From: Eugen Leitl (eugen@leitl.org)
Date: Thu Jun 27 2002 - 08:32:30 MDT
On Thu, 27 Jun 2002, Louis Newstrom wrote:
> Even if, by reading a "source of entropy", the simulation could get
> truly random numbers, I think that that would be possible to detect.
Well, it's easy enough to check. Pipe the output of /dev/random with true
entropy (not all implementatioins mantain an internal entropy pool,
though) though SHA-1, or just keep forcefeeding SHA-1 with its own
excreta, and try to tell both apart with statistical methods.
> Just as an example, if we did a statistical correlation between two
> unrelated events (like meteor showers vs. robbery attempts) and found
> that they had the same distribution pattern, that would be suspicious.
There are a lot of wholly unrelated distributions (e.g. those following
power law) which show the same patterns. We typically don't get suspicious
when we see this, but instead marvel at the beautiful order underlying and
permeating all things (cough, cough). Also, you seem to assume that we'd
be using a coarse fake, and not accurate models (I can run a very detailed
model of the solar system with a fraction of a resources necessary to
model a mind accurately). And robberies are made by people, so I don't see
how you would need a random or pseudorandom source for that.
Another point brought up in the thread: if I upload you, I of course can
upload (scan) the entire world, making a model just sufficiently accurate
to foil monkey senses. It is actually not necessary to do it right away,
because all it takes is uploading the part of the world you think you're
exploring. If you're going to visit a room not yet in our model, we'll
slow you down, scan the room, and you resume execution when you enter it.
Denizens of the virtual world are surrounded with bubbles of carbon-copied
reality, and they leave trails of scanned reality behind them for later
use. Cities would wind up completely cached very quickly, areas of remote
wilderness need not to be represented at all.
> If we correllated other random events, and found that all random
> events had the same distribution pattern, that would be a clue that
> they are being generated by the same random generator. That would be
> an indication that we are in a simulation.
You're postulating a very coarse, sloppy numerical model. I could equally
expect rendering artifacts in the sky, or imperfect volume exclusion.
This, however, is not a modern video game.
I agree that we're probably not a simulation, but I disagree in principle
that we can tell. I'm just sticking with Ockham's razor here.
> This might be happenning, and no one noticed yet. But I don't buy the
> argument that it would be impossible to tell. I think that if we ARE
> in a simulation, we will eventually find out.
Um, no. Look at what digital physics folks are claiming: if you've got
discreteness at Planck level, the reasons for that are completely
arbitrary. The machinery is entirely transcendent. Immanence is perfectly
sandboxed.
This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 09:15:03 MST