Martin Davies wrote an article for the Guardian Unlimited yesterday that discusses ‘griefing’ in online games. It’s a good article with some interesting ideas about self-policing communities and the pitfalls of such. I know it is a problem that I started thinking about back in the days of playing Meridian 59.

Unfortunately I think he is a bit too broad in his definition of ‘griefers’. He seems to want to use the term for anyone who causes grief, which I suppose makes it a valid definition, but not a very useful one.

If I kill a mob that someone else has been trying to find for the past two hours then I’m going to cause them some grief. Does that automatically make me a griefer? I wouldn’t think so. There could be plenty of valid reasons for me to kill it. Maybe it attacked me and I couldn’t get away. Maybe I’ve been waiting for it even longer. Maybe I didn’t know he was waiting for it and only found out when he came rushing up after combat had started. All of these things are going to cause him grief but shouldn’t make me, or anyone else, a griefer.

What about narrowing down the definition to someone who deliberately does something, knowing it will cause grief? Well, that removes the case of me attacking the mob when I didn’t realize someone else was waiting for it, but it still doesn’t seem to be a good definition. That would make pretty much everyone who plays a PvP game a griefer at some time or another since being killed (especially if there’s penalties associated with it) causes grief.

Instead I would propose that the term ‘griefer’ is used to refer to someone who deliberately inflicts grief for their own enjoyment. Players in PvP will not normally be griefers because although their target feels grief they aren’t deliberately inflicting it purely for their own enjoyment. The grief they inflict is only a byproduct of normal play. Should there be circumstances in which they kill an opposing player solely for the purpose of causing grief (such as attacking a player when nothing will be gained because they are too far below your level) then it turns from PvP to griefing.

Of course this means that someone who cheats another player out of items or money isn’t a griefer, but that’s fine. Griefer doesn’t have to be, nor should it be, a term to cover anyone who has socially disagreeable habits. We can apply other terms to those people; scammers, spammers, exploiters, etc.

So what’s the big deal? Why is it important to define what a griefer is? Because by defining them it helps us to understand the problem. Broad definitions lead to broad solutions, and broad solutions lead to problems.

At this point we can go one of two ways. We can try to come up with solutions to deal with griefers or we can try to come up with solutions to deal with people with socially disagreeable habits. Of course that doesn’t mean that the two possibilities are exclusive of each other. Just because we develop a method of dealing with the broad category of socially disagreeable people doesn’t mean we can’t also come up with a more specific method for dealing with griefers but while we are developing a method of dealing with them we should probably focus on either the group or the subgroup and not both at once.

Given that griefers cause grief for enjoyment rather than as a side effect of play we have something that will help us to spot them. People who initiate combat against players when there is no apparent gain should probably be suspect. Players who attack already engaged targets who cannot gain any benefit from the target should be suspect if their action has a negative effect on the other player (lowering rewards from the encounter).

Of course there may be valid reasons for the action, so a single occurrence probably shouldn’t flag a player as a ‘griefer’. One solution might be to have players accumulate points which slowly decrease over time. As more points accrue the player becomes more penalized. Since the object of griefers is to inflict grief, usually through superior stats, the best penalties probably revolve around decreasing their stats. By setting a reasonable threshold players who accidentally accumulate a few points won’t be penalized.

For the broader category of socially disagreeable people I’m not completely convinced there is a good solution. Most systems will probably be worked out by players at which point some people will undoubtedly try to ‘game’ the system. I don’t think that means we shouldn’t try, however. Instead it just means that we should remain cognisant of the limitations and possibility of exploits.

At the heart of it any system that deals with socially disagreeable players is going to have to rely on human input, I think. Human interaction is simply too complicated to have a coded system that tries to evaluate actions as ‘bad’ or ‘good’, so there will have to be some way for players to ‘tag’ others as one way or another.

Of course this almost inevitably leads to problems. One example I remember is in the Sims Online. There was a group of players who would actively extort other players. If you didn’t give into their demands they would all put you onto their ‘enemies’ list. Since certain character abilities were tied to having more friends than enemies suddenly having fifteen people add you to their enemies list could definitely effect your character. Even worse there was a form of competition that occurred in who could have a higher score of friends minus enemies. I can’t recall if there was an in game bonus to being high on the ladder but even if there wasn’t there was still the prestige.

Another issue with vote based systems is voting rings, groups where everyone votes for each other by either tacit or explicit agreement. While this may seem fairly benign if there is no gain for having a high number of positive votes it does have the problem that it could be used to offset negative votes, allowing players to grief more without penalty.

To make a voting system usable it would probably need to have a fairly complicated formula for weighing votes. Google does something like this when it builds a page rank. Pages that are linked to from lots of other pages are given a higher page score than pages linked to from fewer pages at the basic level. However the system is a great deal more complex than simply that. Links that come from pages with high page scores are worth more than links that come from pages with low page scores. Further complicating it, pages with lots of links out effectively dilute the value of each link, and of course pages that come from the same site are completely ignored.

The exact formula is something of a mystery and is, I believe, a closely guarded secret of Google. However it still gives us a good idea as to how to start. The system should probably work along the lines of players having positive and negative outgoing connections. The weight of those connections are weighted base on various criteria.

To break up voting circles the weight of connections should be decreased based up the relation of links between people. Reciprocal connections (ones where two players either mark each other as good or bad) should be weighed less than one way connections. By walking back through players incoming connections it would be possible to find connections that are multiple steps away (Alice connects to Bob and Bob connect to Charlie) and reduce the weight so that Charlie’s connection has more weight for Alice than Bob but it has the greatest weight for Dave who isn’t part of the chain at all. 

Connections from players with a higher positive value should have more weight than players with a lower value but at the same time the more outgoing connections a player has the less weight each connection is given.

If the connection is a positive connection then the weight of it is added to a player’s value and if it is negative it is subtracted. Thus a player with a high value will have more effect with a connection, up or down, than a player with a low value. Likewise two friends linking to each other will not have much effect, nor will two enemies.

At any rate I’ve rattled on far longer than I meant to. Obviously such a system would need to have its exact values hidden from the players and would need a great deal of tuning but it seems like the basic idea should work.