User:OrenBochman/WGT/Ham Spam Game

Simple Ham Spam - Creator facing a New Page Patroller

A new articles has n memes etimated as 5. They are either contributed as ham or spam and they are either accepted as ham or rejected as spam. By introducing Reputation and Face into the payoff the game can be seen to be an assymetrical Prisoner's dillema (iterated or not).

	Partroller (accept)	Partroller (reject SD)	Partroller (reject Afd)
ham	$n-work$ $n$	$-work$ $-work({\frac {1}{10}}+{\frac {1}{\lVert Community\rVert }})$	$-2work,$ $-work(2+{\frac {1}{\lVert Community\rVert }})$
spam	$Ln-{\frac {work}{3}}$ $-Ln$	$-{\frac {work}{3}}$ $-{\frac {work}{10}}$	$-{\frac {work}{3}}$ $-2work$

• The information is perfect so the patroller can identify spam from ham.
• Deleting SD requires a minimal expenditure in communication, coordination or time.
• Afd requires voting, communication and reaching consensus within a week.
• L is the lemon factor and is greater than one.

The ham spam games can be followed by:

• a single round of CSD/AfD game. High (40%) and on optional round of Xfd (resoration request) (5%)low.
• a single round of a Ban Game below. If spam count > 3.
• a single round of chicken - an edit war with another editor. (would be better modeled using a ham/opinion/spam).

However these have complex probabilities of taking place.

Empirical questions

1. What is the distribution of memes in a new article. Avarage and Standard diviation.
2. Is Accept/Reject dependent on number of memes.
3. The probabilities of CSD AFD have been studied
4. The probability of Bans are more difficult but can also be estimated by reviewing edit histories and counting events.
5. The death event is of interest as well.

Actual deletion can be more complicated for both sides - as can be seen from the chart (developed Alexandre Passant & Jodi Schneider) below:

this chart suggests that the normal form game occurs with multiple agents and multiple rounds at Afd and CSD.

Practical Conerns

Modeling a the playesrs as both a bad faith and a good faith actor simply reflects that each user may have his a conflict of interest with that of wikipedia. This includes pai editing under an alias AKA the invisible pink unicorn. Such an editor can quickly establish a IPU identity with low cost and a minimal reputation to do the spam edits.

Since this is a Prisoners dillema some questions arise.

Both players can may wish to increase their payoffs resulting in a non parto efficent mutual deffection equilibria (spam spam). A social engineer would inquire how a (ham,ham) could be encouraged. This is generaly solved by punishing users who defect.

the IPU is a form of abuse more pravelent amonst advanced users. These experienced users are able to game the system and place articles into wikipedia with minimal work. This has significant value for forign elements who offer significant temptations to have the creator intoroduce spam. Such editors will not longer be impatial when taking the other side as patroller. The risk of spamming is far greater for a patroller since he could loose his privlages. In practice great abuse is tollerated (a moral hazzard)

One case which is not tolerated is and therefore not without risk — once IPUs are exposed users can inccur signficant penalties. and lose their real identity and privileges.

So when they wish to behave badly they will use disposable accounts sometimes refered to as Invisible Pink Unicorns or (IPU). These accounts are also detectable and depending on how badly the IPU is abused - the curtain may be lifted.

Recomendation

• Use of stylometric and editing metadata to identify IPUs.
• A better COI resolution - let there be good spam. Users should be allowed to do much broader work within the scope of their main account, and not have to resort to hiding their identities. This means having a more liberal policy and being clearer about COI. However this will not solve all COI problems.
• So users should also have a stylometric print or another method of authentication which they cannot forge. This is a commited stylometric identiy.
• But revealing identity is punished in the long-term — epsecialy when dealing with controversial matters. So these two concerns must also be resolved.

Spam Ham Game under perfect information

Combine the above in an extensive form.

Simple Ham Spam - Non-Creator facing a Revent Changes Patroller - Sub Game

• This is a close relative of the ham/spam game. However in this case the edit is a smaller departure from the status quoe and thus percieved as less controversial. An quickest rejection of an edit is the a reversion. It is possible for a deletion prorocol to be initated as well.
• On the other hand users are cessured against making many small edits which increase work load at recent changes.
• generaly under knowledge symmetry (the longer the edit the more likely it is to be reverted). this is due to problems like CPVIO risk (though it can also be) due to other reasons.
• Revisions add new memes or replace memes. edits can be ham or spam.

• $P_{revert}={\frac {\lVert Memes\rVert }{Meme_{max}}}*reputation$

	Partroller (accept)	Partroller (revert)
ham, defend	$n-work$ $n$	$-2work-2face$ $-work({\frac {11}{10}}+{\frac {1}{\lVert Community\rVert }})$
ham, abandon	$n-work$ $n$	$-work-face$ $-work/10$
spam, defend	$nL-work/3$ $-nL$	$-2work$ $-work({\frac {11}{10}})$
spam, abandon	$nL-work/3$ $-nL$	$-work/3$ $-work/10$

• remember the information is still symmetric so the patroller can identify ham from spam.
• reversion after defending is a double effrontery resulting in a loss of the editor
• this is a sub game which would be followed by another round where the Patroller bans the Editor with probability

$p_{ban}=\sum edits_{ham}-L\sum edits_{spam}$

• can this become a war of attrition?