[extropy-chat] game theory and decisions

[Damien Sullivan]

This is longer than I expected when I started.  The first half is about game
theory and some of what Axelrod says about the Prisoner's Dilemma, with some
crude implied applications to real life.  The second half is about how we make
decisions, and how that might challenge ideas of minimally restricted freedom.

On Tue, Jul 12, 2005 at 08:37:49AM -0700, Jef Allbright wrote:
> Dirk Bruere wrote:

> >I suggest simple game theory indicates otherwise.
> >Namely, that a group A acting consistently in concert will win every 
> >time over either individuals or a looser group B that sometimes act 
> >for their own group and sometimes for A depending on circumstance.
> >

> I'm glad to see that you and I are now apparently in agreement on the 
> big-picture benefits of cooperation, however, game theory has often been 
> used to demonstrate the opposite point--that at least in the short-term, 
> narrowly defined domain of the game, rational behavior means ruthless 
> identification with the local self and immediate goals.  This leads, as 
> most on this list are well aware, to the so-called Prisoner's Dilemma 
> paradox, where the paradox appears to arise precisely because of the 
> limited scope of the game compared to our real-world experience of an 
> open-ended and not fully knowable web of possible future interactions.

Game theory can be used to show a bunch of things, depending on the
conditions.

The paradox doesn't 'appear' to rise; it rises.  Limited scope describes some
situations: single anonymous interactions, first-strike winner-take-all, being
able to kill your interactee, being able to choose to break relations and run
away.  I read Axelrod's _The Evolution of Cooperation_ recently, and he talks
about a possibility of bacteria being sensitive to the health of their host,
so that they play nice while it is healthy (taking a long-term strategy of
commensalism or mild parasitism, shedding bacteria over a long time), but
multiply all out and grab resources when the host gets sick or badly injured,
figuring the host is going to die soon, it's time to grab what you can and
run.

Iterated Prisoner's Dilemma of course leads to Tit for Tat as one robust
stable strategy, though not the only one.  (BTW, Pinker in _The blank slate_
mapped Tit for Tat into cultures of machismo and respect and reputation, as in
gangs or the Mafia or Southern feuds or international politics.  Which made me
think that multiple private protection agencies might not be such a great
thing.)

And humans go beyond the IPD to having things such as reputation, and labels.
Axelrod discusses other stable strategies.  One is a pair, bully and meek.
The bully alternates cooperation and defection, unless the other defects, in
which case the bully will always defect with that partner.  The meek strategy
cooperates unless it gets defected on twice, in which case it switches to
defecting.

As a pair, this is almost but not quite as productive as a pair of
cooperators.  Most of the benefit goes to the bully.  But it's not to the
benefit of the individual meek strategy to switch to tit for tat.

Analogies to human interactions where an actor may vary being exploitative and
playing nice, but punish retaliation very severely, are left to the reader.

Given labels, Axelrod describes populations of agents A and B.  A's
tit-for-tat with each other but always defect with Bs.  Bs t-f-t with each
other and defect with A's.  Any individual A trying to switch to playing fair
with everyone won't get far, as the Bs will just keep defecting.  In real life
of course, the other A's may add their own punishment to the deviant.  He also
mentions status hierarchies being stable, if agents bully those beneath them
and act meekly to those above them; I think this is just an elaborate version
of the earlier example.

Pernicious effects are easy to find in these two examples: if Bs are a
minority they'll suffer defections more often from running into the more
numerous A's, unless they can congregate.  And in the status hierarchy those
at the bottom will make much less profit than those on top.  No news there,
but at least it's a concrete mathematical example.

What's my point?  Not sure I have a specific one, other than to elaborate a
lot on game theory.  Tit-for-tat (or, nice reciprocal cooperation in general)
is cool, but it's not the end of the story.

> (slight) issue with another poster's statement that "if survival is the 
> goal, then you get there by rational behavior", and pointing out that 
> this is true only in the limited set of those cases where one has 
> sufficient contextual information and sufficient processing power and 
> time to come to a "rational" decision.  Very often, in a situation where 

Indeed.  Quite often you get to survival by instinctive or habitual behavior,
because that behavior is fast.  Conscious look-ahead deliberation, modelling
different futures and picking the best one, can be powerful, but it's also
slow.

Which I think sheds light on failures of "willpower", or "free will", which I
think is best cashed out in everyday speech as the ability to override
short-term desire for the sake of long-term goals.  It's not willpower after
all, but speed: presented with an attractive stimulus the fast reflexive
pathways can command action and attention before the slow rational ones
finish processing.  (Worse, the rational pathways need attention to run with
any speed, but attention may be hijacked.)

Which has been leading me toward ideas about practical limits on individual
freedom.  One pragmatic justification for liberalism/libertarianism is the
idea that the individual will most reliably have the most information about
her needs and abilities on the spot, and thus will make the best decisions for
herself, more so than anyone else.  But, sometimes, it maybe that someone may
make better decisions for you precisely because they are not on the spot, not
subject to the short-term-desire stimulus which is distracting you from the
rational goals you'd normally prefer to follow.

Toy example: someone wants to diet.  That is, they say they want to diet, they
make visible efforts to diet, they feel regret when they break their diet.
But they do often break their diet, typically when some attractive food is in
sight or smell.  The fast habit of eating the food on perception trumps the
slow memory of being on a diet.  Success depends not on willpower -- if you
get that far, you've already lost -- but on cultivating better habits and
inhibitions ahead of time.  Or on avoiding the food.

Or on letting someone else block your behavior when you go for the food.

I note that an economist might say that you have a revealed preference for not
dieting, especially if you buy the food.  But this seems flawed to me.  Yes,
some processes are revealed which prefer the food.  But the stated desires to
diet, and the regret after failure, reveal other processes which prefer to
diet.  Identifying the process which can make impulse purchases with the whole
person seems dubious to me.

-xx- Damien X-)