Space Colony Survival (was: RE: Yudkowsky's AI (again))

Billy Brown (
Mon, 29 Mar 1999 10:00:17 -0600

den Otter wrote:
> > From: Billy Brown <>
> > What good would that do?
> Um, how about survival?

As I understand it, you want a Mars colony as insurance against the destruction of the human race via gray goo or other 'dumb' ultratech disasters (as opposed to a smart disaster, which would build its own spacecraft and come looking for you).

That would be a really good idea, except that we can't possibly set up a colony that could actually survive. The best we can manage is to set up a base that would be completely dependant on regular resupply from Earth, and would die off within a few years of the disaster.

Here is my reasoning:

  1. Simply surviving on any other body in the solar system will require an elaborate technological infrastructure. Any colony that cannot maintain its own infrastructure will die off very quickly.
  2. We want the base to be able to produce its own singularity if Earth is depopulated. Unless you believe in a magical, one-invention-that-solves-all-problems genie machine, that means you need a group of researchers big enough to perpetuate a large majority of our current scientific knowledge. An individual can introduce one innovation, but there are so many different things to be done that you still need a large group of people to make progress possible. That means we need tens of thousands of the best minds on Earth, or (more likely) a few hundred thousand researchers with a more normal distribution of ability.
  3. We need a functioning economy that can support the researchers and make use of their results. Right now you would need several hundred thousand people just to achieve a minimal level viability, and there would still be many different kinds of things that could not be manufactured locally. To get true versatility you probably need millions of top-flight people, or tens of millions of average citizens. Now, these numbers should start to decline as better forms of AI emerge, but that is going to be a gradual process. The only way around this requirement is to make such strong assumptions about AI that the whole exercise is pointless - if human-equivalent AI is cheap and easy, SI will be a problem before gray goo is even possible.
  4. Right now, a massive worldwide effort might be able to set up a few hundred people on Mars, with minimal equipment and no long-term survival prospects. If launch costs come down dramatically in the next 20 years we might increase the mass we can afford to move by two orders of magnitude, which would give us a few thousand people with the ability to handle some basic necessities (i.e. local production of air, water and food). This is still far short of even the most minimal requirement (by a factor of at least 10^2 in people, and 10^4 or worse in mass).

To sum it all up, it seems to me that interplanetary colonization requires either a vast increase in wealth (by a factor of 10^2 or more), nanotechnology (which would vastly reduce our minimum throw weight), or sentient AI (which would vastly reduce the need for human staff). Since the appearance of nanotech or sentient AI would make the whole project moot, and the wealth increase seems unlikely without ultratech, the idea of building an emergency survival colony in our current era does not seem workable.

Now, as an aside, I do think it is a good use for the early products of nanotech research. If we get big improvements in industrial automation and some moderately advanced nanotech, but sentient AI turns out to be a very hard problem, then interplanetary colonies would become feasible. Unfortunately, that means we have to solve the gray goo problem first.

Billy Brown, MCSE+I