🚀 go-pugleaf

Hal's Extropians Digest      Fri, 19 Jun 1992       Volume 92 : Issue   0011

    Selected messages from the Extropians list.  To join, send to:
    <extropians-request@gnu.ai.mit.edu>.  Messages from the Extropians
    list may not be cross-posted without permission from the poster.
    Such permission has been obtained for all messages in this Digest.

    Hal Finney
    Extropians Digest Moderator

Today's Topics:
		    Newcomb's Paradox (8 messages)

----------------------------------------------------------------------

Date: Tue, 16 Jun 92 08:55:30 PDT
From: uunet!ghsvax!hal (Hal Finney)
To: Extropians@gnu.ai.mit.edu
Subject: Newcomb's Paradox

Thanks to Mike for posting about the paradox, but I think his version
was lacking an important ingredient.

As Mike said, you are facing a SuperEntity who will predict your
actions.  I'll say more about this in a moment.  There are two boxes
on the table, and before you came into the room the Entity put money
into them.  Your choices, as I remember it, are to pick either box A,
or _both_ boxes A and B.

In box B the Entity always puts $1,000.  In box A, the Entity has put
either $10,000 if it predicted that you would choose A, or only
$1,000 if it predicted that you would take both A and B.

Now, if the Entity is right, then if you pick just A you will get
$10,000, but if you pick both A and B you will get $2,000.  So
assuming that the Entity is right you should just pick A.

OTOH, when you come into the room the Entity has already filled the
boxes.  The money is sitting before you on the table.  Nothing you do
now can change the amount of money in those boxes.  Either there is
$10,000 or $1,000 in box A, and there is $1,000 in box B.  By taking B
along with A you will always get $1,000 more than if you just picked A
by itself.  So apparently you should pick both A and B.

It is said that most people find one or another of these two arguments
so convincing as to be utterly obvious, yet people disagree in roughly
equal numbers as to which argument is right.

Many people will question whether such a SuperEntity could exist.  For
the sake of the argument, you are asked to believe that it does.
Suppose you have played this game or other games many times, and each
time the Entity has predicted your actions correctly with 100%
accuracy.  In technological terms, we might say that the Entity has an
accurate model of your brain and is running it to see what you will
do.  (If you still don't believe that such an Entity could exist, then
don't pay any attention to the paradox.)

The two arguments for why you should pick only A, or why you should
both A and B, remind me of the arguments about whether you should
Cooperate or Defect in the Prisoner's Dilemma game.  I choose
Cooperate in the single-shot PD, and by what feels like similar
reasoning, I would take only box A in Newcomb's paradox.  I wonder if
there is a correlation in the answers on these two paradoxes.  Do
Defectors in the PD take both boxes A and B?  Do other Cooperators
take just box A?

Hal Finney
ghsvax!hal@uunet.uu.net				...!uunet!ghsvax!hal
Permission is granted to freely reproduce this message.

------------------------------

Date: Tue, 16 Jun 1992 11:13-0700
From: Robin Hanson <uunet!CHARON.arc.nasa.gov!Hanson>
To: Extropians@gnu.ai.mit.edu
Subject: Newcomb's Paradox

>I choose
>Cooperate in the single-shot PD, and by what feels like similar
>reasoning, I would take only box A in Newcomb's paradox.  I wonder if
>there is a correlation in the answers on these two paradoxes.  Do
>Defectors in the PD take both boxes A and B?  Do other Cooperators
>take just box A?

Er, the experimental economics literatures says *everyone* eventually
chooses Defect in the Prisoner's Dilemma.  Set up a game where say 20
strangers are repeatedly paired up to play the game, pairing is random
each time and anonymous, and players are paid substantial real money
based on the outcome.  Some people take longer than others to come
around, but eventually just about everyone switches to defect.  Heavy
moralizing overtones can slow it down, but ...  Economists switch faster
as do people from more prestigious universities.  Don't believe me?
Try it yourself!

For obvious reasons, Newcomb's paradox behaviour is harder to test.

Robin Hanson

------------------------------

Date: Tue, 16 Jun 92 12:32:54 PDT
From: uunet!ghsvax!hal (Hal Finney)
To: Extropians@gnu.ai.mit.edu
Subject: Newcomb's Paradox

   From: Robin Hanson <uunet!CHARON.arc.nasa.gov!Hanson>

   Er, the experimental economics literatures says *everyone* eventually
   chooses Defect in the Prisoner's Dilemma.  Set up a game where say 20
   strangers are repeatedly paired up to play the game, pairing is random
   each time and anonymous, and players are paid substantial real money
   based on the outcome.  Some people take longer than others to come
   around, but eventually just about everyone switches to defect.  Heavy
   moralizing overtones can slow it down, but ...  Economists switch faster
   as do people from more prestigious universities.  Don't believe me?
   Try it yourself!

