search

maxkw / evolution

0 stars 0 forks source link

mean RA agents vs. altrusic #4

Closed maxkw closed 8 years ago

maxkw commented 8 years ago

run experiments that show how the mean RA agents play against an altruisic agent. How long does it take for them to realize the other guy is cooperative and start defecting. show graphs with different curves depending on the priors of each agent type.

polyguo commented 8 years ago

In this scenario RA, AA, and SA play prisoner's dilemma.

When the RA prior is high every RationalAgent is expecting that everyone else is rational, and that they expect the same. So RA will expect the other agent to be nice to everyone. If K=0, RA will never figure out AA's type. screenshot from 2016-07-19 12-58-42

It should be the case that by increasing K, RA will be able to figure out who AA is by realizing that AA is not being mean to SA despite having evidence of SA's type. screenshot from 2016-07-19 12-40-50 Which is indeed the case :)

maxkw commented 8 years ago

I don't totally understand the setup - but I suspect this is a very very nice result!

polyguo commented 8 years ago

I don't totally understand the setup - but I suspect this is a very very nice result!

3 agents play Prisoner's Dilemma. The world has only these 3 agents. It's basically protection, but with 3 agents. The selfish agent is just there to give away the AA. RA vs AA just looks like the top one. The discovery is dependent on the prior. I'm curious as to why it won't converge to 1. maybe it's the beta... RAvAA.pdf nope...

maxkw commented 8 years ago

I'm still not clear how the three agents interact. The experiment I was proposing was a single "ReciprocalAgent" plays against a single "AltruisticAgent".