Eigenmorality

#1

Post by bdr » Sat, Jun 21, 2014, 4:31 PM UTC

Re: Eigenmorality

#2

Post by knowtheory » Sat, Jun 21, 2014, 5:53 PM UTC

This is very long but worth reading.

The modeling exercise herein is basically attempting to use a game theoretic model to test out some really dumb/simplified models of cooperation and whether the behaviors observed approximate anything approaching what our intuitions might say is moral behavior, up to and including an 'eigenjesus' and 'eigenmoses' up against tit_for_tat bots and the like.

Re: Eigenmorality

#3

Post by MarkPNeyer » Sat, Jun 21, 2014, 6:45 PM UTC

my friends and i had started on this already. i had a hard time explaing to people why it was valuable; looks like scott has done it for us.

please help us!

https://github.com/neyer/dewDrop

right now all we have is a way to state which facebook users a person trusts. there's a chrome extension to help with this. it's extremely basic.

i have a server running at https://dewdrop.neyer.me - we need a lot more help!

i'm just putting it on github now - so i'll update the readme in a few minutes.

Re: Eigenmorality

#4

Post by hhm » Sat, Jun 21, 2014, 7:01 PM UTC

This Tolkien quote builds a similar circular definition of "worth", which might be amenable to the same kind of analysis. https://twitter.com/JRRTolkien/status/480127254857400320

Re: Eigenmorality

#5

Post by rm445 » Sat, Jun 21, 2014, 7:04 PM UTC

knowtheory wrote: post #2

This is very long but worth reading. The modeling exercise herein is basically attempting to use a game theoretic model to test out some really dumb/simplified models of cooperation and whether the behaviors observed approximate anything approaching what our intuitions might say is moral behavior, up to and including an 'eigenjesus' and 'eigenmoses' up against tit_for_tat bots and the like.

It's very long but written in a light, readable style.

What I took away from it was the point late-on about, if we had a PageRank-type system for establishing trust in issues that drive policy (such as climate change), then it would show where a group had left the consensus and formed a separate talking-shop that was trying to shout down the greater consensus.

I think this actually applies back to the Iterated Prisoner's Dilemma games. The 'morality' calculation might not be able to perfectly establish what is moral and immoral behaviour, for the reasons given in the article, but it ought to be able to establish where groups are not working to the moral codes of the majority, perhaps because they are discriminating or discriminated against, but also possibly because they are co-conspiring.

Re: Eigenmorality

#6

Post by wodenokoto » Sat, Jun 21, 2014, 7:04 PM UTC

hhm wrote: post #4

This Tolkien quote builds a similar circular definition of "worth", which might be amenable to the same kind of analysis. https://twitter.com/JRRTolkien/status/480127254857400320

“All have their worth and each contributes to the worth of the others.”

Re: Eigenmorality

#7

Post by hhm » Sat, Jun 21, 2014, 7:18 PM UTC

wodenokoto wrote: post #6

hhm wrote: post #4

This Tolkien quote builds a similar circular definition of "worth", which might be amenable to the same kind of analysis. https://twitter.com/JRRTolkien/status/480127254857400320

“All have their worth and each contributes to the worth of the others.”

I now realize I should have copied the text into the comment in the first place.

Re: Eigenmorality

#8

Post by derefr » Sat, Jun 21, 2014, 7:19 PM UTC

As Aaronson points out, PageRank has a few edge-cases when used to do this analysis, basically because it treats its graph as a closed, internally-solipsistic system--it has no definition of morality other than what each of its nodes prefer of one-another. This works if you have a diverse spectrum of preference functions distributed among the nodes (the result tends toward a "live and let live" meta-ethics), but if your analysis is aimed at a preferentially homogeneous group (e.g. Nazi Germany), PageRank won't give you the solution of "move the 'evil' majority toward the tenets of the good minority." It'll instead suggest that the optimal system would have the 'good' minority give up and become 'evil'.

Scott Alexander suggests (http://slatestarcodex.com/2014/06/20/ground-morality-in-part...) you could instead use DW-nominate, the tool that does meta-cluster-analysis to mathematically detect "party lines" in congress (which are basically just clusters in human-utility-function-space anyway), to find what preference-subfunctions (e.g. helping old ladies cross the street, returning a wallet you find laying on the ground) correlate together into a cluster (that might be called 'goodness') -- and then grounding/normalizing the PageRank analysis with that, such that you can tell whether the system as a whole is in a 'good' or 'evil' state.

Re: Eigenmorality

#9

Post by bryan_rasmussen » Sat, Jun 21, 2014, 7:38 PM UTC

It seems to me that this would only be of interest if it can be shown that an immoral person is not someone that cooperates with other immoral people but not with moral people.

Re: Eigenmorality

#10

Post by miguelrochefort » Sat, Jun 21, 2014, 7:41 PM UTC

MarkPNeyer wrote: post #3

my friends and i had started on this already. i had a hard time explaing to people why it was valuable; looks like scott has done it for us. please help us! https://github.com/neyer/dewDrop right now all we have is a way to state which facebook users a person trusts. there's a chrome extension to help with this. it's extremely basic. i have a server running at https://dewdrop.neyer.me - we need a lot more help! i'm j…

I'm happy to find other people that share my vision.

Drop me a line: username @ gmail