Let's say Mr. Blue, and Ms. Red have each
been arrested for some minor crime. The police think they committed a more serious
crime but they don't have enough evidence to convict them. They need a confession. They take them and put them in separate rooms
so they can't talk, and play a little game. To try to force a confession the police give
them each a choice. Admit your partner committed the crime, and
you will go free. We'll pardon you for the minor crime but your
partner will have to spend 3 years in prison. If you stay silent and your partner lets us
know that you were the one who really did it then you're going to have to go away for
3 years. They know that the police don't have any evidence
and if they both stay silent then they will only go to prison for 1 year each for the
minor crime. If they both betray each other then they'll
both go to prison for 2 years each. OK, each partner can do 1 of 2 things. Stay Silent, or Betray. Staying silent would be cooperating and betraying
would be defecting. If they both stay silent, they each spend
a year in prison. If one betrays and the other stays silent,
then the betrayer goes free and the silent spends 3 years in prison. If they both betray then it's 2 years each. So what are they going to do? Well they should cooperate. That's the best option for the group, if we
add the total number of years in prison. But let's take it from Red's perspective. If she thinks blue is going to stay silent,
then she should betray so she can go free. Going free is better than a year in prison. If she thinks he's going to betray her then
she should definitely betray, 2 years in jail is better than 3 and being made a fool of. Blue is in the exact same situation and will
think the exact same thing, he should betray if she stays silent and he should betray if
she betrays. They should have both cooperated, but from
an individual stand point they noticed they could always gain by defecting. If they have no control over what the other
person is going to do. So they'll both defect to try to better their
own situation. But come away not only hurting the group,
but themselves. Individually they're worse off than if they
both cooperated. This situation is pretty made up, but it has
some real world analogues. A common example is with marketing .
Let's say 2 cigarette companies, Red Strikes, and Smooth blue, are deciding how much money
they should spend on advertising. Since the product they each make is identical
to one another, advertising has a huge impact on sales. For simplicity let's say their choices are:
to advertise a bunch, or not advertise at all. And there's just 100 people in this society
and they all smoke. If both don't advertise, then just by random
chance picking cigarette boxes, 50 people buy Red Strikes and 50 people buy Smooth blue. At $2 a pack they each make $100. Let's say advertising costs $30. If one person advertises and the other does
not, then 80 people will buy the cigarettes from the ads and 20 people buy the other ones. The advertiser makes $160 minus $30 for ads,
and comes away with $130. The non advertiser didn't spend money, but
only made $40. If they both advertise, again half will buy
Red Strikes, and half will buy Smooth blue. But since they both spent $30 on advertising,
they only come away with only $70 each. Same deal, both people cooperating and not
advertising is the most preferable situation, but both company can see that advertising
will always make them more money. But unlike the prisoner's in jail, these companies
can talk and try to influence each other. From here Blue would be better off if Red
didn't advertise. Red wouldn't go for that because that would
be worse for them. Blue could try to convince Red that they would
both not advertise, the only other situation where they're both better off. But without any real obligation to each other,
there's nothing that's stopping them from trying to advertise to gain more of the market
anyway. If you think your opponent's going to not
advertise, you're better off advertising. Although we're still making assumptions to
make this situation work too. With this model we're assuming they only play
once. The game changes when the players have a chance
to build a relationship and work together to get more gains over time, or punish each
other by not cooperating. Also to make the model work we have to make
up rules for the players. Assume they're basically computer programs
with predictable actions. These guys are creepier than they were in
my head. They were supposed to be cute. For the prisoner's dilemma and other similar
models, we're assuming they are Rational Agents. A rational agent is a hypothetical person
that will always pick the option that they predict will work out best for them. They're not really thinking about the gains
of someone else. Seems selfish but it something that real people
will generally do too. People always want what's best for themselves
and we don't like to made a fool of. But if you put real people in the prisoner's
dilemma, people don't always defect like the model predicts. In one study, 40 people playing prisoner's
dilemma games, through a computer, without ever meeting or talking, only playing each
opponent once, these are one off games, using a payoff matrix that looks like this, cooperated
an averaged 22% of the time. These people never cooperated. These are people always cooperated. These guys cooperated on half of their games
and everyone else is in between. This is a lot of cooperation coming from a
model that predicts no cooperation. The largest group did act like rational agents,
but most people tried to cooperate at least once. It's because there's more to real people. We are social creatures and even in a one
off scenario with no guarantees and obligations and no chance to build a relationship, we're
still thinking about how the group might decide. We're actually thinking from the perspective
of the group, and making an optimistic decision. Cooperating an average of 20% of the time
might not seem very optimistic, but remember this is with absolutely no communication or
obligations. Anyways, that's not really the point. Using the rational agent is still useful. The model is just trying to point out the
dilemma of certain specific situation where people actually hurting themselves when counter-intuitively,
they're only thinking about themselves... and that's why we're modelling using the cold
robotic sociopaths.
Person Anthony, a fantasy sci-fi writer, used the same delima in his book Golem in Gears
I've been fascinated with the concept since
I meant to look up what the Prisoner's Dilemma was after seeing it mentioned whilst reading the Baby Eaters sci fi short story a while back, but forgot about it and never got around to it. This was a pretty good concise explanation
Link to Baby Eaters, in case anyone's interested (its a pretty good read) http://lesswrong.com/lw/y5/the_babyeating_aliens_18/
The follow-up video is much more interesting:
The Iterated Prisoner's Dilemma and The Evolution of Cooperation