[extropy-chat] Superrationality

[Lee Corbin]

Damien wrote

> On Thu, May 18, 2006 at 03:15:22PM -0700, Lee Corbin wrote:
>
> > Well, were it in our power to redefine what Hofstadter meant, then
> > yes, it might be an improvement to call it advice---or, what is the
> > same thing in game theory, a strategy.
> 
> What he said, which I can because I just checked, was that you should
> realize you're a typical member of the society, hence will make the same
> decision as the other, hence should cooperate.  One flaw is that
> thinking this way is in fact not typical of members of our society, at
> least put in this abstract way.

Thanks for checking on Hofstadter's original definition. Let me
take a swing at it line by line.

"One should realize that one is a typical member of some class
and so hence will [helplessly] make the same decision as any
other member of that class."

We must acknowledge the cases in which there *does* exist such
a class---and so superrationality exists, may I posit---and
those cases in which there is no such class.

Consider my earlier questions: (In the NIPD), would you Cooperate
with a ruthless gangster who was used to exploiting others and who
knew that you were not a ruthless gangster? Would you Cooperate
with someone who was on record as expressing doubts about
Superrationality?

The answers to these are clearly "no". (Alas, I'm afraid that it
is absolutely essential for readers to UNDERSTAND THAT THIS IS A
ONE-SHOT PRISONER'S DILEMMA WHICH IS *NOT* RELATED TO REAL LIFE.
IN REAL LIFE THERE IS NO EXPLICIT PAYOFF MATRIX THAT TAKES INTO
ACCOUNT ALL ONE'S VALUES; THIS IS A SINGLE PLAY OF A MATHEMATICAL
SITUATION IN ABSTRACT GAME THEORY.)

On the other hand, if you were playing with a duplicate of yourself
that had been created five minutes ago, or were playing with a 
mirror-image of yourself, then the answers are most assuredly
[and helplessly] "yes". For, were you to have doubts about whether
superrationality were the correct strategy when playing against a
close duplicate, then so would your duplicate!  Again, we are talking
about strategy for an *instance* of one in a particular game; we are
*not* talking about the more subtle question of what is a *person*.
(In the latter case, by the way, I do consider myself to be the
same person as my duplicate; but I am *not* the same instance.)


Is it superrational to vote?  After all, if you decide to not vote,
will that change the outcome by only one vote?   No!  Not necessarily!
Suppose---just as in the case of superrationality---that whatever
causal influences come to you from the outside (say the latest
scandals involving certain politicians) that motivate you to vote
may also motivate others to vote in exactly the same way. Therefore,
it is sheer arrogance to assume that you have free will, and can
make decisions uncorrelated with the rest of the universe. I always
figure that when I do or don't vote, the totals are changed by perhaps
5-50 votes in a large election.

One way of looking at Hofstadter's superrationality, then, is to
consider whether or not this meme of superrationality (or meme of
Cooperation under certain circumstances) is one of these "outside
causal influences" that constrains your "free will". In a nutshell,
if you know that you are playing a certain individual X in a NIPD,
then is it or is it not the case that (as in some hypothetical
voting cases) your behavior is correlated with his?

As I have said, the *only* cases I know of where superrationality
can be correct is when it is true that your behavior is positively
correlated with that of your opponent X. Now if X is Douglas
Hofstadter in 1983 (where Doug1983 happens to know that you are
an Extropian from 2006 who is fully up to speed on Superrationality),
then you know he's going to Cooperate. Hence you should Defect. I
therefore repeat my earlier conclusion: Cooperation is possible if
(1) you don't happen to know what your opponent will do, and
(2) there is reason to believe that his behavior is strongly
correlated with yours.

> In more usual situations, as my tipping question was meant to
> suggest, many of us may have superrational habits.

I would suggest that this is *not* a case of superrationality
at all. This is a case of genuine altruism. (I join those who
praise genuine altruism, and strongly disagree with that brand
of libertarian/individualist who embraces selfishness to such
a degree that altruism is seen as unwise.) It's not any kind
of NIPD because by the time you've got the check, the waitress
has already played her move.

> What he also said, in his postscript to the final column, was that
> societies might be divided into Type I and Type II.  Type II we're
> familiar with.  Type I members believe in the rationality of
> one-shot cooperation *with other members of Type I societies*.

Yes. Suppose that you *know* that the adversary in the NIPD is
a Type I individual who believes in the rationality of one-shot
cooperation. Then you should Defect. (All one has to do to
understand this is read the payoff matrix again.) Likewise,
suppose that your adversary knows that *you* are superrational;
then he *should* exercise his free will and Defect.

Even in the case that you both know, correctly, that the other
is superrational, and that each knows that the other knows, and
each knows that the other knows that the other knows, and so on,
then if a meme from the outside drifts in and says to you "PAL,
JUST THIS TIME, DEFECT!", then you *should* be affected by that
meme. Because according to the payoff matrix, you'll be better
off.

> Or in other words, the superrational should cooperate with the
> superrational, and defect with the merely rational.  Of course, knowing
> (or even having a high probability) that the other is superrational is
> probably a side-channel bit which isn't part of the general PD.  It is
> part of the original PD, though, where you and the other prisoner really
> are accomplices.

In the world of real-life prisoners, who have internal payoffs such
as personal loyalty, or even "thieves' honor", and who are human
beings to have a self-image and a self-concept of themselves, then
quite rightly such a prisoner may Cooperate without being irrational.
But even if you are superrational, and it's a one-shot PD, and your
payoff *is* magically captured by the entries in the payoff matrix,
then to ignore the outside meme "PAL, JUST THIS TIME, DEFECT!",
is not rational.

Lee