If you want a moon base, you probably want to build it before anyone has to live there. This creates a double bind: who builds the moon base? Someone has to show up and start shoveling, and then there’s nowhere to sleep.
What would really be great is if we could send some robots up there to build a moon base for us. Then it would be comfy when we arrive. No shoveling required.
Of course, we’re probably talking about robotic bulldozers and such, heavy stuff to launch into space. And they only work so fast; it might take a long time. We might have to send a lot of them if we wanted, say, a city on the moon, and then we’d have to keep sending them from earth when they broke down. Annoying.
What would really be nice is if we could just send up one of these things, and then tell it to build another one. You know, melt some regolith, refine some metals, grow some silicon crystals, carve out all the parts, stick them together. Now you have two lunar construction robots. Do that a few times and you could really build something big, fast.
This notion is called a Von Neumann probe, after John Von Neumann, who first considered this kind of machine in an abstract way, shortly before computers and space probes started to become real things.
Von Neumann machines, or self-replicating machines, have a remarkable property: in a convenient environment, their numbers can grow exponentially. That can be very convenient for, say, colonizing an inhospitable planet. They also have a big problem: no one’s ever figured out how to build one.
It’s easy to see why it might be hard to build such a thing. Say you start with a bulldozer. Maybe you have a plan to use computer controls to carve molds out of sand and then fill them with molten iron, which will become parts for the new bulldozer. Great. (Drexler would call this a clanking replicator.)
This process is going to be complex, so you’ll need a pretty sophisticated digital logic system to drive it — a computer, essentially. But how are you going to make a computer? Current computers are made out of silicon doped with phosphorus and boron, so you’ll have to find those elements and purify them, a step which requires additional equipment that you’ll also have to carry around. Then there’s the lithography step, which currently requires a gigantic clean-room, an army of solid-state quantum physicists, and a bunch of very carefully tuned machinery that is nigh-impossible to build.
You have to carry all of this with you, and all the equipment required to build it, and all the equipment required to build that equipment … one might wonder whether any finite-size machine can actually achieve this.
Shortly after Von Neumann wrote his mathematical treatment of this idea, showing that it was possible in principle, in a bare mathematical sense, Watson and Crick published their analysis of DNA. Within a few years, it was clear enough: Von Neumann machines really are possible, because every living creature is a Von Neumann machine.
We can start from the simplest cell. A cell consists of (1) a DNA strand encoding some information, (2) a cell membrane (made of lecithin) and (3) some other stuff like ribosomes and RNA polymerase for reading the DNA. It’s abundantly clear that this thing is a self-replicating machine. If you don’t believe me, just watch The Inner Life of the Cell.
How does the cell avoid the limitless expansion of required equipment? I like to think that it relies on the discreteness of matter. Rather than build at macroscopic scale, out of chunks of solid stuff, cells are built atom-by-atom by ultra-miniaturized construction equipment. The simplest cells start to look distinctly digital, with countable numbers of atomic-level components that can be copied or deleted as needed. The problem becomes discrete instead of continuous, and therefore finite instead of infinite.
It’s really cool.
This seems like good news for Von Neumann machine enthusiasts. We know there’s at least one kind, so maybe there are more! But are there?
The thing about Von Neumann machines is that they copy themselves, which means pretty soon you have a lot of them. On earth, we see them everywhere, and typically refer to them by subtypes, like trees, people, mildew, etc.
But we only see one kind. All of these things are Life, which is just one kind of machine. They all have the same lecithin-based membranes, and DNA-based memory units, and closely related other machinery. They all formed from a single family tree.
When we look in rocks, we don’t see little rock-machines tunneling around making copies of themselves. There are no ice-robots, magma-swimmers, solar jellyfish, or moon men. We would know if there were, because once these things get going, they don’t stop until they run out of stuff, and then they start fighting for the remaining stuff, eating each other, diversifying into predator and prey, etc.
Instead, all we see is silence.
Life as we know it formed less than a billion years after the Earth itself, and maybe a lot faster than that. The universe has been around for 13 billion years, and there are countless billions of planets and stars. Where are all the strange self-replicating things? (This is a variant of the Fermi Paradox.)
I think we should entertain a possible answer: they’re here, and we’re it. We have exactly one known type of Von Neumann Machine, and some good reason to believe that if others were possible, we would be able to spot them. We also know that all the other types we’ve tried to design have some serious problems that smell like fundamental physics.
So maybe, just maybe, the laws of physics only permit one kind of self-replicating machine, and it loves cake and hates bleach. Or to be more precise, maybe the space of physically allowed self-replicating machines is connected and fundamentally resembles Biology.
If so, scifi fans are in luck: to learn much about the universe, we really will have to build spaceships, and colonies, and all the rest. The robots won’t get to have all the fun.