> This brings up an interesting conundrum that I raised three or four torture
> experiments ago. Given 10 instantiations of a person having an unpleasant
> experience E ... for example 10 sentient programs running in parallel, is
> it better, if we aim to reduce suffering, to (a) terminate 9 of the 10
> programs and leave one still running and experiencing E, or (b) stop 5 of
> the 10 programs from experiencing E, but leave them running,
Ah ha! And having what sort of experience? It's crucial: I shall
assume that they are having very mildly positive experiences, e.g.,
that their lives are barely worth living in this condition.
> and leave the other 5 programs continuing to experience E?
I will say that it is better to terminate 9 of the 10,
because we are given that the experience E is horrible:
therefore the total benefit to the person is calculated
5*(-1000) + 5*(2) < 1*(-1000)
where, say, E is worth -1000 to you, and the so-so day
worth 2. This uses the additivity of benefit. It also
accords with common sense, in that if you had to sign
up to go through in sequence the experiences of the 10
(five tortures and five so-so days), or instead sign up
to go through just 1 torture, you'd sign up for the latter.
Especially if you were forced to have a run-through of
each ahead of time!
> If you do the total suffering equation assuming that each
> instantiation is separate, (a) is better.
And that is the answer I arrived at.
> But I would argue that if you are one of the suffering victims,
> (a) does you no good at all: subjectively, you will continue to
> suffer, since the one remaining program that is running will
> serve as continuation for any of the 9 terminated ones.
Here is the dreadful "closest continuer" method of Nozick and
others. I claim it gives the wrong answer. Look, the "continuation"
happens anyway, whether you die here or not! Especially if the
events are outside each other's light cones, how can what happens
here possibly affect what happens there? Just because you, say,
are *not* terminated here does not mean that you don't "continue"
there just as much.
What happens in box P far, far away from box Q does not affect what
happens in box Q. If you are in pain in box P, then it doesn't
matter to you in P what goes on in Q. Your benefit seems to me to
be just the sum of the two, much as if you experienced them as a
single copy, but had memory erasure between the experiences.
> In fact, there is no way for someone inside the simulated system
> to know that any of the instantiations had been terminated, as
> long as at least one keeps running.
Well, even if they are ALL terminated, the subject does not know.
Knowing is an activity, and so you don't know if suddenly you all
> On the other hand, with (b) there is subjectively a 50% probability that
> your suffering will end.
People who use probability when discussing duplicates seem to talk
as though an executing process had a magical serial number generated
by God. When the subject dies, it's as if the serial number is
instantaneously transferred to *one* other of the possible systems
that could support him, but not to the others. But there are no souls.
There are no serial numbers. You become all the others equally, and
with 100% probability for each. You even become them if "you" do
What will happen if you choose A is that you will experience E through
one bad session. If you choose B, you will experience E through five
bad sessions, and E' through five so-so sessions. Given that E' is of
no particular value either positive or negative, A is the better choice.
(This is just using different language to do the same calculation as above.)