start 2023 with a nice prophecy? Sometimes people tell me that there is just too much nasty stuff here on Gray Mirror. Maybe they are right. Maybe people are tired of reading about coups and dictators and kings all the time. Can’t we be optimistic?

Well, at least—someone can. All the best prophets steal. Let me steal an idea from a better and more optimistic prophet, Robert Heinlein, and propose a thesis: remote hands are the future of physical work.

Picture a future with a standard high-fidelity haptic input rig. This rig gives your hands and fingers motion tracking with sensation and resistance, and your arms just motion tracking. With a VR display, your upper torso can animate an output robot of any scale and strength, dextrous enough to use hand tools appropriate to that scale.

Through such a rig, you could play the violin, cook a meal, or screw in a screw. Or you could play a 50-foot-long violin with a 30-foot bow, cook a meal for a hundred people in a pan the size of a ping-pong table, or use a millimeter-long screwdriver with micron bolts and nuts to screw two human hairs together. Of course, everything in physics changes with scale—but it only changes so much.

Why is this better? At present, when we work in the physical world at human scale, we use our general-purpose hands with special-purpose hand tools. When we work at an inhuman scale, either larger or smaller than is suited to hand tools, we use custom machinery—whose haptic bandwidth, in input and output, is far more specialized. This low I/O bandwidth vastly increases the human labor cost of the endeavor.

Imagine you are replanting a baby redwood tree, three feet tall. A few strokes of the shovel makes a hole big enough for the root ball. Holding the tree by the trunk with one hand, you fit the ball loosely into the ground as you use a bag of soil to fill in the gap between the hole and the roots.

Now imagine doing the same thing with a young mature redwood, a hundred feet tall. The scale is different. The task is essentially the same. But the hole cannot be dug with a few strokes of the shovel; it needs earth-moving machinery. Bulldozers, etc.

Both the input and the output of a bulldozer are far more narrow in bandwidth than the input and output of a human hand guiding a shovel. As a result, far more human operations are needed for the big pit than for the little hole. As always, there are physical differences of scale—but try to imagine humans replanting the little tree with little remote-controlled bulldozers, etc. It would be little less laborious.

Imagine building a miniature house, with all the same parts as a regular house, but the size of a dollhouse, on a desk in a shop. (Some parts at this scale would be hard to manipulate because they are too small—but you could work at any scale.) Imagine how much easier it would be to put together this model, than to build an actual house.

For any mechanical problem, the right scale is a scale that lets you use hand tools easily—not so small that precision becomes a problem, not so large that strength becomes a problem. With our present-day toolkit, when we work on this optimal scale, we use general-purpose motors (hands) with special-purpose tools. When we work on any other scale, we use special-purpose tools with specialized motors.

When you see the future, see the big tree being planted by a big landscaping robot. Robot and tree both arrive on the biggest possible trucks. The robot is a swiveling skeletal torso, fifty feet tall, with diesel-electric power, rising from the truckbed. It starts the job by anchoring its own shoulders to the ground with external guywires, so that it can act stably with enormous force. When it digs, it uses a shovel the size of a telephone pole. It is driven by some guy in his cluttered office-bedroom in Kentucky. Its hands are black and rubbery and a yard long and almost frighteningly humanlike. Caterpillar’s biggest general-purpose bot is two hundred feet tall and if you give it an elephant, a blast furnace, and the right knives and stuff, it can kill and clean and cook and serve the thing almost just as fast as your country cousin can process a rabbit.

Whereas if you have a brain tumor the size of a marble that needs to be removed, the ideal way to remove it may be a team of twenty surgeons, each the size of an ant. The landscaping operator in Kentucky will probably not be on that team. But in principle, he may be using the same input rig. Isn’t that a cool future? I think it’s a cool future.

One of the reasons it is a cool future—to connect this future back to kings and stuff—is that kings, since they own their subjects, must concern themselves economically not just with the products of their subjects work, but the impact of production on the subject. Otherwise your king is ignoring appreciation and depreciation, which is just an accounting mistake—pretty embarrassing in a king.

A human being appreciates, becoming a more worthy person and thus worth more, by growing deeper in skill—and the more general a skill is, the greater depth is found in it. Charlie Chaplin in Modern Times played a human being used as a primitive industrial robot, just pulling a lever a thousand times a day. Here again we see profoundly limited haptic bandwidth. The activity is stultifying and soul-destroying.

When action in the real world at every scale is performed by general-purpose haptic interfaces, more cool things get done better, and there is more demand for physical labor that—because of its broad haptic bandwidth—is creative, satisfying, and not soul-destroying but soul-building. This is capital appreciation in human capital.

While a king may appreciate this improvement through his spiritual capacity as a servant of God, his finance minister appreciates it in his professional capacity as a servant of Mammon. Broadly skilled artisans are just better people than machine serfs, an improvement to the property that may be hard to measure but will keep paying off for many years and generations.

Of course, it is possible that these hands will only be trained by humans—but, in actual reality, controlled by mere filthy statistics—aka “artificial intelligence.” But it seems unlikely. Making art is easier than it looks, but manipulating objects is harder.

For humans, writing a great pop song is harder than folding a shirt. For computers, folding a shirt is harder. Generative AI can amaze us with its creativity, but it cannot help us in driving a car, folding a shirt, or planting a tree.

Managing moving objects in the real world seems to always ramify into a set of obscure corner cases in which there is not enough training data. Generative AI works where errors—the wrong number of fingers in a painting, math errors in prose—do not matter. If it was as bad for a text generator to make a logical mistake as for a self-driving car to crash, no one would use text generators.