From: Hal Finney (hal@finney.org)
Date: Thu Mar 06 2003 - 15:50:58 MST
Wei writes:
> Perhaps a better analogy is this: trying to decide whether re-runs are
> good is like trying to decide whether solipsism is correct. Today
> solipsism can be refuted, by Occam's Razor (or Kolmogorov's Razor to
> borrow Eliezer's term), but realizing this required the invention of
> algorithmic information theory (which includes a definition of simplicity
> that is mathematically rigorous and more or less self-evident). Before
> that, different people had different intutions about what was simplier:
> that only I exist, or that the whole universe exists? And there was no way
> to convince someone whose intuition differed from you.
This is a good point, these tools give us powerful new insights into some
old ethical problems. However, solipsism was not a successful philosophy
even before anyone had heard of Kolmogorov complexity. People were able
to reject it and make progress on moral arguments without that tool.
It is in that spirit that I think we can make progress on the issue
of copies. We may not have mathematical proof, but we can still come up
with arguments that give us good reasons to believe one way or the other.
In the case of copies, consider an experiment where we are simulating
someone and can give them either a good or bad experience. Let's not
even suppose these are replays, they are new experiences which we can
accurately anticipate will be pleasant or unpleasant.
Suppose we are going to flip a biased quantum coin, one which has a 90%
chance of coming up heads. We will generate the good or bad experience
depending on the outcome of the coin flip. I claim that it is obvious
that it is better to give the good experience when we get the 90% outcome
and the bad experience when we get the 10% outcome. That's the assumption
I will start with.
Now consider Tegmark's level 1 of parallelism, the fact that in a
sufficiently large volume of space I can find a large number of copies
of me, in fact copies of the entire earth and our entire visible universe
(the "Hubble bubble"?). When I do my quantum coin flip, 90% of the copies
will see it come up heads and cause the good experience for the subject,
and 10% will see tails and cause the bad experience.
I will also assume that my knowledge of this fact about the physical
universe will not change my mind about the ethical value of my decision
to give the good experience for the 90% outcome.
Now the problem is this. There are really only two different programs
being run for our experimental subject, the guy in the simulation. One is
a good experience and one is bad. All my decision does is to change how
many copies of each of these two programs are run. In making my decision
about which experiences to assign to the two coin flip outcomes, I have
chosen that the copies of the good experience will outnumber copies of
the bad experience by 9 to 1.
But if I don't believe that the number of copies being run makes a
difference, then I haven't accomplished what I desired. The fact that
I am running more copies of the good program than the bad wouldn't make
any difference. Therefore there is no actual ethical value in what I
have done, I might have just as validly reversed the outcome of my coin
flips and it wouldn't have made any difference.
In this way I reach a contradiction between the belief that the number
of copies doesn't matter, the belief that the existence of distant
parallel copies of myself doesn't make much difference in what I should
do, and the idea that there is value in making people happy. Of these,
the most questionable seems to be the assumption that copies don't matter,
so this line of reasoning turns me away from that belief.
I can come up with similar contradictions from simpler cases like
our own observations of subjective probability. The fact that I do
experience a subjective 90% chance of seeing the quantum coin come
up heads corresponds very well with the fact that 90% of the copies
of me will see heads - but only if I assume that the multiplicity of
the copies matters. After the coin flip, in a certain voume of space
there are 90 copies of me that see heads and 10 copies that see tails.
But within the two groups all copies are identical (neglecting other
quantum events which would further split me). If the multiplicity
doesn't count, then there are really just two outcomes and I might
expect to subjectively experience equal probability for them.
This is a variant on an old argument against the MWI, but in that case I
always felt the answer was "measure", that some of the outcomes occured
in a quantum branch which had this intangible quality which made it count
more. In this case I can't invoke any such magic; all the copies of me
are running in the same universe and with equal quantum amplitude. I have
to resort to counting the instances separately and assuming that each one
makes its own independent contribution to my subjective experiences, in
order to gain correspondence with subjective probability. Therefore it
is most consistent to say that separate runs of identical programs do
"count", they do add to the measure of the subjective experience.
Hal
This archive was generated by hypermail 2.1.5 : Thu Mar 06 2003 - 15:57:52 MST