So I’ve come to the conclusion that the most significant thing to focus on right now is getting clanking replicators up and running: programmable machines based on bulk-material-processing technology (i.e. the way all machines except for a few experimental STM and AFM setups in research labs operate) that can reproduce themselves much faster than the economic growth rate, say in hours to weeks rather than decades.
I think this is possible now, there’s a sort of “arms race” underway to get it to happen, and whoever succeeds will have a massive economic advantage, comparable to but larger than the discontinuity in the shift from hunter-gatherer and pastoral-nomad lifestyles to agriculture.
Freitas is the one who’s written the most about this in the past. Sipper also has a page on the issue.
I’ve just downloaded Freitas and Merkle’s 2004 book KSRM, which seems to be the latest survey of the space; it probably inspired Adrian Bowyer to start RepRap. Freitas in particular seems focused on molecular nanotechnology now, but it seems likely that MNT is somedistance further down the road.
The race is unremarked and can be carried out at small scale and without exotic materials or, probably, much special resources. Once a clanking replicator is created, exponential growth should be rapid.
Many words are available from fiction and more careful speculation for such a project: Autofacs, Second Variety, Screamers, Berserkers, Auxons, and Replicators. Auxon is a positive term (from Lackner and Wendt’s 1995 proposal), and Autofac is sort of neutral; the others are all nightmares.
I have a lot of reading to do now.
I think mechanical control systems might be adequate, and they won’t require the exotic high-purity materials that semiconductor devices do. Freitas I guess didn’t think that was going to be a problem in 1980.
However, this requires a mechanical system capable of universal computation. Reif wrote a survey chapter in 2008 on mechanical computation which seems to suggest that nobody has built a mechanical universal computer yet.
The Curta I calculator had only 571 parts, while Vaucanson’s swan (according to Freitas) had over 1000. I think Calculus Vaporis could probably be implemented with a similar parts count using lookup tables for combinational logic.