-Prisoners dilemma, tit for tat, and
cheating.
Today I want to look at how the
(Darwinian) Theory of Evolution can model morality from purely
egoistic assumptions.
Let us quickly recap the assumptions of
Darwin:
- There are more offspring than can survive (there would be exponential growth if everyone survived) (survive means: have offspring, not survive forever).
- There are differences between offspring.
- These differences are heritable (if your father is taller than average, we expect you to be – this only needs to be statistically true)
The conclusion then follows:
Those traits we see after many
generations are the traits that is well suited to survive the
environment (survive still means: have offspring). Note also that
you cannot develop a complicated trait without there being several
stages of beneficial traits leading up to this complicated trait.
In morality we have a principle known
as "tit for tat". Tit for tat is: "Do unto another as
another has done unto you". If Sam helps you, then the next time
you help Sam. If Sam steals from you, then you steal from Sam.
We will discuss how this principle (tit
for tat) can follow from the Theory of Evolution and purely egoistic
assumptions. First let us define our environment/setup/problem, what
is known as the iterated prisoner's dilemma.
The prisoner's dilemma has its name for a reason, but for reasons of clarity of exposition I will
present it differently. It goes as follows. You and Joe find some
food in the jungle (think about our ancestral environment). You and
Joe get two options: Fight or Cooperate. You can choose different
actions. Results are as follows:
- Both cooperate: Both get 3 (units of) food
- Both fight: Both get 1 food
- A fights while B cooperates: A gets 4 food, B gets 0 food.
Note how this is supposed to model how
food is ruined (the prey escapes etc) when someone fights (in case 2
we loose 4 food, in case 3 we loose 2 food). We assume that the
choice you and Joe picks are independent (you cannot wait and see
what he chooses).
How is the prisoner's dilemma soled by
two egoistical entities? If you and Joe are egoistic you will always
choose to fight. Does this make any sense when both could have more
food if both of you cooperated? Yes, as the choice is independent. No
matter which choice Joe picks, it's always better for
you to fight than cooperate. In that way outcome 2) is recognized as
a stable Nash equilibrium (anyone seen the film "A beautiful mind"?).
Before we go on to iterate this
problem, I want to take a small digression. How can we model the
solution of a single prisoner's dilemma? Well, we can have
utilitarianism (everyone's welfare is important) instead of egoism,
but that was not our assumptions. One common solution is "kin
selection". If Joe is your brother, then his survival will bring
your genes (statistically about half of them) down the line. So his
survival is half as important as yours: He getting 4 food has value 2
to your genes, but you getting 1 food each has value 1.5. This makes
cooperating always better
for your genes. (We just assumed probability of survival is linearly
dependent on amount of food.) But what about your friend, they are
not of your blood, can you still have kin selection? Before modern
transportation it was highly likely that your children would have
kids with the children of one of your friends some day. This might
(with a slightly different setting, and some more assumptions) make
cooperating with your friends a good strategy. From now on we assume
that the two '
'prisoners'
are complete strangers with nothing in common.
What
its iterated prisoner's dilemma? Well, you go hunting with Joe every
week for a few years. Now, every week you make a new choice,
independent of Joe, whether you should fight or cooperate. But now
you are allowed to remember everything that has happened up to this
point. Now you are allowed to choose "Because Joe did X last
time, I will do X now", which we called the tit for tat
strategy. NB: you always start by cooperating. What is special about
this strategy? It is essentially unbeatable.
Let us disregard small deviations (like tit for tat with forgiveness
or extra randomness). What does unbeatable mean? If you have a
population of 100 people, each with their own strategy; some of them
with the tit for tat strategy, some of them with completely different
strategies then no one will get more food than you
if everyone plays the iterated prisoners dilemma against everyone
else (and you are using tit for tat). (This also relies on the
assumption that the world is not dominated by a big number of tit for
tat haters whose strategy is discovering the 'tit for tat strategy
users' and killing them.)
Now
the Theory of Evolution concludes that after many generations,
everyone will have something close to the tit for tat strategy. Yes,
this is dependent on even more assumptions. Maybe we should model this
on the computer one of these days?
No comments:
Post a Comment