SI clusters (was RE: seti@home is SORTA WORKING)

Robert J. Bradbury (
Mon, 12 Jul 1999 08:06 PDT

> Rob Harris Cen-IT <> wrote>

> Intelligent barnacles sat on a rock on some remote coastline might say the
> same ("why haven't SI's (humans) altered our rock?)
> Why the hell would we ? Even if there was some reason why we might want to
> change the configuration of pieces of rock, we surely wouldn't get to EVERY
> piece of rock, and considering the vast number of solar systems in this
> universe alone, this makes the rock on earth analogy a massive probability
> in comparison.

This is a good point. SIs could colonize everything *if* they wanted to. Even humans could do it without nanotechnology if the launch costs get low enough and enough people want to do this.

Since we still see stars (assuming the Universe is not a TV screen), then they don't exist or don't desire to colonize (everything). That implies that it isn't to their benefit to colonize everything (just as it is not to our benefit to separate all the rocks from all the barnacles).

I believe this has a lot to do with speed of light delays. There is a measure of parallel processing performance called the logP model, that I won't go into the details of here. The model however allows you to compare the performance of various parallel processors taking into account network bandwidth, time to access the network, CPU speed, etc. Think of an M-brain saying "I need more computational capacity". So send a nanoprobe over to the nearest star with instructions to build another M-brain, great, that gets you 2x your current processing capacity. But you get killed in the logP analysis because of the delays passing the data back and forth between the M-brains.

Consider what would happen to your "mind" if you moved the neurons a foot apart, a kilometer apart, an astronomical unit apart. Your computational throughput is going to slow down significantly.

The only solution is to put the stars *very* close together (but not so close as to collide or produce a black hole). Well, this architecture sounds a lot like a "globular cluster" (hundreds of thousands or millions of stars in the space of a few hundred light years). GCs have stars that are metal-poor, implying they are either very old (and natural) or perhaps relatively young but the metal has been artificially removed. If true, you have one "activity" that SIs do -- strip stars of metals for comutronium/memorenium construction and dump the H/He back into the stars for further processing.

My favorite theory of M-brains says that they hide the stars, but there may be some architectures [esp. long term memory storage units] where you have little use for the energy but metal (C/Fe/etc.) is in really short supply. What I don't know at this time is whether or not the masses of GCs significantly exceeds the estimates from the visible stars. If this were true then visible GCs are the hard drives or tape backups. There might be "hidden" GCs that would be the Super-SI CPU "clusters" that think about really "long-term" problems (sure, the LogP parameters are poor (inter-CPU times of weeks to months), but they are better than a single-SI and you've got billions of years anyway).