*Everyone*?  "Just about" everyone, maybe, but I'll bet it depends on
the group involved.  If heavy moralizing can slow it down, what about
heavily moral people?  What if you had twenty Mother Theresa's playing
the game?  Don't you think they might all end up cooperating?

Pretty silly of them, too.  They could have been smart like those
economists from all those prestigious universities and gone home with...
75 cents?  Instead of the fifty bucks poor dumb little Mother Theresa
ended up with. ;-)

Anyway, it would be an interesting social experiment to see how
personality traits correlate with behavior in this situation.  I might
guess, for example, that very religious people would be more likely to
cooperate than people on certain mailing lists...

I've also often thought of how interesting it would be to run a
Newcomb's Paradox experiment.  The hard part would be to convince the
player that his choice really was going to be accurately predicted.
This doesn't necessarily mean that you have to accurately predict it,
only that he believes that you will.  One possibility, for example,
would be to pay people to lie to the subject and say that they had
just finished playing it and had had their moves accurately predicted.
Then you wire the guy up with a headband and electrical probes so he
really thinks there's something to it.

Another approach would be to use magic tricks to make it _look_ like
the money is safely in the boxes, untouchable by the experimenter, but
to make a switch at the list minute after he's made his choice so that
he always gets the right amount.  Then you could repeat the experiment
a few times, enough to convince him that it really works, and only
count the results from the last run.

Hal Finney
ghsvax!hal@uunet.uu.net				...!uunet!ghsvax!hal
Permission is granted to freely reproduce this message.

------------------------------

Date: Tue, 16 Jun 1992 13:49-0700
From: Robin Hanson <uunet!CHARON.arc.nasa.gov!Hanson>
To: Extropians@gnu.ai.mit.edu
Subject: Newcomb's Paradox

Hal Finney writes:
>>>I choose Cooperate in the single-shot PD,

>>Er, the experimental economics literatures says *everyone* eventually
>>chooses Defect [...]  Some people take longer than others to come
>>around, but eventually just about everyone switches to defect.

>*Everyone*?  "Just about" everyone, maybe, but I'll bet it depends on
>the group involved. [...]  What if you had twenty Mother Theresa's playing
>the game?  Don't you think they might all end up cooperating?

No, actually I don't.  And I believe that you would eventually defect
when playing with random people.  Let's just agree that no experiment
has yet identified a type of people who don't eventually settle on
defect.  Maybe you have an idea for a new interesting experiment, which
will overturn the expectations created by all the experiments to date.
If so, good luck with it.  Till then I'll stick with conventional wisdom.

Robin

------------------------------

Date: Tue, 16 Jun 1992 18:58-0700
From: Robin Hanson <uunet!CHARON.arc.nasa.gov!Hanson>
To: Extropians@gnu.ai.mit.edu
Subject: Re: Newcomb's Paradox

O.K.  I'll bite.  If SuperEntity has already made its guess about
whether I'll open B or not (according to Hal's formulation) and stuffed
A accordingly, and there are no other consequences to consider, such as
keeping my word to someone or trying to influence future SuperEntity's
guesses about whether I would do similar things in similar
circumstances, then it is obvious that I want to open B.  If I open
B I get $1K more for sure, no uncertainty about it.

But, if I am here before the SuperEntity makes its estimate, and am
trying to decide if I should change myself in such a way that I won't
pick B, I might in principle want to change myself to become "irrational"
in this certain way because SuperEntity rewards such folks, if in fact
the rewards were large enough.  I say "in principle" because there are
huge risks here.  If I make myself irrational in some way, what future
opportunities will I miss because I make the wrong decision there, where
there is no SuperEntity to reward such mistakes?

A variant of the paradox that is more direct to me is this:  Imagine a
SuperEntity who will give you $1B in a year if he is fairly confident
you will act through the rest of your life as if you sincerely believed
that "P = NP" (or any other statement for which there is not anywhere
near enough evidence for a rational belief in yet, but which doesn't
affect your daily life very much).  Would you try to make yourself such
that you believed this, and would continue to believe it?  You have a
year and drugs, brain-washing, etc. are all available for you to try.

I think this general issue is more relevant than it might seem because
most of us are under great social pressures to be irrational in various
ways.  Until recently, and perhaps still, great social rewards are given
to those who believe in religions without sufficient evidence, or who
pledge allegiance to dominant local ideologies.  It is quite easy, and
perhaps even reasonable, to not pay as much attention as one rationally
aught in questioning such things.  Even, or perhaps especially, in small
counter-culture social groups such as the Extropy group here there are
rewards for proper enthusiasm for the dominant beliefs, and penalties
for being too much a "wet blanket".

