Page 1 of 1

The Science of Betrayal

PostPosted: 29 Jan 2019, 13:52
by super_dipsy
Not like me to post here :) . However, I recently was chatting to Amby (he of the Diplomacy podcasts) and he mentioned an article he had read. I thought others might find it interesting, although to be honest it more whets the appetite rather than having too much concrete in. But I found it interesting anyway

https://www.sciencenews.org/blog/culture-beaker/few-key-signs-betray-betrayal

Re: The Science of Betrayal

PostPosted: 29 Jan 2019, 18:43
by Phlegmatic
Very interesting, thanks Dipsy!

This is also a great article that you might like:

http://grantland.com/features/diplomacy-the-board-game-of-the-alpha-nerds/

Re: The Science of Betrayal

PostPosted: 29 Jan 2019, 22:26
by DirtyHarry
That was an interesting article, but I wish there was more "meat". More examples on how the comms changed before something bad was about to happen :-)

Re: The Science of Betrayal

PostPosted: 30 Jan 2019, 01:09
by Tortellacci
DirtyHarry wrote:That was an interesting article, but I wish there was more "meat". More examples on how the comms changed before something bad was about to happen :-)


^^ This. Granted the article was about betrayal and only used Diplomacy as an example, but regardless I wish there was more meat. Very interesting though. The program they mentioned predicted stabs over 50% of the time.. that's very good!

Re: The Science of Betrayal

PostPosted: 31 Jan 2019, 09:22
by Calavera
I might go against everyone's opinion, but this article is plain terrible. For instance, they have links every time they put a direct quote, but this link actually leads not the the interview or a paper, but to that person home page. Very similar are their other links - they don't lead to events or science papers, but to home pages of respective institutions or persons, or even Wikipedia articles about songs unrelated to the topic. The only actual link they provide is a grad student's site with article preprint.

At first I thought this is a classic scientist rapes reporter case, but the study they link to is not too rigorous itself. Its flaws are numerous. First, they do t-tests on random variables which they don't even try to qualify as normally distributed (quite a common thing in machine learning circles nowadays, but still very bad). Second, they have somewhat dubious definitions on what is friendship and what is betrayal in Diplomacy. Third, they state that "victims" of betrayal see it coming with nearly zero probability, but don't support this claim with anything. Fourth, they correctly indicate that length of a friendship is extremely important factor in betrayal chance, but neither incorporate it in their model, nor try to exclude it. Fifth, they don't specify a score measure. If we assume that 57% is a commonly used F1 score, then I believe regular Diplomacy players anticipate betrayals and non-betrayals with much greater accuracy (though I, obviously, don't have proof to support this).

Overall, a decent article for a grad student, maybe even publishable in local university journal. But not a serious study, and Science News website is not very scientific