Read on Twitter
Thinking about self-replicators recently, thread with thoughts. Why do I care? Because a reasonably compact machine that can power itself, produce a broad range of things (including itself) and recycle things into other things is basically a sustainable civilization in a box.
(power itself = solar power or similar). If we had such a thing, we would have drastically fewer things to fight about. No fighting over oil/gas, or even material possessions in general. When you can manufacture things at will, the only thing that matters is the source code.
That's all well and good, but is such a machine even possible? Well, human civilization/tech today has the capability to replace/improve any machine. And sustainable energy is a thing. So, we have a proof of concept, but it's massive, and hard to define. Can we do better?
Yes! We can define the set of all machines & people that are required to reproduce all machines in this set. This would be vastly smaller and simpler than "all the machines and people we have". So we're making progress.
Maybe the complete set, if composed today, would fit in a huge factory, or a small city. We need the ability to manufacture, from raw materials, anything we would need to manufacture any/all of these machines.
From there, assuming access to the designs of these machines, we can optimize. Find shared primitives, remove redundant pieces/technologies, as well as their supporting machinery. Find machines with broad capabilities and use them to replace a variety of specialized machines.
And then iterate. Every simplification brings potential for consolidation and simplification. We get to a much smaller set of machines with all capabilities to manufacture themselves (and much more) from raw materials, power sustainably, & reclaim material from discarded machines
This would be the top-down way to get to a small-ish civilization in a box. At least it works to create an intuition of what's possible. The bottom up way is much more speculative (but likely much more practical). Think of a soft robotic arm:
There are 3d printers today that are essentially robotic arms. Using soft robotics gets rid of the problem of printing motors etc. A tentacle-like 3d printer that can print with material similar to its own, can very plausibly create structures very similar to itself.
The aim is to get the machine(s) to manufacture something as complex as itself or more. You can make progress by improving the capabilities of the machine or reducing its complexity. You'll likely need to do a bit of both.
Soft robotics can most likely pushed far beyond its current state by using compliant mechanism design principles:
In particular, combining compliant mechanism design with generative design can push the capability/complexity ratio even higher:
Obligatory tweet about how deep learning or whatever the next AI wave is can optimize the hell out of whatever it is we're making and make it do more with less. Not spending too much thought on this since I don't love invoking genies and have little non-obvious to add.
To get to a full-blown sustainably-powered self-replicator with excess capacity to create other things too, you'll need to go all the way down to manufacturing the integrated circuits ( https://www.nano-di.com/blog/2019-3d-printing-integrated-circuits-whats-possible-now-and-in-the-future), and the solar cells. ( https://all3dp.com/4/swedish-start-t3dp-might-solved-problem-3d-printing-solar-cells/) .
Nascent technologies, but not unthinkable. And with the RISC-V community incubating open source IP for modern CPUs, you may even have something to print that's worth printing!
There's still the problem of materials of course, and whether exotic ones are required to manufacture some of these advanced elements, but the point isn't to give the full design. Just to argue that it's possible, and the technology is either here or a few years away.
So far in this monster of a thread, we've seen sketches of a top-down approach, a bottom-up approach, but both seem genuinely hard to do for a small team. How could one bootstrap such a thing? You'd likely want to focus on a subset of the problem, from where you can expand out.
I've got a few ideas, but nothing fully formed enough to be worth adding here. The point is that such technology is very likely feasible, and time makes it more likely to happen. And if it did, it would reduce the incentives for inter-human conflict by orders of magnitude.
There is a lot of talk about technology-based existential risk (and such a machine would also be dangerous in a specific way), but I've not heard of any substantial reasons for technology-driven optimism in a while. I owe this line of thought to discovering the #gameb community.
I'm sure someone somewhere has thought of or written similar things, please do link me to them if you know about them. I've put this here as a marker in time, and a call for related ideas to anyone who reads this. Thanks for reading this far!
