Skip to content

gfarrell/eye-for-an-eye

master
Switch branches/tags
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
app
 
 
src
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

An Eye for an Eye

Evolution simulator for the prisoner's dilemma.

Purpose

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:

  • A cooperates, B cooperates;
  • A cooperates, B defects;
  • A defects, B cooperates;
  • A defects, B defects.

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.

Nuances in the simulation

Imperfect world

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.

Beg forgiveness

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.

El vivo vive del bobo

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.

Fitness inheritance

In this model, children inherit scores from their parents, to mimic how advantage and disadvantage can be inherited in the real world.

About

Evolutionary prisoner's dilemma simulator

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published