>From: "Hal Finney" [mailto:[EMAIL PROTECTED]
>Another way to think of it is that all bit strings
>exist, timelessly; and some of them implicitly specify computer programs;
>and some of those computer programs would create universes with observers
>just like us in them.  You don't necessarily need the machinery of
>the computer to run the program, it could be that the existence of the
>program itself is sufficient for what we think of as reality.

Brent Meeker replies:

> In what sense does "the program" exist if not as physical tokens?
> Is it enough that you've thought of the concept?  The same "program",
> i.e. bit-string, does different things on different computers.  So how
> can the program instantiate reality independent of the computer?

Yes, I think it is enough that I have thought of the concept!  Or more
accurately, I think it is enough that the concept is thinkable-of.

What I mean is, a bit string plus the concept of a computer is enough
to imply a universe with a time coordinate (or more than one!)  and all
the complexity we perceive.  In the Platonic sense both the bits and
the computer-concept exist in the abstract, as both are informational
entities.  So in principle that should be enough.

Now, as to the problem of which computer to use to interpret a given
bit string, what I think you also need to do is to imagine all possible
(abstract) computers as well as all possible bit strings.  Then these
produce all possible universes.

So this exposes a weakness, which is that we seem to need a measure over
all the computers, in order to compute a measure over the universes they
compute.  And if we look at this more closely, we see that I have glossed
over another assumption, which is an implied measure over bit strings.
Robin Hanson on another list challenged me on this point: we need to know
which bit strings are more likely in order to deduce which universes are
more likely.

In the case of the bit strings, there is an obvious symmetric choice,
which is that each bit has independent, equal probability of 1/2.
This is not the only possibility, though, and we might get different
probability assignments for our universes if we assume a different measure
over bit strings.  But this measure is certainly staring us in the face
and has an obvious appeal.

For computers, it is much harder to come up with a natural measure which
will tell us that some computers are "more likely", have more measure,
than others.  My hope is that with further understanding and philosophical
exploration of this issue, we will either come up with an obvious measure
(as in the case of bit strings) or decide that it doesn't matter.

The theory of algorithmic complexity shows that, from a sufficiently
removed perspective, almost all computers compute essentially the
same complexity for a universe.  These wiggle words "almost all" and
"essentially" do leave room for the fact that certain computers compute
completely different complexities for a given universe.  But that's
only a small fraction of computers - at least, I'd like to say that,
but that again seems to require a measure over computers.

So I do think this is an area where some work is needed, but certainly
this result from AIT gives hope for a solution.  It cannot be a
coincidence that almost all computers work almost the same on almost
all universes.  That comes really close to letting us say, the computer
doesn't matter.  Maybe future philosphers will give us better grounds for
letting us be agnostic about the choice of computers.  And maybe we will
even find that the question of bit string measure is similarly irrelevant.

Hal Finney

Reply via email to