On 30 Oct 2013, at 06:35, Jason Resch wrote:
On Tue, Oct 29, 2013 at 3:12 PM, LizR <[email protected]> wrote:
I suggested doing this on FOAR (I used HAL from 2001). It simply
makes it easier to visualise if you forget about biological
creatures. Assuming comp, an AI is exactly equivalent to a human
person, so anything you can do to an AI could be done (in theory) to
a human by a teleporter, or to a human by MWI style splitting.
What should the AI expect to see? It should expect to see the ball
turn red and remain red.
Should it expect (expect as in place a high probability on) that?
Only 1 of the 256 actually see that happen. It is far more likely
to see an incompressible pattern.
There are copies of it which see the ball go blue at various points...
However this answer doesn't assume comp.
The existence of a conscious AI implicitly assumes comp
It assumes only "strong AI". Strong AI might be true, yet comp false.
It is not because machines can think that only machine can think. may
be angel and gods can think too. Of course, if comp is true, some
machine can think, and thus string AI is true.
(at least for some types of observers, you could still like Craig
argue that computers cannot support *your* experience, only some
limited class of experience).
Well, the puppet's experience, I presume.
We know that a Turing universal puppet can cut his links with the
manipulator :)
According to comp it doesn't know what "it" will see, or to be more
exact it knows that "it" will see all combinations, but by that time
it will no longer be an "it" but a "them". Technically - in this
case - we know which ones are the copies and which ones aren't -
however comp says that the AI will experience becoming many AIs,
with varied experiences.
====================
I think we can all agree on this (LizR, Bruno, Clark, Chris, myself,
etc.):
If the AI (or all of them) went through two tests, test A, and test B
A) The test described where the simulation process forks 8 times and
256 copies are created and they each see a different pattern of the
ball changing color
B) A test where the AI is not duplicated but instead a random number
generator (controlled entirely outside the simulation) determines
whether the ball changes to red or blue with 50% probability 8 times
Then the AI (or AIs) could not say whether test A occurred first or
test B occurred first.
===================
If you agree with this, that is sufficient to reach the main point
of step 3, which is the two tests are subjectively
indistinguishable. Expecting the ball to change color at random
(test B), and being iteratively duplicated and seeing all
possibilities in different instances (test A), are absolutely
indistinguishable from any point of view that exists inside the
simulation. No one inside the simulation can determine whether test
A was happening, or whether test B was happening. It is a very
simple point, and I don't think anyone here would argue that an
observer within the simulation could distinguish between the two
cases.
If you happen to disagree that an entity within the simulation could
distinguish between test A or test B (that is to say, that they
could guess whether test A or test B was happening with greater than
a 50% probability) then please state how that can be done.
Otherwise, you understand the point of step 3 sufficiently to move
on and there is no more need to argue about pronouns, personal
identity, which you you happen to be, etc.
If anyone does not provide an argument for how the AI, or AIs, (or
any observer or entity) within the simulation could distinguish
these cases, and continues to argue about pronouns, personal
identity, etc., then I think the only conclusion that remains is
that such a person has little or no interest in advancing their own
or anyone else's understanding and is simply being a troll.
The point is crystal clear and indisputable in this situation, it
doesn't matter how the AI is programmed: there is no way for any
entity in the simulation to distinguish between an inherently random
process (test B) from a wholly deterministic one (test A). If you
think you know a way, then please tell us how. If you see no way,
then you accept step 3, which is that the appearance of subjective
indeterminacy can arise in an objectively deterministic processes.
(Note the above is not aimed at any person in particular. If anyone
can show where the reasoning is wrong, please do so.)
Indeed.
In any case, although one copy is the original, that doesn't really
help, because an AI, by its nature, is probably being constantly
swapped into different parts of computer memory (or stored on disc),
parts of it are being copied, other parts erased, and so on. Comp
says none of this matters - that its experiences are at a
fundamental level exactly like ours.
