Evolution simulator for the prisoner's dilemma.
The point of this project is to simulate scenarios in which Agents
play out the prisoner's dilemma but with a conception of memory (so
they respond to what their counteragents last did to them). In order
to test the "fitness" of various strategies, the simulator also has
an appreciation of reproduction, such that Agents achieving certain
scores will be able to reproduce.
The basic idea is that an Agent has two possible actions, Cooperate
and Defect. In the simulation, there is a RewardsVector which will
configure the calculation for what happens in the following scenarios,
for two Agents A and B:
Acooperates,Bcooperates;Acooperates,Bdefects;Adefects,Bcooperates;Adefects,Bdefects.
These outcomes will have different weightings, which will be used on
each round to calculate the Agent's score, and therefore give it a
probability of reproducing.
The simplest behaviour is a "tit-for-tat" algorithm (or "an eye for an
eye"), in which the Agent will, by default, cooperate, but if the
counteragent defects, then it will defect in retaliation.
No natural Agent is perfect, so the world will introduce a "fuck-up
factor" or "mistake factor". This will be the probability of an Agent
which, intending to do one action, does the opposite.
Agents can also have a tolerance of defections in their counteragents,
by adjusting the generosity factor, which is the likelihood that
an Agent will cooperate even in the face of a defection from the
counteragent.
Some people might believe that by being selfish, one can get ahead, and
therefore one should defect more often in the hope that some sucker
cooperates anyway. The selfishness factor of an Agent is the
propensity to defect no matter what.
In this model, children inherit scores from their parents, to mimic how advantage and disadvantage can be inherited in the real world.