A Drippingly Wet Blanket, Robin Hanson

------------------------------

Date: Wed, 17 Jun 92 09:14:53 PDT
From: uunet!ghsvax!hal (Hal Finney)
To: Extropians@gnu.ai.mit.edu
Subject: Newcomb's Paradox

> A Drippingly Wet Blanket, Robin Hanson

Well, if our worst problem is Robin acting like a wet blanket I think
we're in pretty good shape!  Seriously, I think there is more
diversity here than some of the posts indicate.  But still, we do
share some points of view, which is what gives the list a reason for
existance.

> O.K.  I'll bite.  If SuperEntity has already made its guess about
> whether I'll open B or not (according to Hal's formulation) and stuffed
> A accordingly, and there are no other consequences to consider, such as
> keeping my word to someone or trying to influence future SuperEntity's
> guesses about whether I would do similar things in similar
> circumstances, then it is obvious that I want to open B.  If I open
> B I get $1K more for sure, no uncertainty about it.

OK, this is another data point - Robin would take both boxes, and I
gather from his earlier posts that he would Defect in the single-shot
PD without hesitation.  I wonder if there is anyone who would take
just box A in Newcomb but quickly Defect in the PD, or vice versa.
I'd be interested to hear their perspectives.

> But, if I am here before the SuperEntity makes its estimate, and am
> trying to decide if I should change myself in such a way that I won't
> pick B, I might in principle want to change myself to become "irrational"
> in this certain way because SuperEntity rewards such folks, if in fact
> the rewards were large enough.

I don't completely understand this.  You are talking about "changing
yourself" so that you will pick just box A, thereby gaining $10,000.
Presumably you will do this through some special process of altering
your brain.  But it seems to me that if you can decide to do this,
because you think it's a good idea, then the same considerations will
simply allow you to pick A without going through that rigamarole.

In other words, if you can decide to change yourself to pick A, you
can decide to pick A.

Now, Robin's other example is harder, where he has to act a certain
way for a whole year.  If this is really something that would come
unnaturally to him, then maybe he would be afraid that he might forget
himself and behave wrongly.  So in that case it would make more sense to
take special measures.

But in general, I think that if you can change yourself to act a
certain way, you can act that way without changing yourself, at least
in a short-term situation like the Newcomb paradox.  It might take a
little will power.

(One counter-example I can think of is someone who needs a drink to
act sociable at parties.  They are changing their brain to cause
certain behavior that they are in principle capable of producing
without that brain change.  I don't know whether this is the kind of
thing that Robin was thinking of.)

Hal Finney
ghsvax!hal@uunet.uu.net				...!uunet!ghsvax!hal
Permission is granted to freely reproduce this message.

------------------------------

Date: 17 Jun 92 23:09:35 EDT
From: Michael Clive Price <uunet!CompuServe.COM!100034.3077>
To: Extropians@gnu.ai.mit.edu
Subject: Re: Newcomb's Paradox

> I wonder if there is anyone who would take just box A in Newcomb
> but quickly Defect in the PD, or vice versa. I'd be interested to
> hear their perspectives.

Assuming that there was real money on it I would definitely go for box A
and Defect in singe-shot PD (unless I was playing against a copy, in
which case I'd go for Cooperate).  I would feel a bit of a cheat in the PD
case, but the logic seems inexorable.  In the iterated PD I would play
tit-for-tat.

Mike Price

------------------------------

Date: Thu, 18 Jun 92 10:18:40 PDT
From: uunet!xanadu.com!tribble (E. Dean Tribble)
To: Extropians@gnu.ai.mit.edu
Subject: Newcomb's Paradox

   The two arguments for why you should pick only A, or why you should
   both A and B, remind me of the arguments about whether you should
   Cooperate or Defect in the Prisoner's Dilemma game.  I choose
   Cooperate in the single-shot PD, and by what feels like similar
   reasoning, I would take only box A in Newcomb's paradox.  I wonder if
   there is a correlation in the answers on these two paradoxes.  Do
   Defectors in the PD take both boxes A and B?  Do other Cooperators
   take just box A?

But in the single prisoner's dilemna, there is no advantage to
cooperating.  No matter what you're partner does, you *will* be better
off Defecting.  Period.  It's only in the iterated situation that
Cooperating makes sense.

On a side note: One of the reasons why reputation agents will
eventually become extremely important is to put sellers in an iterated
situation with the market.  Therefore if the Defect with one customer
(screww, provide shoddy service, etc), they will feel it's effects
through many customers.

dean

------------------------------

End of Hal's Extropians Digest Volume 92 Issue #0011
****************************************************