So. What's wrong with this picture, if anything?
What do you mean by one copy is the original? How can you
distinguish an original from a copy?
I don't think Liz did that.
Note that someone disbelieving comp, and drugged in Helsinki, and then
forced to do the WM-duplication, will wake up in both city, both
pretending that they are the real original one, perhaps even
disbelieving they have been teleported. They will not recognize
themselves in the copy, as that might not be that easy. (Especially if
your face is quite dissymmetrical, like in the novel "Despair" by
Nabokov).
Bruno
Jason
On 30 October 2013 09:41, Jason Resch <[email protected]> wrote:
On Tue, Oct 29, 2013 at 2:06 PM, meekerdb <[email protected]>
wrote:
On 10/29/2013 8:19 AM, Jason Resch wrote:
Chris,
Perhaps it is simpler to think about first person indeterminacy
like this (it requires some familiaraity with programming, but I
will try to elaborate those details):
Imagine there is a conscious AI inside a virtual environment (an
open field)
Inside that virtual environment is a ball, which the AI is looking
at and next to the ball is a note which reads:
"At noon (when the virtual sun is directly overhead) the protocol
will begin. In the protocol, the process containing this
simulation will fork (split in two), after the fork, the color of
the ball will change to red for the parent process and it will
change to blue in the child process (forking duplicates a process
into two identical copies, with one called the parent and the other
the child). A second after the color of the ball is set, another
fork will happen. This will happen 8 times leading to 256
processes, after which the simulation will end."
It is 11:59 in the simulation, what can the AI expect to see during
the next 1 minute and 8 seconds?
I don't see that as any different.
It is similar, but it never hurts to look at the same problem from
different angles. What is a little more evident in this case is
that of the 256 possible memories of the AI about to meet its doom,
none contain the memory of seeing all 256 possibilities, an in fact,
the majority of them see the ball change color back and forth at
random. Only 2 see it stay all red or all blue for the last 8
seconds. None of them can predict from the view inside the
simulation, whether the ball will stay the same color or change
after the next fork occurs.
The problem is still what is the referent of "the AI". As John
Clark points out "the AI" is ambiguous when there are duplicates.
Personal identity is less of an issue in this case, because it
concerns the AI or anything/anyone else inside the simulation who
might also be viewing the ball. In this way, it is slightly more
analogous to MWI since it is the environment which is duplicated,
not just the person, and so the apparent random changing of the ball
color is also something that can be agreed upon by the group of
observers within the simulation.
Sometimes Bruno talks about "the universal person" who is merely
embodied as particular persons. So on that view it would be right
to say *the* universal person sees Washington and Moscom.
But not "at the same time" or as "an integrated experience", so the
appearance of randomness still arises from the first person
perspective(s).
But then that's contrary to identifying a person by their memories.
My view is that "a person" is just a useful model, when there is no
duplication - and that's true whether the duplication is via Everett
or Bruno's teleporter.
What model should be used in a world with duplication, fission
machines, mind uploading, split brains, biological clones, amnesia,
etc.? Or does personhood no longer make sense at all in the face of
such situations?
Personally I believe no theory that aims to attach persons to one
psychological or physiological continuity can be successful.
Jason
--
You received this message because you are subscribed to the Google
Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it,
send an email to [email protected].
To post to this group, send email to [email protected].
Visit this group at http://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/groups/opt_out.
--
You received this message because you are subscribed to the Google
Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it,
send an email to [email protected].
To post to this group, send email to [email protected].
Visit this group at http://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/groups/opt_out.
--
You received this message because you are subscribed to the Google
Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it,
send an email to [email protected].
To post to this group, send email to [email protected].
Visit this group at http://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/groups/opt_out.
http://iridia.ulb.ac.be/~marchal/
--
You received this message because you are subscribed to the Google Groups
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To post to this group, send email to [email protected].
Visit this group at http://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/groups/opt_out.
