Virtual Earth is not the goal

Rudy Rucker has an interesting and convincing post about the improbability of creating a mirror virtual Earth to replace the "real" one. Unfortunately, I think it conflates several issues and he's only right one of them--fortunately it's the his main point!

To get the main argument out of the way, he's absolutely correct that there's very little cost-benefit in simulating an exact duplicate of the Earth, and that such a simulation is probably doomed to failure because of cost, complexity, and theoretical physical limits.

However, that doesn't mean that the Earth won't end up as computronium anyway. I can easily see a chain of events whereby we create a virtual world (low fidelity and incomplete), load ourselves up into it, create new types of reality, and decide that the VR world is "better" along some cost-benefit axes and continue to convert all available matter. This means we can end of never creating a good simulation of Earth, but that we don't keep the old one around either.

I think Rudy is placing some value on the existing world such that, since we can't perfectly recreate it, we'll keep the old one around because it's so great. While I sympathize and love the Earth as it is, I think the whole point of the Sigularity is that we can't know the minds of post-singular beings and what value they will place on anything. It's very plausible that high-speed simulated worlds are more valuable--maybe even to the point of a necessity for Darwinian survival--that the slow-paced "real reality" is at best forgotten.

I don't think Stross speculates much on what post-singular beings do in computronium, but he does say it focuses more on completely new and complex forms of social interaction and economics. One need look no further than today's Web to see how little simulation is required to allow that to happen.

I'm not placing value judgements on these potential paths, but it's a safe bet that "matching nature" is not the value function that the post-singular beings will attempt to maximize.